University of Southampton Research Repository Eprints Soton

University of Southampton Research Repository ePrints Soton

Copyright © and Moral Rights for this thesis are retained by the author and/or other copyright owners. A copy can be downloaded for personal non-commercial research or study, without prior permission or charge. This thesis cannot be reproduced or quoted extensively from without first obtaining permission in writing from the copyright holder/s. The content must not be changed in any way or sold commercially in any format or medium without the formal permission of the copyright holders.

When referring to this work, full bibliographic details including the author, title, awarding institution and date of the thesis must be given e.g.

AUTHOR (year of submission) "Full thesis title", University of Southampton, name of the University School or Department, PhD Thesis, pagination

http://eprints.soton.ac.uk

UNIVERSITY OF SOUTHAMPTON

FACULTY OF MEDICINE

MEASURING THE MENTAL HEALTH OF THE VETERINARY PROFESSION: PSYCHOMETRIC CONSIDERATIONS

DAVID JAMES BARTRAM

Thesis for the degree of Master of Philosophy

December 2011 i

UNIVERSITY OF SOUTHAMPTON ABSTRACT FACULTY OF MEDICINE Master of Philosophy

MEASURING THE MENTAL HEALTH OF THE VETERINARY PROFESSION: PSYCHOMETRIC CONSIDERATIONS By David James Bartram

The UK veterinary profession has an elevated risk of suicide. The ability to measure the psychological health of this occupational group is an important first step towards suicide prevention. The aim of this research was to assess the suitability of the 14- item Warwick-Edinburgh Mental Well-being Scale (WEMWBS) in this context. The psychometric properties of data quality, scaling assumptions, targeting, reliability and validity were rigorously evaluated by the complementary application of traditional and modern psychometric methods to data from two large cross-sectional samples of the veterinary profession (n = 1796 and n = 8829) derived from independent postal surveys. Where possible, the findings were cross-validated between samples. External construct validity was assessed by testing a priori hypothesised associations between WEMWBS scores and other measures of psychological health or psychosocial work characteristics. Internal construct validity was examined using traditional factor analytic techniques and Rasch analysis. The WEMWBS did not show floor or ceiling effects. As hypothesised, scores were negatively associated with anxiety and depressive symptoms and suicidal ideation. Data for the original 14 items deviated significantly from Rasch model expectations (chi-square = 558.2, df = 112, p = < 0.001, PSI = 0.918). A unidimensional 7-item scale (Short WEMWBS, SWEMWBS) with acceptable fit to the model (chi-square = 58.8, df = 56, p = 0.104, PSI = 0.832) was derived by sequential removal of the most misfitting items. Fit statistics for SWEMWBS were consistent across four further random subsets of data drawn from both samples. Interval-level measurements on SWEMWBS retained associations with scores on comparator instruments. The analyses provided a range of evidence of the psychometric properties of WEMWBS and SWEMWBS, consistent with published results for general population samples. The measurement invariance of SWEMWBS between samples of this occupational group and the general population was supported. SWEMWBS has robust measurement properties which support its suitability as an overall indicator of population mental health and well-being in the UK veterinary profession. iii

LIST OF CONTENTS

Abstract...... i

List of contents...... xiii

List of tables...... xiii

List of figures...... xiii

Declaration of authorship ...... xiii

Acknowledgements...... xiii

List of abbreviations ...... xvii

Chapter 1: Introduction ...... 1

1.0 Introduction ...... 2 1.1 Elevated suicide risk among veterinary surgeons...... 2 1.2 Measurement of psychological health...... 3 1.3 Need for quality measures of the veterinary profession’s mental health...... 5 1.4 Where to begin...... 5 1.5 Research questions ...... 6 1.6 Aims and objectives ...... 6 1.7 Approach...... 7 1.8 Contributions to research and practice ...... 9 1.9 Document structure...... 10 Chapter 2: Veterinary surgeons and suicide: a structured review of possible influences on increased risk...... 11

2.0 Introduction ...... 12 2.1 Literature search methodology...... 12 2.2 Suicide statistics...... 13 2.2.1 United Kingdom...... 13 2.2.2 Rest of the world ...... 20 2.3 Risk factors for suicide and suicidal behaviour ...... 22 2.4 Protective factors against suicide and suicidal behaviour...... 25 2.5 Conceptual models of suicidal behaviour...... 25 2.6 Possible influences on suicide risk among veterinary surgeons ...... 27 2.6.1 Access to means of suicide...... 27 iv

2.6.2 Attitudes to death and euthanasia...... 29 2.6.3 Suicide ‘contagion’ ...... 30 2.6.4 Cognitive and personality factors ...... 30 2.6.5 Work-related stressors ...... 33 2.6.6 Effects of gender ...... 39 2.6.7 Perceived stigma...... 41 mental health...... 41 2.7 Towards a hypothetical model to explain suicide risk in veterinary surgeons: a schematic representation ...... 48 2.8 Shortfalls in existing published work ...... 50 2.9 Summary...... 50 Chapter 3: Measurement properties of health rating scales ...... 53

3.0 Introduction ...... 54 3.1 Psychometric theory: traditional and modern...... 55 3.2 Traditional psychometric methods ...... 55 3.3 Modern psychometric methods ...... 57 3.4 Rasch analysis ...... 58 3.4.1 Limitations of the Rasch model ...... 63 3.5 Quality criteria for the measurement properties of health rating scales ...... 63 3.5.1 Six psychometric properties: data quality, scaling assumptions, targeting, reliability, validity and responsiveness ...... 64 3.6 Conclusion ...... 70 3.7 Summary...... 71 Chapter 4: Materials and methods...... 73

4.0 Introduction ...... 74 4.1 Source of the datasets ...... 75 4.2 Ethical considerations ...... 76 4.3 Instruments ...... 77 4.3.1 The Warwick-Edinburgh Mental Well-being Scale (WEMWBS)...... 77 4.3.2 The Hospital Anxiety and Depression Scale (HADS)...... 78 4.3.3 The Alcohol Use Disorders Identification Test alcohol consumption questions (AUDIT-C)...... 80 4.3.4 Three items on suicidal ideation sourced from the second National Survey of Psychiatric Morbidity ...... 80 4.3.5 The Health and Safety Executive Management Standards Indicator Tool (HSE MSIT)...... 81 v

4.3.6 Negative and positive work-home interaction (WHI_N and WHI_P) sub- scales of the SWING instrument ...... 82 4.4 Data analysis...... 83 4.4.1 Direction of scoring ...... 83 4.4.2 Treatment of missing data...... 83 4.4.3 Evaluation of psychometric properties ...... 86 4.5 Summary...... 108 Chapter 5: Results ...... 111

5.0 Introduction ...... 112 5.1 Demographic and occupational characteristics of study samples...... 112 5.2 Traditional psychometric analyses ...... 113 5.2.1 Data quality ...... 115 5.2.2 Scaling assumptions ...... 116 5.2.3 Targeting ...... 116 5.2.4 Reliability...... 118 5.2.5 Convergent construct validity ...... 121 5.2.6 Discriminant construct validity...... 122 5.2.7 Group difference validity ...... 122 5.2.8 Exploratory factor analysis (EFA)...... 125 5.2.9 Confirmatory factor analysis (CFA) ...... 129 5.3 Modern psychometric methods ...... 129 5.3.1 Rasch analysis ...... 129 5.4 Summary...... 150 Chapter 6: Discussion...... 153

6.0 Introduction ...... 154 6.1 Psychometric properties of WEMWBS: traditional psychometric methods .. 154 6.2 Psychometric properties of WEMWBS: Rasch analysis ...... 156 6.3 Comparison and contrast of findings: traditional vs. modern paradigms ..... 157 6.4 Transformation of raw scores to interval measurements ...... 158 6.5 Convergent construct validity ...... 159 6.6 Remaining gaps in the evidence base ...... 161 6.7 Validation is a process, not an outcome ...... 162 6.8 Strengths and limitations...... 163 6.8.1 Strengths...... 163 6.8.2 Limitations ...... 165 6.9 Implications for research ...... 168 vi

6.9.1 Implications of psychometric evaluations...... 168 6.9.2 Implications of the wider findings of the literature review...... 172 6.10 Implications for the profession ...... 174 6.10.1 Implications of psychometric evaluations...... 174 6.10.2 Implications of the wider findings of the literature review...... 175 6.11 Conclusions...... 188 Chapter 7: Conclusions...... 191

7.0 Towards meaningful measurement...... 192 7.1 Revisiting the study aims and objectives ...... 193 7.1.1 Defensibility of comparing scores across other population groups ...... 193 7.1.2 Potential utility for research, including monitoring of trends...... 194 7.2 A beginning rather than a conclusion...... 196 Appendix 1: The Warwick-Edinburgh Mental Well-being Scale (WEMWBS)...... 201

Appendix 2: Survey 1 questionnaire ...... 205

Appendix 3: Study synopsis - Survey 1 ...... 213

Appendix 4: Survey 2 questionnaire ...... 221

Appendix 5: Ethical approval and sponsorship - Survey 1...... 239

Appendix 6: Ethical approval and sponsorship - Survey 2...... 245

Appendix 7: Further results of multiple regressions to estimate associations between scales administered in Survey 1 ...... 249

Appendix 8: Synopsis of qualitative studies...... 261

Appendix 9: Related publications...... 267

Glossary...... 267

References...... 275 vii

LIST OF TABLES

Table 2.1: High risk occupational groups for suicide (suicide and open verdicts), men aged 16 to 64 years 1979-1990, men aged 20 to 64 years 1982-1987, 1991- 1996 and 2001-2005, England and Wales...... 16

Table 2.2: Suicide and undetermined intent deaths for men aged 16 to 45 years and 46 to 64 years in Scotland in 1981-1999 ...... 17

Table 2.3: Mortality from suicide in male veterinarians, medical practitioners and dental practitioners, and in female veterinarians and medical practitioners, aged 20 to 74 years, in England and Wales in 1979-1990 (excluding 1981) and 1991-2000 ...... 18

Table 2.4: Relative risk (RR) of suicide in high risk occupational groups for men aged 16 to 44 years and 45 to 64 years and females aged 16 to 64 years, in England and Wales in 1990-1992, compared with reference groups ...... 19

Table 2.5: Percentage distribution of suicides by method for veterinarians and combined suicides for all occupations, for males aged 20 to 64 years and for females aged 20 to 59 years, in England and Wales, 1982 to 1996...... 20

Table 2.6: Suicide rates for men aged over 20 years per 100,000 person years in Norway in 1960-2000...... 21

Table 2.7: Social and demographic risk factors for suicide and suicidal behaviour . 23

Table 2.8: Psychiatric risk factors for suicide and suicidal behaviour ...... 23

Table 2.9: Situational risk factors for suicide and suicidal behaviour...... 24

Table 2.10: Biological risk factors for suicide and suicidal behaviour ...... 24

Table 2.11: Psychological risk factors for suicide and suicidal behaviour...... 24

Table 2.12: Protective factors against suicide and suicidal behaviour...... 25

Table 2.13: Summary of studies included in this review with findings relevant to mental health in veterinarians ...... 45

Table 4.1: Summary of psychometric assessments applied to two samples...... 87 viii

Table 4.2: Psychometric properties, their definitions and criteria used to indicate acceptable performance...... 90

Table 5.1: Summary of sample demographic and occupational characteristics .... 113

Table 5.2: Traditional psychometric methods – data quality, scaling assumptions, targeting, reliability ...... 114

Table 5.3: Traditional psychometric methods – convergent and discriminant construct validity and group difference validity...... 123

Table 5.4: Summary of factor analytic assessments applied to two samples...... 126

Table 5.5: Rasch fit statistics for each of the 14 items...... 131

Table 5.6: Differential item functioning for gender ...... 135

Table 5.7: Differential item functioning for age group ...... 137

Table 5.8: Summary of model fit statistics for original scale, successive revisions and cross-validation ...... 145

Table 5.9: Principal component analysis of Rasch item residuals for the derived 7-item scale...... 146

Table 5.10: Rasch fit statistics for the 7 retained items ...... 146

Table 5.11: Raw score to interval measure conversion table for the Rasch-derived 7-item scale...... 149

Table 6.1: Proposed individual-oriented interventions with potential to help alleviate psychological distress among veterinary surgeons...... 189

Table 6.2: Proposed organisation-oriented interventions with potential to help alleviate psychological distress among veterinary surgeons ...... 190

Table A7.1: Associations between scales and the estimated odds of at-risk drinking ...... 251

Table A7.2: Associations between scales and the estimated odds of reporting having experienced suicidal thoughts in the previous 12 months...... 252 ix

Table A7.3: Associations between scales and WEMWBS or the Rasch-derived 7- item interval scale (SWEMWBS) score...... 253

Table A7.4: Associations between scales and the estimated odds of HADS-A possible or probable caseness (score ≥ 8) ...... 254

Table A7.5: Associations between scales and the estimated odds of HADS-A probable caseness (score ≥ 11)...... 255

Table A7.6: Associations between scales and the estimated odds of HADS-D possible or probable caseness (score ≥ 8) ...... 256

Table A7.7: Associations between scales and the estimated odds of HADS-D probable caseness (score ≥ 11)...... 257

Table A7.8: Pearson correlation coefficients between HADS, WEMWBS and HSE MSIT psychosocial work characteristics sub-scales ...... 258

Table A7.9: Pearson correlation coefficients between HADS, the Rasch-derived 7-item interval scale (SWEMWBS) and HSE MSIT psychosocial work characteristics sub-scales...... 259 xi

LIST OF FIGURES

Fig. 2.1: Schematic representation of a hypothetical model to explain the risk of suicide in veterinary surgeons...... 49

Fig. 3.1: Rasch equation ...... 60

Fig. 3.2: Probability of a person affirming an item...... 61

Fig. 3.3: Interval between adjacent scores on an ordinal scale is not equal...... 62

Fig. 4.1: WEMWBS items ...... 78

Fig. 4.2: HADS items ...... 79

Fig. 4.3: AUDIT-C items...... 80

Fig. 4.4: Suicidal ideation items ...... 81

Fig. 4.5: HSE MSIT items ...... 82

Fig. 4.6: WHI_N and WHI_P sub-scale items ...... 83

Fig. 4.7: Samples used in the Rasch analysis and output validation ...... 101

Fig. 4.8: The Rasch analysis process ...... 102

Fig. 4.9: Ideal response option thresholds ordering...... 104

Fig. 5.1: Proportion of item-level missing data [a] Survey 1; and [b] Survey 2...... 115

Fig. 5.2: Score distributions: [a] WEMWBS raw score, Survey 1; [b] WEMWBS raw score, Survey 2; and [c] Rasch-derived interval scale, Survey 1 ...... 117

Fig. 5.3: Proportion of respondents endorsing each WEMWBS item response category – Survey 1 ...... 119

Fig. 5.4: Proportion of respondents endorsing each WEMWBS item response category – Survey 2 ...... 120

Fig. 5.5: Scree plot for 14-item WEMWBS [a] Survey 1, and [b] Survey 2 ...... 127

Fig. 5.6: Scree plot for Rasch-derived 7-item scale [a] Survey 1, and [b] Survey 2 ...... 128 xii

Fig. 5.7: Response option thresholds ordering for item 9 ...... 130

Fig. 5.8: Threshold map for the 14-item WEMWBS...... 130

Fig. 5.9: Item characteristic curves for: [a] items 8, [b] item 12, and [c] item 9 ...... 132

Fig. 5.10: Item scoring for respondents with [a] a high negative person fit residual, and [b] a high positive person fit residual...... 134

Fig. 5.11: Item characteristic curves by gender for [a] item 7, and [b] item 10...... 136

Fig. 5.12: Item characteristic curves by age group for [a] item 1, and [b] item 6.... 138

Fig. 5.13: Item characteristic curves by gender for [a] item 1, [b] item 3, and [c] items 1 and 3 combined as a subset ...... 140

Fig. 5.14: Item characteristic curves by age group for [a] item 1, [b] item 3, and [c] items 1 and 3 combined as a subset...... 142

Fig. 5.15: Person-item threshold distribution for the derived 7-item scale ...... 147

Fig. 5.16: Item map for the derived 7-item scale...... 148

Fig. 5.17: Scale characteristic curve for the derived 7-item scale...... 149

Fig. 6.1: HADS-D score floor effect in Sample 1...... 155

Fig. 6.2: The mental health spectrum ...... 176

Fig. 6.3: An example of advertising to raise awareness of support services ...... 180

Fig. 6.4: An example of signage (sticker)...... 187 xiii

DECLARATION OF AUTHORSHIP

I, David James Bartram declare that this thesis and the work presented in it are my own and have been generated by me as the result of my own original research.

MEASURING THE MENTAL HEALTH OF THE VETERINARY PROFESSION: PSYCHOMETRIC CONSIDERATIONS

I confirm that:

1. This work was done wholly or mainly while in candidature for a research degree at this University;

2. Where any part of this thesis has previously been submitted for a degree or any other qualification at this University or any other institution, this has been clearly stated;

3. Where I have consulted the published work of others, this is always clearly attributed;

4. Where I have quoted from the work of others, the source is always given. With the exception of such quotations, this thesis is entirely my own work;

5. I have acknowledged all main sources of help;

6. Where the thesis is based on work done by me jointly with others, I have made clear exactly what was done by others and what I have contributed myself;

7. Parts of this work have been published as:

BARTRAM, D.J. & BALDWIN, D.S. (2010) Veterinary surgeons and suicide: a structured review of possible influences on increased risk. Veterinary Record 166, 388-397

BARTRAM, D.J., SINCLAIR, J.M.A. & BALDWIN, D.S. (2010) Interventions with potential to improve the mental health and well-being of UK veterinary surgeons. Veterinary Record 166, 518-523

ACKNOWLEDGEMENTS

I am sincerely grateful to Professor David S. Baldwin and Dr. Julia Sinclair of the Mental Health Group, Faculty of Medicine, University of Southampton for their enthusiastic academic supervision of the work presented in this thesis. Regretfully, due to a change in work circumstances, I am unable to complete the data analysis and writing up of two related qualitative studies which they also supervised. However, the alternative arrangements in place for the completion of these studies will help to ensure that their potential contributions to the understanding of suicide in the veterinary profession are not lost.

This research would not have been possible without the support of the many veterinary surgeons throughout the UK who invested their time into completing and returning the questionnaires. I offer in return, the assurance of my best efforts to ensure their responses are a catalyst for continuing endeavours to improve the psychological health of the profession.

Veterinary Business Development Ltd., publisher of Veterinary Times, provided use of their Vetfile® database as the sampling frame, and printed and posted the questionnaires and covering letters for the survey in 2007.

The Royal College of Veterinary Surgeons agreed to include the Warwick-Edinburgh Mental Well-being Scale in the RCVS Survey of the Profession 2010 and supplied the response data in a suitable format for analysis.

I thank the tutors on the statistics modules for postgraduate students in 2009 at the Faculty of Medicine, University of Southampton, and the on the introductory course in Rasch analysis in 2010 at the Psychometric Laboratory for Health Sciences, Department of Rehabilitation Medicine, University of Leeds, for inspiring my interest in quantitative methods.

Special thanks are owed to my wife, Ann-Marie, for her forbearance and for helping to proof-read the manuscript.

This thesis is dedicated to the memory of two colleagues who were undergraduates with me at the Royal Veterinary College, University of London, from 1983 to 1988, and have since died by suicide. Their tragic and untimely deaths, like those of many others who take their own lives, were potentially preventable.

xvii

LIST OF ABBREVIATIONS

AUDIT-C Alcohol consumption items: Alcohol Use Disorders Identification Test CFA Confirmatory factor analysis CFI Comparative fit index CI Confidence interval CTT Classical test theory DIF Differential item functioning EFA Exploratory factor analysis HADS Hospital Anxiety and Depression Scale HADS-A Anxiety sub-scale of HADS HADS-D Depression sub-scale of HADS HSE MSIT Health and Safety Executive Management Standards Indicator Tool ICC Item characteristic curve ICCC Intra-class correlation coefficient IRT Item response theory KMO Kaiser-Meyer-Olkin measure of sampling adequacy LOGIT Log odds probability unit MS Mean square OR Odds ratio PCA Principal components analysis PMR Proportional mortality ratio PSI Person separation index RCVS Royal College of Veterinary Surgeons RMSEA Root mean square error of approximation SD Standard deviation SE Standard error SEM Standard error of measurement SMR Standardised mortality ratio SWEMWBS Short Warwick-Edinburgh Mental Well-being Scale WEMWBS Warwick-Edinburgh Mental Well-being Scale WHI_N Negative work-home interaction WHI_P Positive work-home interaction 1

Chapter 1:

Introduction

2 Chapter 1: Introduction

1.0 Introduction

This chapter presents an overview of the current knowledge of suicide risk and psychological health in the profession, and an introduction to psychological health measurement and the psychometric properties of rating scales, as a background to the research. The research questions, aims and objectives of the study are described and the methodology is summarised. An overview of the study’s contributions to research and practice is provided. The chapter ends with a summary of the structure of the thesis.

1.1 Elevated suicide risk among veterinary surgeons

In the UK, mortality due to suicide is higher in the veterinary profession than in the general population, the proportional mortality ratio for suicide being around four times that of the general population and twice that of other healthcare professions (Bartram and Baldwin 2010, Platt et al. 2010b). This pattern of excess mortality from suicide relative to other causes of death appears to have remained fairly stable across recent decades. The relative risk of suicide across occupational groups is often explained by differences in demographic factors but veterinary surgeons have a higher risk of suicide even when these are taken into account (Charlton 1995).

Little is known about the mechanisms of increased suicide risk in the profession. It is uncertain whether the increased risk derives from the characteristics of individuals entering the profession, the work environment, or other factors known to influence suicide. In common with other high risk occupational groups, veterinary surgeons have ready access to effective means of suicide which may play an important aetiological role (Hawton 2007).

Suicide rate is sometimes used as an imperfect proxy indicator of population mental health status (Bray and Gunnell 2006). The increased suicide risk among veterinary surgeons may be an indicator of increased psychological morbidity within the profession. Bartram et al. (2009a, b, c) conducted the first comprehensive cross- sectional survey of the extent of psychological morbidity among veterinary surgeons Chapter 1: Introduction 3

in the UK using several standardised instruments embedded into a postal questionnaire. The high levels of psychological distress reported suggest that ready access to effective means of suicide is probably not operating in isolation to increase the suicide risk.

Research into mental health and suicide among veterinary surgeons is important, not only with a view towards enhancing the well-being of individuals within the profession and the negative effects that their mental ill-health can have on their work colleagues, friends and families, but also to safeguard veterinary public health and the health and welfare of animals under their care. Research in this professional group might provide additional insights into the influences on mental health and well- being in other occupations. Moreover, mental ill-health can have financial implications for the affected individual, their employer and healthcare provider.

1.2 Measurement of psychological health

Not everything that counts can be counted and not everything that can be counted counts. Albert Einstein (attributed) [1879-1955]

Many dimensions of psychological health cannot be measured directly, such as attitudes and beliefs, intentions and motives, mood states or behaviours that occur in a private setting. Instead, the attributes are measured indirectly by operationalising them into multi-item self-report instruments. Each item is considered an imperfect measure of the attribute of interest but, as a whole, a set of similar items is hypothesised to provide valid indirect assessment of the targeted attribute. Responses to items that are believed to represent a single shared attribute are usually summed to form a composite score to indicate the level of severity of that attribute. Because the attributes of interest are not directly observed, they are referred to as latent variables or constructs. In this thesis, the term ‘rating scale’ is used as the umbrella term to cover any instrument that conforms to a questionnaire- style structure, and is used to obtain scores from a person’s responses to statements or questions, which in turn are considered to be measurements of a specific variable.

There are three domains on which the quality of a health rating scale can be assessed – validity, reliability and responsiveness (Mokkink et al. 2010a, b, c). Validity is the degree to which an instrument measures the construct(s) it is purported to measure. Reliability is the extent to which scores for patients who have 4 Chapter 1: Introduction

not changed are the same for repeated measurement under several conditions. Responsiveness is the ability of an instrument to detect change over time in the construct to be measured.

However, even if these psychometric properties are considered to be acceptable for a particular scale, individual scores for each item on the questionnaire and the composite score derived by summing each of the item scores, are ordinal in nature. That is, the scores have a rank order but the intervals between each score are not necessarily equal (Stucki et al. 1996, Hobart et al. 2007). The Hospital Anxiety and Depression Scale (HADS) (Zigmond and Snaith 1983) can be used as an example. Each item has four possible response categories and, although each category represents a greater severity of anxiety or depressive symptoms than the previous category, the difference between sequential categories in terms of a quantifiable amount of psychological distress is unknown. The categories are assigned sequential integer scores which imply incorrectly that the differences are equal. In fact the categories are essentially arbitrary observations and the scores do not constitute interval measures (Hobart and Cano 2009).

Ordinal scores are sufficient to separate respondents into groups based upon the magnitude of the construct being measured. For example, HADS can be used to categorise respondents into non-case, possible case or probable case of clinical anxiety and/or depressive illness (Zigmond and Snaith 1983). Ordinal scores are also sufficient to indicate whether the severity of the health dimension has improved or worsened. However, it is difficult to interpret the extent of the change because the distance between each score on the scale does not necessarily reflect equidistant steps in the severity of the underlying construct (Stucki et al. 1996). Moreover, it is not possible to make meaningful score comparisons between subgroups of respondents because, although it is possible to estimate whether the score for one subgroup is significantly different from the score for another subgroup, the extent of the difference in the severity of the underlying trait is unknown. Furthermore, the application of parametric statistical methods is limited by the fact that such methods assume the intervals between each score are equal (interval-level scaling).

A further issue which has potential to confound the interpretation of scores is the extent to which evidence for the psychometric properties of validity, reliability and responsiveness for an instrument can be generalised across different groups of people. For example, without evidence that construct validity (the extent to which Chapter 1: Introduction 5

item sets are an indirect measure of the latent variable) is invariant across population groups, differences between groups might reflect differences in the measurement properties of the instrument used rather than true differences in the latent variable (Meredith and Teresi 2006). Quantitative group comparisons are not defensible if the instrument scores cannot be interpreted in the same way across the groups concerned (Gregorich 2006). Differences in the construct validity of the instrument between different groups of people could be due, for example, to differences in the meaning of the items and their relations. Consequently it is appropriate to examine the consistency of the construct validity of an instrument in populations which may differ from those in which this property has been previously assessed, using both traditional (Gregorich 2006, Meredith and Teresi 2006) and modern (Lambert et al. 2011) psychometric methods. For example, Rasch analysis of HADS data has been reported for several populations including cancer patients (Smith et al. 2006), rehabilitation (Pallant and Tennant 2007), Parkinson’s disease (Forjaz et al. 2009), motor neurone disease (Gibbons et al. 2011) and caregivers of cancer survivors (Lambert et al. 2011). Variation in the psychometric properties of instruments can also occur across different occupational groups (Leiter and Schaufeli 1996, Vandenberg and Lance 2000).

1.3 Need for quality measures of the veterinary profession’s mental health

The ability to make high quality measurements of mental health and well-being in the profession is a requirement for the meaningful interpretation of results from studies which aim to explore potential influences on the elevated suicide risk, examine changes in mental well-being over time, or evaluate the effectiveness of interventions. Yet a structured review of the literature (presented in Chapter 2) has identified that several of the questionnaire instruments that have been used in reported studies to explore mental health and well-being in the veterinary profession are self-formulated and no attempt has been made to establish their measurement qualities. Moreover, the utility of existing standardised instruments with known psychometric properties relies on the potentially flawed assumption that properties established on representative general population samples are generalisable to the veterinary profession.

1.4 Where to begin

The potential use of an existing scale could have advantages over the development of a veterinary profession-specific instrument, provided certain psychometric criteria 6 Chapter 1: Introduction

were fulfilled, because it would enable comparison with data from other population groups. The Warwick-Edinburgh Mental Well-being Scale (WEMWBS) (Tennant et al. 2007) is a 14-item instrument developed for assessing positive mental well-being at a general population level. The instrument captures a wide conception of mental well-being including affective-emotional aspects, cognitive-evaluative dimensions and psychological functioning. A copy of the instrument is included in Appendix 1. The scale offers promise as a tool for measuring mental health and well-being at a population level within the veterinary profession.

The cross-sectional study of mental health and well-being reported by Bartram et al. (2009a, b, c) used a questionnaire in which the WEMWBS was embedded with several other instruments which are purported to measure related constructs. The WEMWBS was also embedded subsequently in a principally employment-related questionnaire administered by the Institute of Employment Studies on behalf of the Royal College of Veterinary Surgeons (RCVS): the RCVS Survey of the Profession 2010 (Robertson-Smith et al. 2010).

Assessments of the psychometric properties of WEMWBS to date are limited to student and adult general population samples (Tennant et al. 2007, Stewart-Brown et al. 2009, Clarke et al. 2011).

1.5 Research questions

Three research questions are addressed:

1. Is WEMWBS a valid and reliable instrument with potential as an overall indicator of mental health and well-being in the UK veterinary profession?

2. Is it possible to derive an interval-level measure for WEMWBS in this occupational group?

3. Does the instrument retain its validity as an overall indicator of mental health and well-being when the raw score is transformed to an interval-level measure?

1.6 Aims and objectives

The study aims to examine the measurement properties of WEMWBS in the context of the veterinary profession to inform judgements concerning: Chapter 1: Introduction 7

[i] the defensibility of comparing WEMWBS scores for the veterinary profession with those for other population groups, and

[ii] the potential utility of the instrument for future research, including, for example, the monitoring of clinically meaningful trends in mental health and well-being in this occupational group over time.

The objectives are to:

[a] Explore the properties of WEMWBS using traditional psychometric methods based on classical test theory (CTT).

[b] Evaluate the internal construct validity of WEMWBS using Rasch analysis, attempt to derive an item set which fits model expectations, and compute a raw score to interval scale transformation for the fitting item set.

[c] Re-examine the external construct validity of WEMWBS using the interval- level measurements derived from the Rasch analysis.

1.7 Approach

The study aims to adopt a comprehensive approach to evaluate the psychometric properties of WEMWBS in the context of the veterinary profession, which includes the complementary application of both traditional and modern psychometric methods.

The comprehensive evaluation of a rating scale should involve the examination of six psychometric properties: data quality, scaling assumptions, targeting, reliability, validity and responsiveness (McHorney et al. 1994, Ware and Gandek 1998, Hobart et al. 2001a, Lohr 2002, Hobart and Cano 2009, Cano and Hobart 2011). Although several psychometric analyses provide evidence for more than one property, and there is overlap among properties, this list of properties provides a useful framework to describe the methods and results of rating scale evaluations (Hobart and Cano 2009) and is used to guide evaluation of the measurement properties of WEMWBS in the present study.

To further increase the robustness of the evaluation, analyses were conducted with two large, independently-derived datasets and, where possible, the findings were validated across both samples. 8 Chapter 1: Introduction

The first dataset was derived from a postal questionnaire survey mailed in October/November 2007 to a large stratified random sample of 3200 veterinary surgeons practising in the UK: A cross-sectional study of mental health and well- being and their associations in the UK veterinary profession. This number represents approximately 20% of the membership of the RCVS, excluding those practising overseas or retired. The survey yielded a response rate 56.1% (1796/3200) (Bartram 2009, Bartram et al. 2009b).

The overarching objectives of this survey were to explore the mental health and well-being of the UK veterinary profession to provide empirical evidence to assess their contribution to the elevated suicide risk and inform the development of strategies to improve psychological health and attenuate suicide risk. Analysis of the results was limited to the examination of associations between scores for the instruments employed and demographic or occupational factors such as gender, age, type of work, employment status and number of hours worked and on-call. Academic supervision for the study was provided by Professor D.S. Baldwin, Division of Clinical Neuroscience, Faculty of Medicine, University of Southampton. This work was submitted by David Bartram (DJB) in part fulfilment of the requirements of the RCVS Diploma of Fellowship, awarded November 2009 (Bartram 2009a). Examination of associations between scores on different scales and assessment of the psychometric properties of the scales was outside the scope of this body of work.

The second dataset was derived from the RCVS Survey of the Profession 2010 which was mailed in January/February 2010 to all the veterinary surgeons listed on the RCVS Register. The survey yielded a response rate of 37.4% (8829/23,594). The aim of the survey was to provide the RCVS and other interested parties with contemporary information on the structure, activities and changes within the profession to inform effective governance. Consequently, the content of the questionnaire was mainly employment-related. The decision to embed WEMWBS into the questionnaire was in response to discussions between DJB and the RCVS based on the results from the cross-sectional study of mental health and well-being in the profession reported by Bartram et al. (2009a, b, c). The RCVS recognised the value of routinely including in their surveys an instrument with potential to assess population-level trends in mental health and well-being over time.

Chapter 1: Introduction 9

1.8 Contributions to research and practice

• Extends knowledge of the psychometric properties of WEMWBS by providing supportive evidence for measurement invariance between the general population and an alternative sample population.

• Corroborates existing evidence of the instrument’s psychometric properties derived from student and general population samples.

• Provides evidence to support the validity of WEMWBS as an overall indicator of population mental health and well-being in the UK veterinary profession.

• Derivation of a raw score to interval measurement transformation to facilitate meaningful examination of differences between subgroups of the profession and the interpretation of change scores.

• Provision of a baseline of measures from which to monitor future change and trends in mental health and well-being among veterinary surgeons in the UK.

• Commitment from the RCVS to include WEMWBS in future surveys of the profession and monitor scores over time.

• Contribution to the general body of knowledge on measurement of population-level mental health and well-being in occupational groups.

• Raised awareness of the mental health and well-being within the profession which has prompted the development of targeted interventions, including the development and implementation of the RCVS Health Protocol (RCVS 2010a).

• The quantitative measurement of psychological health within the profession facilitated by this research is complemented by two ongoing qualitative approaches to the identification of possible influences on the elevated suicide risk. The studies were undertaken by the author and data collection is completed. A synopsis of them is given in Appendix 8 but they are not otherwise reported in this thesis.

10 Chapter 1: Introduction

1.9 Document structure

Chapter 1 presents an introduction to the research and its importance for the veterinary profession. Relevant literature relating to suicidal behaviour and mental health among veterinary surgeons is reviewed in Chapter 2. Chapter 3 outlines the theory which underpins techniques to evaluate the quality of psychological measurement instruments. The methods and results of the research are reported in Chapters 4 and 5 respectively. The results are discussed in Chapter 6 and the strengths and limitations of the study are indentified. The chapter concludes with a discussion of the implications for research and the profession of both the psychometric analyses and the literature review of potential influences on the elevated suicide risk. Proposals for interventions which have potential to improve psychological health and attenuate suicide risk are included in the discussion of the implications of the study for the profession. The study conclusions are presented in Chapter 7. Supplementary documentation including: copies of survey instruments; confirmation of ethical approval; the results of tests of association between scales that are not reported elsewhere in the thesis; an outline of related qualitative studies undertaken by the author; and study-related publications are presented in the Appendices. 11

Chapter 2:

Veterinary surgeons and suicide: a structured review of possible influences on increased risk

12 Chapter 2: Structured review of possible influences on increased risk

2.0 Introduction

There have been concerns for many years that mortality due to suicide is higher in the veterinary profession than in the general population (Mellanby 2005).

This chapter reviews the published data on suicide risk in the veterinary profession and compares it with that in other healthcare professions and the general population. Studies of suicide statistics across several different time periods in UK consistently report that the veterinary profession has around four times the proportion of all deaths certified as suicide that would be expected from the proportion for the general population, and around twice that for other healthcare professionals. This observation of increased risk is supported by studies in other countries.

The relative risk of suicide across occupational groups is often explained by differences in demographic factors but veterinary surgeons have higher risk of suicide even when these demographic factors are taken into account (Charlton 1995).

There has been much speculation regarding possible mechanisms underlying increased suicide risk in the profession, but little empirical research.

This chapter presents a narrative synthesis of the central themes reported in the literature as possible influences on the increased suicide risk. A review of current knowledge about possible influences on the suicide rate among veterinarians and factors elevating the risk in other occupations and in the general population is used to propose a hypothetical model to explain suicide risk in veterinary surgeons. Based on testable constructs, it attempts to clarify a complex interaction of possible mechanisms across the career life course and thereby facilitate a more systematic approach to research in this field.

2.1 Literature search methodology

Papers relevant to mental health among veterinary surgeons were identified by searching PubMed (which includes MEDLINE) and Google Scholar for articles published after 1970. Search terms included but were not limited to: veterinary and depression, suicide, stress, burnout, distress, abuse, alcohol drinking, substance- related disorders, psychosocial working conditions, coping or psychiatry. Additional articles were identified by scrutinising the reference lists of relevant articles, hand Chapter 2: Structured review of possible influences on increased risk 13

searching of conference proceedings, examining regular automatic customised e- mail alerts for relevant new articles in journals (Ovid Autoalert), and by consulting experts. The search strategy excluded non-English language records. Articles were included as appropriate to provide a comprehensive structured review of the evidence of increased suicide risk among veterinary surgeons and possible influences on the increased risk, including evidence from other occupational groups. Themes were integrated to inform the development of a hypothetical explanatory model based on testable constructs.

2.2 Suicide statistics

2.2.1 United Kingdom

2.2.1.1 General population

The Office for National Statistics defines ‘suicide’ as deaths given an underlying cause of intentional self-harm and deaths due to self-injury or poisoning of undetermined intent. The latter are included as there is evidence that most injuries and poisonings of undetermined intent (‘open verdicts’) are cases where the harm was self-inflicted but there was insufficient evidence to prove that the deceased deliberately intended to kill themselves. Suicide statistics based on suicide verdicts alone are an underestimate because there is evidence that around 75% of undetermined deaths are actually suicides (Linsley et al. 2001, Brock et al. 2006). There is concern that the increasing use of ‘narrative verdicts’ by coroners over the last decade, in lieu of the ‘short form’ verdicts of suicide or ‘open’, may have reduced the reliability of the statistics (Gunnell et al. 2011).

Suicides represent almost 1 percent of the total of all deaths for ages 15 years and over in the UK: 5675 people killed themselves in 2009, a fall of 10 percent from the 1991 total (Office for National Statistics 2011). Three-quarters of suicides in 2009 were of men (Office for National Statistics 2011). Time-trend analysis shows that the male-to-female sex ratio for suicides has remained around 3:1 since the early 1990s, with fluctuations from over 4:1 in the 1880s to 1.5:1 in the 1960s which are thought to be largely due to variations in the availability and lethality of suicide methods more favoured by women than men (Thomas and Gunnell 2010). Suicide was reported to be the most common cause of death among men aged 15 to 44 years in the general population (Brock and Griffiths 2003, Griffiths et al. 2005, General Register Office for Scotland 2007) and although suicide rates in young men

14 Chapter 2: Structured review of possible influences on increased risk

have declined markedly and are at their lowest level for almost 30 years (Biddle et al. 2008), suicide remains a common cause of death among men in this age range.

2.2.1.2 Veterinary surgeons and other high risk occupations

Several studies from a number of countries have found that members of some occupational groups are at greatly increased risk of suicide (Agerbo et al. 2007), this elevated risk being seen in healthcare professionals including doctors (Hawton et al. 2001, Schernhammer and Colditz 2004, Torre et al. 2005), pharmacists (Kelly and Bunting 1998), dentists (Alexander 2001) and nurses (Hawton and Vislisel 1999). Farmers are also at increased risk (Malmberg et al. 1999).

In Great Britain, the standard method used to make comparisons between the death rates by suicide in different occupations and the general population is by calculation of the proportional mortality ratio (PMR). The PMR is a ratio, often expressed as a percentage, of how more or less likely a death in a particular occupation is to be from suicide as opposed to other causes, than a death of someone of the same age (or within the same specified age range) and gender in the general population as a whole. A PMR of 100 indicates that there is no difference in the ratio of suicide deaths to all deaths in the occupation compared with the general population, while a PMR for suicide of 200 indicates that twice the expected proportion of suicide deaths to all deaths was recorded in that occupational group (Kelly and Bunting 1998).

While PMR is a widely used measure, it should be acknowledged that PMR is affected by the relative frequency of other causes of death. A high PMR can indicate lower mortality from other causes, as well as higher mortality from the cause being examined. Lower mortality from other causes may account for the high PMRs for suicide identified for some high social class occupational groups, including healthcare professionals (Kelly et al. 1995, Kelly and Bunting 1998, Hawton et al. 2001, Meltzer et al. 2008). Moreover, lower mortality from other causes may also explain the apparent paradox between high PMRs for suicide identified for some high social class occupational groups and robust evidence of an inverse relationship between occupational social class and risk of suicide; the lower the social class, the higher the risk (Platt and Hawton 2000). However, Meltzer et al. (2008) reported a high standardised mortality ratio (SMR) for suicide – a population risk statistic Chapter 2: Structured review of possible influences on increased risk 15

unaffected by the frequency of other causes of mortality – for female veterinarians, indicating significantly higher suicide mortality than the general population.1

The absolute number of suicides by veterinary surgeons is low due to the small size of the profession (approximately 16,000 veterinary surgeons practising in the UK), but there is a substantial literature indicating that the profession is at increased risk compared to other occupations and the general population. Awareness of the increased risk has existed for many years. For example, a survey of the causes of mortality in veterinarians resident in Britain followed up from 1949-1953 to 1975 reported a two-fold increase in mortality from suicide (Kinlen 1983).

On the basis of proportional mortality ratios (PMRs) in England and Wales (Charlton et al. 1993, Kelly et al. 1995, Kelly and Bunting 1998, Mellanby 2005, Meltzer et al. 2008) and Scotland (Stark et al. 2006a), the chances of a death in this occupational group being due to suicide are around four times that of the general population, and around twice that of other healthcare professionals.

PMRs for the highest risk occupations for men are presented in Table 2.1 (England and Wales) and Table 2.2 (Scotland). The PMR for veterinary surgeons is consistently one of the highest.

1 SMR: 431, 95% CI: 140 to 1005 (England and Wales, 2001-2005)

16 Chapter 2: Structured review of possible influences on increased risk

Table 2.1: High risk occupational groups for suicide (suicide and open verdicts), men aged 16 to 64 years 1979-1990, men aged 20 to 64 years 1982-1987, 1991-1996 and 2001-2005, England and Wales

95% PMR confidence interval

1979-1990 1 Veterinarians 364 NR Pharmacists 217 NR Dentists 204 NR Farmers 187 NR Medical practitioners 184 NR

1982-1987 2 Veterinarians 349 203 to 559 Farmers, horticulturalists, farm 202 180 to 226 managers Librarians, information officers 226 140 to 345 Pharmacists 214 140 to 313 Medical practitioners 175 138 to 218 Dental practitioners 192 117 to 296

1991-1996 2 Dental practitioners 249 161 to 367 Veterinarians 324 148 to 615 Farmers, horticulturalists, farm 144 124 to 166 managers Medical practitioners 147 115 to 185 Garage proprietors 155 112 to 208 Pharmacists 171 111 to 252

2001-2005 §, 3 Dental practitioners 292 173 to 461 Farmers 189 157 to 227 Beauticians and related occupations 297 143 to 547 Medical practitioners 165 125 to 214 Artists 171 124 to 229

Within each year range, occupations are listed in descending order of lower bound of 95% confidence interval so that those most significantly different from the general population appear at the top. § Note: The lower bound of the 95 percent confidence interval of suicide PMR for veterinarians 2001-2005 did not exceed 100 and is therefore not reported by Meltzer et al. (2008). Sources: 1 Charlton et al. (1993); 2 Kelly and Bunting (1998); 3 Meltzer et al. (2008) NR, Not reported Chapter 2: Structured review of possible influences on increased risk 17

Table 2.2: Suicide and undetermined intent deaths for men aged 16 to 45 years and 46 to 64 years in Scotland in 1981-1999

16- to 45-year old men 46- to 64-year old men 95% 95% PMR PMR confidence interval confidence interval

1981-1999 Veterinarians 293 80 to 749 301 36 to 1088 Medical practitioners 180 118 to 265 205 109 to 351 Dentists 128 26 to 374 235 76 to 548 Pharmacists 43 1 to 238 118 14 to 242

Note: When the 95% confidence interval includes 100, the difference between the PMR for an occupation and the general population is not statistically significant (p<0.05). Source: Stark et al. (2006a)

Mellanby (2005) undertook a detailed analysis using PMRs for deaths by suicide in England and Wales among veterinarians, medical practitioners and dental practitioners. The results are summarised in Table 2.3. For both time periods, the PMR was markedly higher for veterinarians. The PMR for suicide among male veterinarians remained stable over time. By contrast, the PMR for suicide among females increased considerably but this must be interpreted with caution due to the small absolute number of deaths involved. Meltzer et al. (2008) also reported a high PMR for suicide among women veterinarians in 2001-2005 (PMR: 609, 95% CI: 198 to 1422).

Mellanby (2005) was unable to complete a detailed analysis of the age, background, previous health history and type of veterinary work undertaken due to restrictions imposed to protect individuals’ privacy. However, suicides by veterinary surgeons do not appear to be confined to a restricted age range (Mellanby 2005, Stark et al. 2006a) as the age group distribution of suicides is not substantially different from all occupations combined (Kelly et al. 1995).

18 Chapter 2: Structured review of possible influences on increased risk

PMR 95% confidence interval

1979 -1990 (excluding 1981) Male Veterinarians 361 252 to 503 Medical practitioners 162 106 to 262 Dental practitioners 194 137 to 266

Female Veterinarians 414 166 to 853 Medical practitioners 193 143 to 254

1991- 2000 Male Veterinarians 374 244 to 548 Medical practitioners 175 147 to 206 Dental practitioners 227 162 to 309

Female Veterinarians 1240 446 to 2710 Medical practitioners 462 339 to 614

Data were incomplete for 1981 so this year was excluded from the analyses Source: Mellanby (2005)

There are many difficulties when comparing the risk of suicide across occupational groups. These include the effects of different socio-demographic factors both between occupations and within occupational specialisms. It is important that confounding by such factors is adequately controlled (Wilhelm et al. 2004). While most differences in suicide risk between occupations are accounted for by differences in income and employment status, the most striking exceptions are for veterinary surgeons, doctors, nurses and pharmacists, all having significantly higher rates of suicide even when demographic factors are taken into account (Charlton 1995, Stack 2001, Agerbo et al. 2007). Charlton (1995) linked data from death certificates and characteristics of the electoral ward of the deceased individuals’ usual residence to undertake a case-control analysis of suicide risk. Table 2.4 presents the relative risk of suicide among veterinarians and other high risk occupational groups after controlling for demographic factors. The risk of suicide relative to death from other causes, among male veterinarians aged 16 to 44 years and 45 to 64 years and female veterinarians aged 16 to 64 years, was elevated by Chapter 2: Structured review of possible influences on increased risk 19

4.6, 5.6 and 7.6 times respectively, in comparison with individuals in the general population with similar demographic characteristics.

Men aged 16-44 Men aged 45-64 Women aged 16-64‡ 95% 95% 95% RR confidence RR confidence RR confidence interval interval interval

Veterinarians 4.61 1.49 to 14.25* 5.62 1.60 to 19.74* 7.62 1.04 to 55.94* Pharmacists 1.15 0.37 to 3.52 4.15 2.00 to 8.58** 1.21 0.27 to 5.35 Dentists 2.26 0.93 to 5.47 5.19 2.29 to 11.76** - - Farmers 0.88 0.60 to 1.30 1.93 1.48 to 2.51** - - Medical * ** 1.50 0.90 to 2.50 2.22 1.35 to 3.65 4.54 2.54 to 8.13 practitioners

Risk is relative to deaths from other causes Reference groups are married; UK-born; corresponding age range; not in the 10 highest occupational groups; in urban wards with owner occupancy >85%; unemployment <5%; <6.5% of occupants changing address per year; <9% of adults below pensionable age living alone ‡ No significant model differences for women under and over age 45 * p<0.01; ** p<0.001 Source: Charlton (1995)

Table 2.5 shows for men and women the distribution of suicides by method for veterinary surgeons and all occupations combined. Deliberate self-poisoning is the most common method of suicide in male and female veterinarians, accounting for 76 and 89 percent of suicides respectively, compared to 20 and 46 percent respectively of suicides in the general population of England and Wales (Kelly and Bunting 1998). Veterinary surgeons and pharmacists have the highest proportions of suicides using this method for all occupational groups; medical practitioners also have an increased risk of this specific method of suicide (Kelly and Bunting 1998, Hawton et al. 2000, Agerbo et al. 2007). There are many reports of the use of medicines for veterinary anaesthesia and euthanasia as suicide agents (for example, Clark and Jones 1979, Cordell et al. 1986, Smith and Lewis 1989, Stowell 1998, Elliott and Hale 1999, Résière et al. 2001, Kintz et al. 2002, Romain et al. 2003, Elejalde et al. 2003, Sterken et al. 2004). Barbiturates are the drugs most

20 Chapter 2: Structured review of possible influences on increased risk

commonly used for suicide by doctors (Hawton et al. 2000) and were used by at least half of male veterinary surgeons who died by suicide by deliberate self- poisoning between 1982 and 1996 in England and Wales (Kelly et al. 1995). Firearms are the second most common method of suicide by male veterinary surgeons, the proportional use of which was also raised relative to the general population, accounting for 16 percent and 5 percent of suicides respectively (Kelly and Bunting 1998).

Suicide method Poisoning Poisoning Hanging Firearms by solid by other and Drowning and Other or liquid gases and suffocation explosives substances vapours

Male

Veterinarians 76 3 5 0 16 0 All men 20 27 27 6 5 16

Female

Veterinarians 89 11 0 0 - 0 All women 46 10 17 9 - 18

Note: Percentages may not add to 100 due to rounding Source: Kelly and Bunting (1998)

2.2.2 Rest of the world

International comparisons of suicide risk by occupation are hindered by variation between countries in the classification and recording of suicides (Andriessen 2006) and the use of a range of different measures of suicide risk across the literature (Platt et al. 2010b).

A study of mortality patterns among US veterinarians from 1947 to 1977 showed significantly elevated mortality from suicide relative to death from other causes for males (PMR: 170), particularly for those in small animal practice (PMR: 357), and with a higher proportion of self-poisonings compared with the general population (Blair and Hayes 1980, Blair and Hayes 1982). Barbiturates were the drugs most commonly used for suicide. In a separate study in California, male and female Chapter 2: Structured review of possible influences on increased risk 21

veterinarians had significantly elevated mortality from suicide of, respectively, 2.5 times and 5.9 times the rate in the general population of that state. Mortality from suicide was significantly elevated for veterinarians who had worked in the profession for fewer than 30 years (Miller and Beaumont 1995). Elevated mortality from suicide relative to death from other causes is also reported for male (PMR: 197) and female (PMR: 180) veterinarians in Washington State, from 1950 to 1999 and 1974 to 1999 respectively (Milham and Ossiander 2001).

In a large study in Norway which examined suicide deaths by occupation over a 40- year period (Hem et al. 2005), the highest rate was among male veterinarians with a suicide rate approaching twice that in the general population. Suicide rates for different healthcare professions are summarised in Table 2.6.

Table 2.6: Suicide rates for men aged over 20 years per 100,000 person years in Norway in 1960-2000

Suicide rate 95% per 100,000 person years confidence interval

Veterinarians 44 25 to 75 Physicians 43 35 to 53 Dentists 33 23 to 46 Pharmacists 29 13 to 64

General population 23 23 to 24

Source: Hem et al. (2005)

Using a similar method, a suicide rate among veterinary surgeons of 45 per 100,000 person years (95% CI: 25 to 82 per 100,000) was reported for two Australian states combined, approximately four times the rate in the general population for the states concerned (Fairnie 2005, Jones-Fairnie et al. 2008). The principal method of suicide was self-poisoning by drugs, mainly injectable barbiturates, and the suicides were not confined to a restricted age range; the age distribution of suicides for veterinary surgeons was not substantially different from the Australian general population.

Hawton et al. (2011) used data from Danish national registers for 1981-2006 to examine risk of suicide in nurses, physicians, dentists, pharmacists and veterinary surgeons compared to teachers and the general population. Crude age- and gender-adjusted rate ratios for suicide compared to teachers were significantly elevated in nurses (RR: 1.90, 95% CI: 1.63 to 2.21), physicians (RR: 1.87, 95% CI:

22 Chapter 2: Structured review of possible influences on increased risk

1.55 to 2.26), dentists (RR: 2.10, 95% CI: 1.58 to 2.79) and pharmacists (RR: 1.91, 95% CI: 1.26 to 2.87), but not veterinary surgeons (RR: 1.04, 95% CI 0.63 to 1.74).

Similarly, the risk of suicide among veterinarians, as assessed by standardised mortality ratios, was not elevated in comparison with the total employed population in an examination of suicide by identified occupational groups in New Zealand over a period of 30 years (Skegg et al. 2010).

It is interesting that in contrast to other countries, there was no indication in Denmark and New Zealand of elevated risk in veterinarians. This suggests that comparative studies across a range of countries of the prevalence of potential risk factors for suicidal behaviour may be of value to help explain the differences.

2.3 Risk factors for suicide and suicidal behaviour

Suicide and suicidal behaviour is thought to result from a complex interplay of risk and protective factors that are biological, psychological, and social in nature. Mental health disorders play the strongest role in the aetiology of suicidal behaviour. Psychological autopsy studies report that around 90 percent of those dying by suicide or making serious suicide attempts have a mental health disorder, most frequently depression and/or alcohol misuse (Cavanagh et al. 2003). However, while a mental health disorder is generally a necessary condition for suicide, it is not sufficient. Most psychiatric patients do not die by suicide (Mann et al. 1999). Additional risk and protective factors are also involved.

An understanding of risk factors for suicide and suicidal behaviour in the general population provides a foundation for elaborating possible mechanisms to explain suicidal behaviour within a specific occupational group.

The risk factors for suicide act either distally (‘upstream’), or proximally (‘downstream’). Distal risk factors affect the threshold for suicide and indirectly increase an individual’s risk when they experience a proximal risk factor. Proximal or trigger factors are temporally closer to the suicidal behaviour and often act as precipitants (Mościcki 1997). The risk factors for suicide and suicidal behaviour can be broadly categorised into social and demographic, psychiatric, situational, biological and psychological and are summarised in Tables 2.7 to 2.11 respectively. Chapter 2: Structured review of possible influences on increased risk 23

Table 2.7: Social and demographic risk factors for suicide and suicidal behaviour

Social and demographic Age The highest suicide rates in the UK are among young men aged 15- 44 years. Among women, the highest rates are for those aged 75 years and over (Brock et al. 2006) Gender Almost three-quarters of suicides in the UK are of men (Brock et al. 2006). Women are twice as likely to experience major depression but their greater inclination to consult friends, readiness to accept help and willingness to change their minds helps protect against suicide (Murphy 1998). Women are more likely to make non-fatal suicide attempts (Miller et al. 2004) Marital status The suicide rate for divorced, widowed or single (never married) men and women is around three times higher than for married men and women (Griffiths et al. 2008) Education Risks are elevated among individuals who have poor or limited education (Gould et al. 1996) Religion Risks are elevated among those without religious affiliation (Neeleman 1998) Socio-economic Risks are inversely related to social class (the lower the class, the status higher the rate) (Platt and Hawton 2000) Unemployment Risks are higher among the unemployed (Platt and Hawton 2000) Occupational group Risk is higher for those working in medical and allied professions, farmers and nurses (Platt and Hawton 2000) Rural location Greater risk of male suicide in remote rural areas relative to urban areas and lower risk of female suicide in accessible rural areas (Levin and Leyland 2005). Rural location may create geographic, psychological, and socio-cultural barriers to the treatment of suicidal thoughts and behaviours (Hirsch 2006) Ethnic origin Suicide rate varies between countries and between ethnic groups within countries (McKenzie et al. 2003)

Table 2.8: Psychiatric risk factors for suicide and suicidal behaviour

Psychiatric Mood disorders Mood disorders are the mental health disorders most commonly associated with suicide. Risk of suicide is increased twenty-fold for those with major depression (Harris and Barraclough 1997) Substance misuse Alcohol misuse predisposes to suicidal behaviour through its depressogenic and disinhibitory effects and promotion of adverse life events (Brady 2006) Anxiety disorders Anxiety disorders are an independent risk factor for subsequent onset of suicidal ideation and attempts (Sareen et al. 2005). Co- morbid anxiety disorders amplify the risk of suicidal behaviour in persons with mood disorders (Sareen et al. 2005, Hawgood and De Leo 2008) Schizophrenia Risk of suicide is increased thirty- to forty-fold for those with schizophrenia (Harris and Barraclough 1997) Personality Personality disorders are associated with increased risk of suicide disorders and suicide attempt, independently and co-morbidly with other mental health disorders (Schneider et al. 2006) Co-morbidity Co-existence of multiple mental health disorders is associated with increased suicide risk (Cavanagh et al. 2003) Previous suicidal Individuals who make non-fatal suicide attempts are at higher risk of behaviour further suicide attempts and completed suicide (Hawton et al. 2003)

24 Chapter 2: Structured review of possible influences on increased risk

Table 2.9: Situational risk factors for suicide and suicidal behaviour

Situational

Stigma Stigma associated with mental illness can reduce help-seeking behaviour (Dinos et al. 2004) Lack of social The association between poor family and/or social integration and support suicide is robust and largely independent of the presence of mental health disorders (Duberstein et al. 2004) Influence of others Direct or indirect exposure to the suicidal behaviour of others can in some circumstances influence attitudes and increase vulnerability to suicide (Maris et al. 2000) Adverse life events Suicidal behaviour is often preceded by exposure to stressful life events (Baca-Garcia et al. 2007) Access to lethal Accessibility of lethal means has a strong influence on suicide rate means (Hawton 2007) Major physical There is an association between suicidal behaviour and physical illness illnesses including AIDS, pulmonary disease, cancer and neurological disorders such as stroke, multiple sclerosis and some forms of epilepsy (Druss and Pincus 2000, Verrotti et al. 2008) Imprisonment There is a five-fold excess of suicides in the prison population of England and Wales (Fazel et al. 2005)

Table 2.10: Biological risk factors for suicide and suicidal behaviour

Biological

Family history of Suicidal behaviour is highly familial and, on the basis of twin and suicidal behaviour adoption studies, heritable (Brent and Mann 2005) Neurotransmitter Neurotransmitter system dysregulation, especially serotonin, is system associated with increased risk of suicidal behaviour (Mann 2003) dysregulation Menstrual cycle The risk of suicide attempt may increase in phases of the menstrual cycle which have lower oestrogen levels (Saunders and Hawton 2006)

Table 2.11: Psychological risk factors for suicide and suicidal behaviour

Psychological

Personality The personality traits of hopelessness, neuroticism, and extroversion characteristics are associated with suicidality (Brezo et al. 2006) Perfectionism There is considerable evidence that aspects of perfectionism are associated with suicidality (O’Connor 2007) Cognitive style Suicide is associated with a constriction in cognitive style which leads to impairments in problem-solving and positive future cognitions (Sheehy and O’Connor 2008) Over-general Depressed and suicidal individuals show a reduced ability to be autobiographical specific in recalling personal memories (Williams and Broadbent memory 1986)

Chapter 2: Structured review of possible influences on increased risk 25

2.4 Protective factors against suicide and suicidal behaviour

Protective factors are those which are thought to buffer against suicidal behaviour during periods of risk such as adverse life events. Empirical evidence on risk factors for suicide and suicidal behaviour is far more abundant than that on protective factors, which remain largely at a theoretical level of discussion (Bertolote 2004).

Those protective factors which are not merely positive configurations of the risk factors outlined above are listed in Table 2.12 below.

Table 2.12: Protective factors against suicide and suicidal behaviour

Protective factors

Reasons for living (e.g. parenthood) Good emotional relationship Survival and coping skills and beliefs Family and social connectedness Problem-solving confidence Internal locus of control Fear of social disapproval Moral objections to suicide Religious participation Physical activity and health

Sources: Malone et al. 2000, Donald et al. 2005, McLean et al. 2008.

2.5 Conceptual models of suicidal behaviour

Many explanatory and predictive models of suicidal ideation and behaviour have been proposed; these have included sociological, psychiatric, biological and psychological explanations. Four models are described in outline here, in order to offer a theoretical context in which to understand the mechanisms that may underlie the possible influences on suicide risk that occur across the course of a veterinary career.

The stress-diathesis model (Mann et al. 1999) posits that suicidal behaviour results from the interaction between stressful life events and an individual predisposition or vulnerability. This vulnerability, itself the product of psychobiological factors, genetic predisposition and past life events, influences how the individual perceives,

26 Chapter 2: Structured review of possible influences on increased risk

interprets and reacts to adverse life events. It is a dynamic system in which stress and diathesis influence each other (van Heeringen 2002).

Consistent with the stress-diathesis model, the ‘cry of pain’ model of suicidal behaviour (also know as the ‘arrested flight’ model) (Williams and Pollock 2000, Williams 2001) conceptualises suicidal behaviour as the response (the ‘cry’) to a stressful situation in which environmental cues result in feelings of defeat, loss or humiliation that give rise to an overwhelming feeling of needing to escape, a sense of being unable to escape (entrapment) and a sense that this state of affairs will continue indefinitely (no prospect of rescue). Judgements regarding perceptions of defeat, entrapment and rescue are determined, at least in part, by psychological variables such as problem solving and the ability to think positively about the future (Williams and Pollock 2000).

The interpersonal-psychological theory of suicide (Riberio and Joiner 2009) proposes that feelings of perceived burdensomeness and thwarted belongingness contribute to suicidality. Furthermore, they suggest that death by suicide is only likely to occur when an individual has acquired the courage and capability to commit suicide through exposure to events that have led to a habituation to pain and fear of death. This habituation process may occur through prior suicidal behaviour in oneself or exposure to suicidal behaviour in others, or through the experience of other traumatic, painful or provocative events. The interpersonal-psychological theory therefore suggests the specific circumstances in which a person may engage in a suicidal act with fatal outcome.

The relationship between occupation and suicide can be conceptualised as a model comprising four components contributing to the differential occupational suicide risk: demographics (the demographic composition of people in the occupation); internal occupational stress (stress associated with the nature of the work); pre-existing psychiatric morbidity (the psychological profile of individuals attracted to the occupation); and opportunity factors (opportunities available for access to lethal means of suicide) (Stack 2001).

Suicidal behaviour can be conceptualised as a continuum of gradually increasing seriousness: feelings that life is not worth living, thoughts of taking one’s own life, seriously considering suicide, suicidal planning, suicidal attempt, and suicide completion (Paykel et al. 1974, Perroud et al. 2010). Another view considers that each of these forms of suicidal behaviour does not together represent a continuum Chapter 2: Structured review of possible influences on increased risk 27

but is instead a discrete category, each with a specific risk factor (Duberstein et al. 2000). A longitudinal study of suicidal planning among doctors suggests that a continuum of severity may operate in this occupational group (Tyssen et al. 2004). Consistent with the continuum hypothesis, for the purpose of this structured review, risk factors for suicidal behaviours in general are considered as possible influences on the increased risk of suicide among veterinary surgeons.

2.6 Possible influences on suicide risk among veterinary surgeons

Risk factors for suicide include depression, alcohol and drug misuse, inherited factors, certain personality traits, and environmental factors such as chronic major difficulties, and undesirable life events (Goldney 2005). Several circumstances may elevate risk in specific occupations and, although the reasons are unclear, an interplay between various potentially malign influences is suggested for the veterinary profession.

The interrelations of work, personality and mental health are well documented (Stansfeld 2002) but reports specific to the veterinary profession (such as Halliwell and Hoskin 2005) have tended to simply present the observations and opinions of concerned individuals. These commentaries offer prima facie compelling concepts, but there is a paucity of research to test their veracity. Little is known about the scope and magnitude of problematic outcomes, possible predisposing factors, or effective interventions in veterinarians.

2.6.1 Access to means of suicide

Availability and acceptability are viewed as two key factors that influence suicide method choice and, therefore, key considerations to suicidal individuals planning an attempt (Farmer and Rohde 1980). Suicidal impulses are often brief and, at the point at which a person feels hopeless and suicidal, ready access to means of suicide may be the key factor that influences translation of suicidal thoughts into an actual suicide act (Hawton 2007). Access to potentially lethal means has a strong influence on overall suicide rate because there is evidence of limited method substitution when access is impaired or removed – decreases in overall suicide rate have been associated with changes to non-toxic domestic gas from coal gas, installation of catalytic converters in cars, smaller over-the-counter pack sizes of paracetamol, installation of barriers on high bridges and the withdrawal of co-proxamol prescribing

28 Chapter 2: Structured review of possible influences on increased risk

(Hawton et al. 2004b, Bennewith et al. 2007, Hawton 2007, Hawton et al. 2009, Thomas and Gunnell 2010).

The rapid rise in the use of domestic coal gas as a method of suicide following its introduction at the end of the 19th century was not offset by declines in the use of other methods, demonstrating how quickly access to and knowledge of a new method can become established within a population to increase overall suicide rates (Thomas and Gunnell 2010, Thomas et al. 2011). A more recent example is the rapid increase in suicides in Hong Kong and urban Taiwan by carbon monoxide poisoning resulting from barbecue charcoal burning in an enclosed space which followed high profile reporting of a case which occurred in Hong Kong in 1998 (Liu et al. 2007, Yip and Lee 2007, Thomas et al. 2011, Law et al. 2011).

Availability and knowledge of medicines is likely to contribute to the suicide risk in doctors (Hawton et al. 2000). Veterinary surgeons also have ready access to medicines with potential for high lethality, because they are typically stored in practice premises, and knowledge of medicines for self-poisoning, offering possible contributory factors for the high suicide risk. Veterinary surgeons are also less supervised in their use of medicines than are doctors (Fishbain 1986).

Methods of suicide for occupational groups are described in Section 2.1.1.2 and Table 2.5. Deliberate self-poisoning is the most common method of suicide in both male and female veterinarians, accounting for 76 and 89 percent of suicides respectively, compared to 20 and 46 percent respectively of suicides in the general population (Kelly and Bunting 1998). Veterinary surgeons and pharmacists have the highest proportions of suicides using this method of all occupational groups; medical practitioners also have an increased risk of this specific method of suicide (Kelly and Bunting 1998, Hawton et al. 2000, Agerbo et al. 2007, Hawton et al. 2011).

An examination of suicide by identified occupational groups in New Zealand over a period of 30 years, demonstrated that nurses, doctors , pharmacists and veterinarians were significantly more likely to use poisoning than were other employed people (3, 4, 5 and 8 times respectively, compared with all others employed) (Skegg et al. 2010). However, suicide risk was not elevated among doctors and veterinarians. This suggests that, while access to means appears to influence the method of suicide chosen, it may not necessarily be a determinant of increased occupational suicide risk, perhaps because of the presence of other factors that confer protection. This is consistent with the observations of Agerbo et Chapter 2: Structured review of possible influences on increased risk 29

al. (2007) who reported that when deaths by medicines were excluded, suicide rates in Danish doctors remained slightly elevated, suggesting that ready availability of lethal means is not the only factor operating to increase their occupational suicide risk (Reichenberg and MacCabe 2007).

Veterinary surgeons working in equine, farm animal and mixed practice have ready access to firearms for euthanasia of large animals. Firearms are the second most common method of suicide by male veterinary surgeons in the UK. The use of this method of suicide is raised relative to the general population, accounting for 16 percent and 5 percent of suicides respectively (Kelly and Bunting 1998). Hawton et al. (2011) observed that whilst numbers were small, there was some indication of greater use of firearms for suicide in veterinarians in Denmark.

Farmers who die by suicide also tend to use methods to which they have easy access, especially firearms (Hawton et al. 1998, Stark et al. 2006b).

2.6.2 Attitudes to death and euthanasia

Veterinary surgeons are frequently responsible for ending the lives of animals, either directly in the case of euthanasia, or indirectly in the case of slaughter of meat- producing livestock. The emotional intensity of the relationships that often develop between people and their pets is such that it is the veterinary profession’s routine experience to discuss, justify the legitimacy of, and ultimately administer euthanasia to animals that are considered by their owners to be virtual persons (Sanders 1995). Familiarity with death and dying may affect attitudes within the profession in regard to the expendability of human life (Charlton et al. 1993). Ninety-three percent of veterinary healthcare workers interviewed in a small-scale study indicated a favourable inclination towards euthanasia of humans (Kirwan 2005). This is a higher proportion than reported in the general population (Clery et al. 2007) and contrasts with prevailing medical opinion (Seale 2009), although comparison of these studies is confounded by dissimilar research methods. Positive associations have been demonstrated between tolerance of suicide (more permissive attitudes towards euthanasia, physician-assisted suicide and unassisted suicide) and suicidal thoughts and behaviour (Neeleman et al. 1997, Etzersdorfer et al. 1998, Zemaitiene and Zaborskis 2005, Gibb et al. 2006, Joe et al. 2007, Eskin et al. 2011).

The theory of cognitive dissonance (Harmon-Jones and Harmon-Jones 2007) – that psychological discomfort arising from conflicting thoughts or beliefs motivates the

30 Chapter 2: Structured review of possible influences on increased risk

modification of existing, or the acquisition of new, thoughts and beliefs to reduce the inconsistency and discomfort – may offer an explanation for any effect of euthanasia attitudes on suicide risk. Veterinary surgeons may experience uncomfortable tensions between their desire to preserve life and an inability to treat a case effectively, which could be ameliorated by modifying attitudes to preserving life to perceive euthanasia as a positive outcome. This altered attitude to death may then facilitate self-justification and lower inhibitions towards suicide as a rational solution to their own problems.

2.6.3 Suicide ‘contagion’

Direct or indirect exposure to the suicidal behaviour of others can in some circumstances influence attitudes and increase vulnerability to suicide (Maris et al. 2000). Knowledge of individual suicides can travel readily through the social networks of a small profession and awareness of specific members of the profession who have completed suicide, or of high levels of suicide in professional peers generally, may normalise suicide and be a contributory risk factor for suicidal behaviour in veterinary surgeons, creating a suicide ‘contagion’ effect among vulnerable individuals within this high-risk occupational group.

2.6.4 Cognitive and personality factors

Cognitive and personality factors are relevant in the understanding of psychological distress and may be important in the prediction of suicide risk (Sheehy and O’Connor 2008). These factors include problem-solving deficits, memory and thinking biases, negative cognitive style, neuroticism, impulsivity, hopelessness, self-criticism and rumination (Williams and Pollock 2002, O’Connor 2007, O’Connor and Noyce 2008).

The relationship between personality factors and vocational interest is well documented (Mount et al. 2005). Individuals may have a preference for, or be preferentially selected into, certain occupations based on their personality or life experiences, which could render them either more vulnerable or resilient to the work environment (Kohn and Schooler 1982, Stansfeld 2002, Wilhelm et al. 2004). Personality profiles of medical students (Meit et al. 2007) and doctors (Clack et al. 2004) differ from those of the general population and there are associations between personality factors and choice of medical speciality (Borges and Savickas 2002): for example, mental health workers are more likely to have early experiences Chapter 2: Structured review of possible influences on increased risk 31

of childhood trauma and family dysfunction than other professions (Elliott and Guy 1993); and women psychiatrists are more likely than women in other medical specialities to report personal or family histories of psychiatric disorder (Frank et al. 2001). It is possible that such self-selection operates in the veterinary profession. The choice of a veterinary career may be subconsciously influenced by factors such as a preference for working with animals rather than people – previous interactions with animals play a critical role in guiding veterinary students into their chosen career (Martin et al. 2003, Martin and Taunton 2005, Serpell 2005, Tomlin et al. 2010a, Amass et al. 2011).

The profession may be particularly vulnerable to suicide because of selection based on the very high academic entry requirements into veterinary schools (Halliwell and Hoskin 2005). Admission requirements lead to an undergraduate cohort in which students with a history of outperforming their peers are grouped with similarly high academic achievers to create an environment in which some students chronically question their own abilities and fear that they will be exposed as intellectual frauds (Zenner et al. 2005). Similar sentiments were expressed frequently by US veterinary students during counselling (Kogan et al. 2005). This situation also applies in students of other health professions and predisposes to psychological problems among those who do not adjust and ensure their expectations of their performance are realistic (Henning et al. 1998).

First-year students at a US veterinary school possessed psychological characteristics consistent with other high-achieving competitive performance populations such as professional athletes (Zenner et al. 2005): students had elevated anxiety levels, placed significant value on positive comparisons with the competence of peers, and harboured a fear of failure. Certain dimensions of perfectionism – high personal standards and perception of parental expectations – were elevated in this student population. There is considerable evidence that aspects of perfectionism including socially prescribed perfectionism (a belief that others hold unrealistic and exaggerated expectations of us which must be met in order to gain acceptance and approval), self-criticism, concern about mistakes, and doubts about action, are associated with suicidality (O’Connor 2007).

Voracek (2004) reported a positive association of intelligence with suicide, but others (such as Gunnell et al. 2005, Andersson et al. 2008) have reported inverse associations. Cognitive performance in childhood appears to be significantly and

32 Chapter 2: Structured review of possible influences on increased risk

inversely related to morbidity and mortality in adulthood, even at the higher end of the intelligence continuum and independent of childhood socio-economic status (Martin and Kubzansky 2005).

Halliwell and Hoskin (2005) conjectured that the highly demanding veterinary undergraduate course has the potential to stifle the development of communication skills and emotional maturity, possibly more so than in the medical curriculum. The considerable volume and pace of learning at veterinary school is a stressor for students (Gelberg and Gelberg 2005). Rice (2008) speculated that a factor influencing the elevated suicide risk could be that undergraduate admission criteria based on high academic achievement might select a greater proportion of students with low ‘emotional intelligence’ (EI), described as the ability to perceive emotions in oneself and others, integrate emotions into thought processes, understand emotions, and moderate emotions in oneself and others (Mayer and Salovey 1997). However, there is a strong positive correlation between several dimensions of EI and academic achievement (for example, Parker et al. 2004, Petrides et al. 2004, Austin et al. 2005) so it seems unlikely that the current admissions procedure inevitably selects a high proportion of students with low EI. A recent meta-analysis of 44 studies demonstrated that EI is positively associated with mental health (Schutte et al. 2007). Although the methodologies of the meta-analysis and the studies on which it is based do not provide evidence regarding causality, it may be that the better perception, understanding, and management of emotion among individuals with higher EI make it less likely that they will experience mental health problems. Fostering EI may have a place in veterinary education to improve the veterinary surgeon-client relationship and act as a buffer against stress in the profession (Gelberg and Gelberg 2005, Timmins 2006). However, Timmins (2006) cautions that the concept of EI is relatively new and is not without controversy. There are disagreements regarding the definition of the term and the reliability and validity of available instruments for measuring it and no direct evidence that improving the EI of students or veterinarians will result in specific outcomes that have been identified as desirable for the profession.

Personality traits of hopelessness, neuroticism and extraversion may be major influences on suicide (Brezo et al. 2006). A longitudinal study of a large cohort of UK medical graduates showed that career stress, burnout and satisfaction were predicted by trait measures of personality assessed five years earlier as medical students (McManus et al. 2004). A six-year longitudinal and nationwide study of Chapter 2: Structured review of possible influences on increased risk 33

Norwegian medical students demonstrated that the traits of neuroticism and high conscientiousness are risk factors for stress (Tyssen et al. 2007). Tyssen et al. (2001a, 2001b, 2004) have shown that specific personality traits and depression are common predictors of mental health problems and suicidal ideation and behaviours among medical students and young doctors, and demonstrated the utility of screening final year medical students to identify a sub-group suitable for intervention.

Little is known about the coping strategies used by veterinary surgeons. Brown (1994) demonstrated that students at a US veterinary school perceived themselves as deploying a range of both problem- and emotion-focused coping styles with relatively equal frequency. By contrast, Australian veterinary students did not consistently employ a range of effective coping strategies to deal with the stressors they encountered during their course of study (Williams et al. 2005). Veterinary surgeons in New Zealand have been shown to make good use of their social networks to seek information and assistance in times of work-related stress, especially from informal sources such as friends, family and colleagues, rather than from resources such as health professionals, counsellors and telephone helplines (Gardner and Hini 2006). The importance of deploying both problem- and emotion- focused strategies to cope with the stresses of veterinary work has been emphasised (Bartram and Gardner 2008).

Studies of the learning styles and personality traits of veterinary undergraduates may help determine whether it is the job itself or the type of person undertaking the job that more significantly influences well-being within the profession (Tomlin et al. 2010b).

2.6.5 Work-related stressors

The demand-control model (Karasek 1979) and the subsequent demand-control- support model (Karasek and Theorell 1990) of job strain predict that the combination of high job demands, low decision latitude and low social support is associated with mental and physical illness. Decision latitude includes decision authority (or control over the working environment) and skill discretion (or variety of work and opportunity for use of skills). High levels of decision latitude have been found to be protective of mental health in cross-sectional studies of veterinarians (Hesketh and Shouksmith 1986). The central tenet of a complementary job stress model, the effort-reward imbalance model (Siegrist 1996), is that the lack of contractual reciprocity in working

34 Chapter 2: Structured review of possible influences on increased risk

conditions characterised by high effort in combination with low reward (financial and career-related rewards, esteem and job security) has adverse mental and physical health consequences, especially in individuals characterised by a motivational pattern of overcommitment to their work.

Work factors associated with psychological ill-health in staff are work demands (long hours worked, work overload and pressure) and the associated effects on personal lives; lack of control over work; lack of participation in decision-making; low occupational social support; and unclear management and work role (Michie and Williams 2003). There is robust, consistent evidence that the combination of high psychological demands and low decision latitude and the combination of high effort at work and low reward are prospective risk factors for common mental disorders (Stansfeld and Candy 2006, Bonde 2008, Netterstrøm et al. 2008, Siegrist 2008).

There is an association between work stress and rates of depression and anxiety: individuals, without any pre-employment history of psychiatric disorders, exposed to high psychological job demands have a two-fold risk of onset of new depression and anxiety compared to those with low job demands (Melchior et al. 2007). Lack of support from colleagues and supervisors and deterioration in work characteristics are also associated with onset of depression (Waldenström et al. 2008).

One putative mechanism underlying associations between work characteristics and mental ill-health is neuroendocrine disturbance, mediated or moderated through psychological pathways including self-esteem and feelings of mastery over work (Stansfeld and Candy 2006).

Veterinary graduates typically move abruptly from the university environment to the relative professional and social isolation of general private practice. Being on-call, financial aspects of the role, and lack of surgical competence are identified as main initial difficulties, and staff turnover is high with one in three leaving their first job within two years (Routly et al. 2002). The first year after graduation is regarded by many graduates and employers as a ‘make or break’ period (Gilling and Parkinson 2009). A link between distress or dissatisfaction in the veterinary workplace and inadequate development of graduate attributes has been suggested (Heath 1998). Non-technical attributes such as communication skills, problem solving and decision making skills, recognition of one’s own limitations and the ability to cope with pressure are considered by recent graduates to be important to ease the transition to working in the veterinary profession (Rhind et al. 2011). Many work with little Chapter 2: Structured review of possible influences on increased risk 35

supervision, do not always have access to assistance from colleagues and make professional mistakes which can have a considerable emotional impact and may be significant in the development of suicidal thoughts (Mellanby and Herrtage 2004). Reactivation of latent negative cognitive schemas (unfavourable core beliefs about oneself that are derived from prior experience and can be activated by specific circumstances), established during adverse experiences at veterinary school or in the early stages of an individual’s career, may play a causative role in later depressive episodes.

Veterinary work is perceived as stressful by over 80 percent of UK veterinary surgeons (Robinson and Hooker 2006). Using a short validated stress evaluation tool to measure and compare a number of work-related stressors and stress outcomes across 26 different occupations in the UK, Johnson et al. (2005) have shown that veterinary surgeons reported lower psychological well-being than workers in most other occupations. A large cross-sectional survey of mental health and well-being in the UK veterinary profession (Bartram et al. 2009b, c) showed that veterinary surgeons have less favourable psychosocial work characteristics (greater risk of work-related stress) in relation to demands and managerial support than the general working population. Number of hours worked, making professional mistakes, and the possibility of client complaints or litigation were the main reported contributors to stress (Bartram et al. 2009c).

A large cross-sectional survey of work-related stress in the veterinary profession in New Zealand showed that those working in small animal practice, female and younger veterinarians reported the highest levels of stress, primarily associated with long working hours, client expectations and unexpected clinical outcomes. Respondents were also stressed by the need to keep up their knowledge and technical skills, and by personal relationships, finances and their expectations of themselves. At some stage during their lifetime, 2 percent of all respondents had made a suicide attempt and 16 percent of respondents had seriously thought about suicide (Gardner and Hini 2006).

A cross-sectional study of veterinary surgeons in Belgium reported only a moderate level of job strain, no higher than reports using the same method of measurement for other professional groups, but higher negative work-home interaction, especially for respondents in bovine or mixed practice (Hansez et al. 2008). Veterinarians were

36 Chapter 2: Structured review of possible influences on increased risk

highly engaged in their work which may have buffered against job strain. However, the results must be interpreted with great care owing to the very low response rate.

Financial debt is a risk factor for mental disorder and suicidal behaviour in the general population (for example, Hatcher 1994, Hintikka et al. 1998, Yip et al. 2007, Jenkins et al. 2008). Twenty-nine percent of working doctors who died by suicide in England and Wales had significant financial problems in the year before death (Hawton et al. 2004a). There may be an association between rising student debts and suicide among veterinary surgeons (Williams 2006). The length of the veterinary course and the requirement for extramural studies means that debt is a particular issue for veterinary students. In 2008, the average debt faced by a final year first- degree veterinary student in the UK was over £20,000 and 35 percent of final year students reported their financial problems as either difficult or severe (British Veterinary Association [BVA] and Association of Veterinary Students [AVS] 2008).

Veterinary practice occupies a difficult and complex moral position because it serves both animal and human interests which may conflict (Tannenbaum 1993, de Graaf 2005). Ethical challenges are commonplace as veterinary surgeons seek to balance obligations to ensure the welfare of their patients while accommodating the owners’ expectations or demands, often within strict economic constraints, the views of professional peers, the wider interests of society as a whole, and the commercial interests of private practice (Rollin 2006). A recent UK survey showed that veterinarian surgeons consider these frequent ethical dilemmas to be stressful (Batchelor and McKeegan 2011). Examples of conflict abound: the sick animal whose monetary value is less than the cost of treatment; the owner who presses for continuing and potentially painful intervention for a dying pet; the owner who demands convenience euthanasia for a healthy animal; the owner who cannot afford the cost of the most appropriate treatment. The ‘moral stress’ of the caring-killing paradox experienced by animal-shelter workers involved with the euthanasia of unwanted animals is associated with increased levels of work-related stress, work- to-family conflict, poor physical health, substance misuse and lower levels of job satisfaction (Rollin 1987, Reeve et al. 2005, Rogelberg et al. 2007). Rohlf and Bennett (2005) demonstrated that different levels of euthanasia-related stress symptoms are not associated with the occupational context in which euthanasia occurs; stress symptoms can be reported regardless of the reasons for euthanasia. Interviews with US veterinary graduates revealed that the most common upsetting experiences involved procedures the students felt were unnecessary, such as when Chapter 2: Structured review of possible influences on increased risk 37

an animal’s suffering was prolonged because its owner did not accept that its disease was incurable and that death was inevitable (Herzog et al. 1989). Procedures causing pain to animals, such as castration of calves without anaesthesia, were also considered stressful to the students and 25 percent found euthanasia of animals personally distressing.

A survey of veterinary surgeons in Finland reported high levels of work-related stress, particularly among those working in urban areas and academia (Reijula et al. 2003).

In a cross-sectional study of the occupational health of veterinary surgeons in Australia, the proportion of respondents with high levels of psychological distress was double that of the general population: significant predictors of high levels of distress included being less than 35 years old and having taken non-prescription drugs in the past 12 months. Working long hours and being on-call after hours were major contributory factors for stress. Veterinary surgeons working in general practice were significantly more stressed than those in other roles (Fairnie 2005). A survey of veterinarians in small animal practice in Australia concluded that this subgroup of the profession was particularly prone to work-related stress (Meehan and Bradley 2007). A survey of veterinarians who graduated from Australian Veterinary Schools between 1960 and 2000 reported one third as having poor psychological health and job-related anxiety. No relationship was found between type of practice and mental health (Fritschi et al. 2009). Hatch et al. (2011) demonstrated that Australian veterinarians report higher levels of psychological distress, including higher levels of depression, anxiety, stress and burnout, than the general working population. The main determinants were female gender, urban location, small animal practice and the first five years after graduation.

In a survey of UK veterinary surgeons, animal deaths caused by illness and euthanasia were reported to provoke significant short- and long-term emotional reactions in a substantial proportion of veterinarians, which may be relevant to the genesis of depression (Fogle and Abrahamson 1990). A study of US veterinarians demonstrated that communicating bad news to clients is stressful and, for some individuals, the feelings of stress are prolonged (Ptacek et al. 2004).

A large cross-sectional study of veterinary surgeons in Germany also recorded high levels of work-related stress, mainly attributable to workload and after-hours on-call duties (Harling et al. 2009). Small surveys of veterinary surgeons in Ireland have

38 Chapter 2: Structured review of possible influences on increased risk

recorded similar outcomes (Connolly 2003, Kinsella 2006). Trimpop et al. (2000a, b) demonstrated that stress associated with long working hours was a predictor of traffic accident rates in German veterinary surgeons: those who worked over 48 hours per week reported significantly higher levels of work stress and driving accidents.

Suicide risk is raised in occupations in which those employed are directly dependent upon clients for their income (Stack 2001). An exploratory study of 36 occupations found that the suicide rate was over 1.5 times higher for persons in client-dependent occupations (such as physicians, dentists and retail proprietors) in comparison with those in non-client-dependent occupations and it was assumed that this increased risk is mediated by client-dependency being a major source of stress (Labovitz and Hagedorn 1971). Veterinary surgeons in private practice work in a client-dependent environment.

Robinson and Hooker (2006) reported that 53 percent of UK veterinary surgeons would still opt for the veterinary profession if they could start their career again, 20 percent would not, and the remaining 27 percent are unsure. Similar figures are reported for veterinary surgeons in Australia (52 percent, 32 percent and 17 percent respectively) (Heath 2007). The apparent disenchantment of a substantial proportion of veterinary surgeons with their chosen career may have a negative impact on their mental health. It has been reported that unrealistic academic or career expectations can foster an unbalanced lifestyle that may lead to emotional exhaustion, depression and addiction (Wolf 1994). However, a recent study of UK veterinary undergraduates suggested that the majority had a realistic understanding of the basic working conditions of a veterinarian in general practice (Tomlin et al. 2010b).

The concept of ‘compassion fatigue’, a form of secondary traumatic stress due to the emotional demands experienced by workers in caring professions, has received some attention in the veterinary literature (Mitchener and Ogilvie 2002, Cohen 2007) but there are no published reports of any measurements of the concept within the profession.

Major incidents concerning animal health, such as the 2001 foot-and-mouth disease outbreak in the UK, can elevate levels of psychological morbidity in affected communities (Peck 2005), potentially including veterinary surgeons through their involvement in large-scale slaughter and disposal, provision of emotional support to farmers, and the economic effects on private practices. Veterinary surgeons were Chapter 2: Structured review of possible influences on increased risk 39

turned to for emotional support during this outbreak by 40 percent of farmers surveyed and were the second most-cited source of support after family, friends and other farmers (Peck et al. 2002, Peck 2005). Calls from the veterinary profession to Vet Helpline2 increased during the outbreak and decreased to pre-outbreak levels by mid-2002 (Peck 2005). Nusbaum et al. (2007) reported a number of negative psychological reactions in veterinarians involved with the outbreak and advocated training in ‘psychologic first aid’ for public health professionals involved in such incidents to limit emotional distress in the rural community and themselves.

Seventy-one percent of working doctors who died by suicide in England and Wales between January 1991 and December 1993 had significant problems at work in the year before death: over one-third of those were facing complaints, which in most cases appeared to have been a key factor leading to suicide, although most were also facing other problems at work and at home. Other common occupational problems included feeling overloaded by the volume of work, working long hours and feeling unable to cope with the responsibility of the job (Hawton et al. 2004a).

The number of complaints received by the Royal College of Veterinary Surgeons from the general public increases steadily each year; the main category being alleged inadequate care (RCVS 2007).

2.6.6 Effects of gender

The PMR for suicide among veterinary surgeons is greater in women than in men (Mellanby 2005), but this may simply reflect the small absolute number of women concerned. This parallels the situation in the medical profession in which female doctors are consistently reported to have higher suicide risk compared to the general population than their male colleagues (Baldwin and Rudge 1995, Hawton et al. 2001, Schernhammer and Colditz 2004, Petersen and Burnett 2008). In marked contrast to the general population in which the suicide rate for men is around three times the suicide rate for women (Brock et al. 2006), the absolute suicide rate among doctors is similar for both genders (Lindeman et al. 1996, Hawton et al. 2001). This differential increase in risk between the sexes requires particular monitoring in view of the large increase in the number of women entering the veterinary profession. There are various theories to explain why women in some professions have an increased risk of suicide including: participation in such

2 Vet Helpline is a peer-support telephone helpline providing emotional support and signposting to the veterinary profession.

40 Chapter 2: Structured review of possible influences on increased risk

disciplines may compete with some of the personality characteristics that are proposed as protective for women, or a relative lack of such characteristics might predispose women to choose careers in these fields; the professions concerned are action-oriented and recruit similarly oriented women; and women may be less able to cope with the social isolation inherent in these professions (Murphy 1998, Bhugra 2006). The theory of status integration suggests that persons in statistically infrequent occupation-based role sets have higher suicide rates than their counterparts (Gibbs 2000) and is supported by studies (such as Evans and Steptoe 2002) which indicate that when men and women occupy jobs in which they are in the cultural and numerical minority, there may be adverse psychological and other health effects that are gender-specific. Accordingly, Hawton et al. (2001) suggested that that the risk among female doctors might be expected to decline as they progressively become the majority in a traditionally male-dominated occupation. This may also apply within the veterinary profession.

Women have a greater preference for self-poisoning as a method of suicide than men (Callanan and Davis 2011), a pattern that is more pronounced in female doctors and veterinarians than in the general population (Kelly and Bunting 1998), so it is possible that in an occupational context of ready access and knowledge, as in the medical and veterinary professions, the risk of suicide may increase more for women than for men. However, male doctors and veterinarians also use self- poisoning for suicide more often than men in general (Kelly and Bunting 1998) so this is unlikely to be the sole explanation for the gender difference. The particular stresses affecting women in these professions are likely to contribute to the difference (Gross 1997, Hawton et al. 2001). By far the largest and most frequent gender-related source of stress for female junior doctors is the conflict they feel between their career and their marital and family life and this is strongly linked to depression (Firth-Cozens 1990). Other sources identified in the survey included a lack of senior female role models and prejudice from patients.

Female veterinary students report higher levels of emotional empathy with animals (Paul and Podberscek 2000), greater concerns for the welfare or rights of animals (Serpell 2005), and attach greater importance to the human-animal bond (Martin et al. 2003, Martin and Taunton 2005), than their male counterparts, and this is reflected in gender differences in veterinary surgeons’ reported emotional responses to treatment failure and carrying out euthanasia (Fogle and Abrahamson 1990), and in attitudes to pain control in animals (Capner et al. 1999, Raekallio et al. 2003, Chapter 2: Structured review of possible influences on increased risk 41

Hugonnard et al. 2004). It may affect the ability of women to cope with the emotional stresses and moral conflicts inherent in veterinary practice (Paul and Podberscek 2000).

Phillips-Miller et al. (2000, 2001) surveyed married or partnered US veterinarians about work satisfaction, work-related stress, marital and family stress, and perceived level of spousal support for their career. Men and women reported similar levels of work satisfaction and work-related stress but women reported significantly greater levels of effect of marital and family stress on career and lower levels of spousal support than their male counterparts. Areas of greatest work dissatisfaction for both genders were income and long working hours. Emergency calls outside office hours and unwillingness of clients to agree to diagnostic or treatment procedures were also cited as being particularly stressful.

2.6.7 Perceived stigma

The stigma associated with mental illness is increasingly recognised as an important factor influencing the accessing of mental healthcare by the general population (Sartorius 2007). Mental illness may be particularly stigmatising for those working in professions where vulnerabilities are not readily tolerated and viewed as a form of weakness with negative career implications. Stigma is recognised as an important factor influencing the seeking of appropriate mental healthcare by medical students (Schwenk et al. 2010) and doctors (White et al. 2006, Worley 2008). Also, suicide risk may be greater in higher income earners who develop mental illness as they may feel more stigmatised than others with lower income (Agerbo et al. 2001). Similar stigma may apply within the veterinary profession with a consequent reduction in help-seeking behaviour by those with concern about their mental health.

2.6.8 Psychiatric illness

The importance of psychiatric disorders as a risk factor for suicide in the general population is described in Section 2.2, Table 2.8.

A large, case-controlled study showed that the risk of psychiatric diagnosis with affective and stress-related disorders is elevated for human service professions relative to all other occupations, after adjustment for an extensive range of sociodemographic variables (Wieclaw et al. 2005, 2006). Specific professions contribute differentially to the magnitude of this increased risk: for example, teaching and social services display the highest risk of stress-related disorders and health professionals

42 Chapter 2: Structured review of possible influences on increased risk

have an elevated risk of depression. The authors suggested that the elevated risk in human service professions may be associated with factors including: job characteristics such as irregular working hours and exposure to distressing events; high and conflicting job demands; self-selection of individuals into occupations who are inclined to display a high degree of commitment; and men and women occupying occupations in which the opposite gender predominates.

Agerbo et al. (2007) explored the impact of psychiatric factors on the relative risk of suicide across 55 occupations. Only modest associations between suicide and occupation were observed among individuals who had been previously hospitalised for psychiatric illness. This implies that mental disorder lies on the causal pathway between occupation and suicide or that occupational differences are less important once a person suffers from a psychiatric illness. However, doctors were a notable exception, at almost four-fold greater risk than other occupations of dying by suicide if they had been previously hospitalised for a psychiatric disorder. Thus the increased risk of suicide among doctors may be related to both an increased risk of psychiatric disorder and the increased risk of those with a psychiatric disorder to die by suicide. Although no cases of suicide among veterinary surgeons were identified in the cohort of suicides examined by Agerbo et al. (2007) it is possible that this dual effect seen in doctors may apply similarly in the veterinary profession.

Observations by Hawton et al. (2011) suggested that the elevated suicide risk in nurses, physicians, dentists and pharmacists was not explained by an excess of persons with psychiatric disorders; indeed, risk relative to the comparison groups was somewhat greater in those without contact with psychiatric services although the relative risks were only statistically different for veterinary surgeons. It is possible, however, that medically-related occupational groups may be reluctant to seek help for mental health problems because of perceived stigma and other factors (White et al. 2006, Worley 2008).

The principal psychological problems experienced by doctors are depression and alcoholism, and these are more frequent than in the general population (Firth- Cozens 2001). Pre-existing psychiatric disorders were present in over 86 percent of doctors completing suicide, mainly depressive illness and alcohol or drug dependence (Hawton et al. 2004a), and over 90 percent of nurses (Hawton et al. 2002). The findings of a recent study of psychopathology in Canadian doctors who killed themselves are consistent with those of Hawton et al. (2004): pre-existing Chapter 2: Structured review of possible influences on increased risk 43

psychiatric disorders were present in 83 percent of doctors completing suicide, mainly depressive illness (Gagné et al. 2011). In a study of UK healthcare professionals referred to a specialist drug and alcohol treatment service, around one-third had previously received treatment for depression and about one-fifth had previously taken an overdose (Gossop et al. 2001). Brooks et al. (2011) reported the diagnoses of the first two hundred healthcare professionals treated by the recently- established, Practitioner Health Programme, set up specifically to provide treatment and support to doctors and dentists with mental health, addiction and/or physical health problems in the London area of the UK – the principal diagnoses were depressive illness (48 percent) and addiction (39 percent), the latter mainly to alcohol. No such data are available for veterinary surgeons but it is reasonable to speculate that psychiatric disorders may be a similar factor in suicides by veterinary surgeons. A ready opportunity exists in both professions for misuse of prescription medications. Among UK veterinary surgeons referred to a health support programme, the order of substance misuse preference is alcohol, ketamine, benzodiazepines, opiates, street drugs (cannabis, heroin, cocaine and ecstasy) and nitrous oxide, and about half of those treated admit to having had suicidal thoughts (Veterinary Benevolent Fund [VBF] 2007). Substance misuse is also reported among Australian veterinarians (Jeyaretnam et al. 2000).

Bartram et al. (2009a, b, c) estimated that compared to the general population, UK veterinary surgeons report higher levels of anxiety and depressive symptoms, higher 12-month prevalence of suicidal thoughts, less favourable psychosocial work characteristics in relation to demands and managerial support, lower levels of positive mental well-being, and higher levels of negative work-home interaction. The level of alcohol consumption did not appear to be a negative influence on mental health within the profession as a whole. The latter is consistent with the findings of Mellanby et al. (2009) that the PMRs for alcohol-related deaths among male and female veterinary surgeons aged 20-64 years in England and Wales 1993-2005 were lower than for the general population, albeit the differences were not statistically significant.

Harling et al. (2009) explored the use of psychotropic substances among German veterinary surgeons. The prevalence of smoking and drug use among veterinarians was respectively lower than and similar to the general population. However, alcohol consumption was elevated, and the prevalence of dangerous levels of alcohol consumption was markedly elevated among female veterinarians.

44 Chapter 2: Structured review of possible influences on increased risk

Hafen et al. (2006, 2008) reported that one-third of first-year students at a US veterinary school had symptoms of depression associated with both academic and non-academic stressors, but it is not clear to what extent these findings can be generalised to veterinary undergraduates elsewhere.

Fishbain (1986) compared the characteristics of four US veterinary surgeons with psychiatric illness with those of published descriptions of psychiatrically impaired doctors. Similarities included the type of drug selected for misuse (mainly opiates); the number of drugs used by an individual (multiple); and source of supply (self- prescribed rather than street-sourced). The veterinary surgeons misusing drugs considered that overwork, fatigue, situational stress, and problems with significant others were precipitating factors. The outcomes must be interpreted very cautiously, however, due to the small number of veterinarians concerned.

Studies with findings relevant to mental health in veterinary surgeons are summarised in Table 2.13

Table 2.13: Summary of studies included in this review with findings relevant to mental health in veterinarians Study and year Design Sample Location n Response Key findings rate (%) Anon 2002 Cross-sectional Veterinary surgeons Australia 313 34 Euthanasia, specific client behaviours or characteristics, Support staff 183 20 and long working hours were important stressors Connolly 2003 Cross-sectional Veterinary surgeons Ireland 46 26 Workload reported as main source of stress Bartram et al. 2009a, Cross-sectional Veterinary surgeons UK 1796 56 Symptoms of mental ill-health elevated relative to b, c general population. High psychological demands and low levels of managerial support Brown 1994 Cross-sectional Veterinary students USA 207 96 Students deploy a range of both problem- and emotion- focused coping styles with relatively equal frequency Elkins and Kearney Cross-sectional Veterinary surgeons USA 572 57 Stress and burnout reported 1992 Fairnie 2005 Cross-sectional Veterinary surgeons Australia 419 43 Proportion of highly distressed respondents was double that of the general population. Distress highest for those <35 years old Fishbain 1986 Case-series Veterinary surgeons USA 4 - Pethidine by intramuscular injection was principal drug abused; history of drug abuse in veterinary school; veterinary-sourced Fogle and Cross-sectional Veterinary surgeons UK 167 56 Animal deaths provoke severe short- and long-term Abrahamson 1990 emotional reactions in veterinarians (76 percent and 20 percent) Fritschi et al. 2009 Cross-sectional Veterinary surgeons Australia 2125 37 One third reported poor psychological health but levels of distress, anxiety and depression similar to other professions Gardner and Hini Cross-sectional Veterinary surgeons NZ 927 49 Those working in small animal practice, women and 2006 younger veterinarians reported the highest levels of stress Hafen et al. 2006, Cohort Veterinary students USA 93 T1: 90 One third of first year students report clinical levels of 2008 78 T2: 84 depressive symptoms. Predictors include academic concerns Hansez et al. 2008 Cross-sectional Veterinary surgeons Belgium 216 9 Mean job strain and job engagement levels not higher than other professions, but greater negative work-home interaction Continued overleaf 45

Study and year Design Sample Location n Response Key findings rate (%) Harling et al. 2009 Cross-sectional Veterinary surgeons Germany 1136 53 Work-related stress mainly attributable to workload, after-hours on-call duties and difficult clients. High alcohol consumption. Hatch et al. 2011 Cross-sectional Veterinary surgeons Australia 1947 28 Veterinarians report higher levels of psychological distress than the general population Heath 2007 Cross-sectional Veterinary surgeons Australia 134 98 Only 52 percent would opt for the veterinary profession if they could start their career again Herzog et al. 1989 Cross-sectional Graduating veterinary US 24 - Distress was associated with prolongation of animal † students suffering, procedures causing pain to animals, and euthanasia Hesketh and Cross-sectional Veterinary surgeons NZ 411 59 Decision latitude protective of mental health; lack of Shouksmith 1986 control over speed of work and discretion are associated with anxiety Johnson et al. 2005 Cross-sectional Veterinary surgeons UK 262 363 Low psychological well-being, intermediate job Ireland satisfaction and good physical health compared to 25 other occupations Kinsella 2006 Cross-sectional Veterinary surgeons Ireland 14 44 Unsociable hours and workload ranked highest as stressors Kirwan 2005 Cross-sectional Veterinary surgeons, UK 15 - 93 percent have favourable inclination towards † nurses and euthanasia of humans receptionists Kogan et al. 2005 Cross-sectional Veterinary students USA 233 44 Non-academic stressors include additional employment, relationship concerns and poor self-care habits Martin et al. 2003 Cross-sectional Veterinary students USA 146 NR Human-animal bond influences decisions to become veterinarians, especially for women Martin and Taunton Cross-sectional Veterinary surgeons USA 415 26 81 percent indicated that the human-animal bond was 2005 important to their decisions to become veterinarians, especially for women Mellanby and Cross-sectional Recent graduates UK 108 27 Poor supervision, minimal support, many mistakes result Herrtage 2004 in distress to veterinarians Continued overleaf

3 The response rate for veterinarians is not reported in Johnson et al. (2005). Percentage obtained from the original source: Connolly (2002).

Study and year Design Sample Location n Response Key findings rate (%) Paul and Cross-sectional Veterinary students UK 319 NR Females have higher emotional empathy with animals Podberscek 2000 which is sustained throughout course for females but declines for males Phillips-Miller et al. Cross-sectional Veterinary surgeons USA 305 54 Women report greater effects of marital and family 2000, 2001 stress and lower spousal support Ptacek et al. 2004 Cross-sectional Veterinary surgeons USA 62 - Communicating bad news is stressful and, for some † individuals, the feelings of stress are prolonged Reijula et al. 2003 Cross-sectional Veterinary surgeons Finland 785 67 Those working in towns or involved in education and research reported the most stress Robinson and Cross-sectional Veterinary surgeons UK 9671 47 Work is perceived as stressful by 80 percent of Hooker 2006 profession Only 53 percent would choose veterinary career again Routly et al. 2002 Cross-sectional Recent graduates UK 58 54 Main difficulties for new graduates are being on call, Senior partners 34 59 financial aspects of the role, and lack of surgical competence Serpell 2005 Cross-sectional Veterinary students USA 302 92 Females have greater concern for animal welfare/rights Interactions with animals influences attitudes and career choice Smith et al. 2009 Cross-sectional Veterinary surgeons Australia 664 64 Stress associated with long working hours, insufficient rest breaks and holidays, client attitudes, lack of recognition and not having enough time per patient Tomlin et al. 2010b Cross-sectional Veterinary students UK 289 68 Veterinary students’ understanding of a career in practice is generally realistic. Limited knowledge of the range of potential job opportunities outside practice Trimpop et al. Cross-sectional Veterinary surgeons Germany 494 NR Stress associated with long working hours was a 2000a, b Support staff 284 predictor of traffic accident rates Williams et al. 2005 Cross-sectional Veterinary students USA 57 41 Narrow range of effective coping strategies Zenner et al. 2005 Cross-sectional Veterinary students USA 61 80 Psychological characteristics similar to other high- achieving populations

NR: Not reported † Interview-based; all other included studies are questionnaire-based 47

48 Chapter 2: Structured review of possible influences on increased risk

2.7 Towards a hypothetical model to explain suicide risk in veterinary surgeons: a schematic representation

A review of current knowledge about possible influences on the suicide risk among veterinarians and factors elevating the risk in other occupations and in the general population has been used to develop a comprehensive hypothetical model to explain suicide risk in veterinary surgeons (Figure 2.1). The model is congruent with the stress-diathesis (Mann et al. 1999), cry of pain (Williams and Pollock 2000, Williams 2001), interpersonal-psychological (Riberio and Joiner 2009) and differential occupational risk (Stack 2001) models of suicide described above.

Based on specific testable constructs, the model attempts to clarify a complex interaction of possible mechanisms across the career life course and may serve as a useful heuristic to facilitate a more focused and systematic approach to research. Research is required to validate or disprove the component hypotheses.

The principal component hypotheses within this model are that the following factors elevate suicide risk in veterinary surgeons:

• Personality characteristics of individuals entering the profession confer a predisposition or vulnerability

• Psychological morbidity attributable to psychosocial factors during undergraduate training and in the workplace

• Familiarity with animal euthanasia leads to more permissive attitudes towards suicide

• Access to and knowledge of lethal means

The authors of a recent systematic review of suicidal behaviour and psychosocial problems in veterinary surgeons (Platt et al. 2010a) confirmed that their findings support this model.

Fig. 2.1: Schematic representation of a hypothetical model to explain the risk of suicide in veterinary surgeons

Source: Bartram and Baldwin (2010)

Reproduced with permission of the British Veterinary Association 49

50 Chapter 2: Structured review of possible influences on increased risk

2.8 Shortfalls in existing published work

The increased risk of suicide among veterinary surgeons is well established (see Section 2.1). However, the available literature that seeks to explain this observation has many shortfalls. These can be summarised as follows:

• The questionnaire instruments are mainly self-formulated and do not use standardised measures with established psychometric properties (such as Gardner and Hini 2006).

• The instruments with established psychometric properties that were used in reported studies have been validated on representative samples of the general population only and not within the context of the veterinary profession.

• A number of reports are based largely on conjecture (such as Halliwell and Hoskin 2005).

• The generalisability of the results is limited by small sample size and/or low response rates (such as Connolly 2003, Kirwan 2005, Kinsella 2006, Hansez et al. 2008, Fritschi et al. 2009) or the study was completed outside the UK (such as Reijula et al. 2003, Gardner and Hini 2006, Hansez et al. 2008, Fritschi et al. 2009).

• The prevalence of any psychological characteristics which might create a predisposition or vulnerability in individuals entering the profession has not been evaluated.

• There has been little attempt to evaluate the prevalence of psychiatric factors such as depression or substance misuse among practising veterinary surgeons.

• Attitudes to suicide and the potential influence of the acceptability of animal euthanasia within the profession on those attitudes have not been evaluated.

2.9 Summary

empirical research. A computerised search of published literature on the suicide risk and influences on suicide among veterinarians, with comparison to the risk and influences in other occupational groups and in the general population, was used to develop a structured review. Veterinary surgeons have a proportional mortality ratio (PMR) for suicide approximately four times that of the general population and around twice that of other healthcare professions. A complex interaction of possible mechanisms may occur across the course of a veterinary career to increase the risk of suicide. Possible factors include the characteristics of individuals entering the profession, negative effects during undergraduate training, work-related stressors, ready access to and knowledge of means, stigma associated with mental illness, professional and social isolation, and alcohol or drug misuse (mainly prescription drugs to which the profession has ready access). Contextual effects such as attitudes to death and euthanasia, formed through the profession’s routine involvement with euthanasia of companion animals and slaughter of farm animals, and suicide ‘contagion’ due to direct or indirect exposure to suicide of peers within this small profession are other possible influences.

Among the shortfalls of existing published work is the use of instruments with psychometric properties that have not been validated within the context of the veterinary profession.

Chapter 3:

Measurement properties of health rating scales

54 Chapter 3: Measurement properties of health rating scales

3.0 Introduction

The structured review presented in Chapter 2 identified that the instruments with established psychometric properties that were used in reported studies of mental health and well-being in the veterinary profession have been validated in representative samples of the general population only, and not within the context of this particular occupational group.

The cross-sectional study of mental health and well-being and their associations in the UK veterinary profession reported by Bartram et al. (2009a, b, c) employed several pre-existing instruments to assess the contribution of mental health and well-being to the elevated suicide risk in this occupational group.

One of the instruments used, the Warwick-Edinburgh Mental Well-being Scale (WEMWBS) (Tennant et al. 2007) was developed as a measure for assessing population positive mental health. Unlike other commonly-used measures which show floor or ceiling effects in general population samples, with most people scoring the optimum level, WEMWBS is better able to distinguish average from good mental health and may have potential for measuring overall changes in population mental well-being (Tennant et al. 2007). The scale offers promise as a tool for measuring mental health and well-being at a population level within the veterinary profession. The scale comprises positively worded items only. This positive orientation towards well-being contrasts with the illness-orientation of other scales which measure similar constructs, and increases its attractiveness as a scale which could be embedded into questionnaires for general rather than health-related purposes, such as the RCVS Survey of the Profession.

However, the psychometric properties of the WEMWBS have been examined in student and representative general population samples only and not within any specific occupational groups (Tennant et al. 2007, Stewart-Brown et al. 2009, Clarke et al. 2011). The demographic structure of the veterinary profession is not representative of the general population, with differences in ethnicity, employment status, educational attainment, income, and social grade (Robinson and Hooker 2006, Robertson-Smith et al. 2010). Moreover, there are also marked population- level differences in the levels of self-reported mental health and well-being (Bartram et al. 2009b). The psychometric properties of psychological instruments are not necessarily stable across different groups of people (reviewed by Vandenberg and Lance 2000). Without evidence for measurement invariance, differences between Chapter 3: Measurement properties of health rating scales 55

groups or across time in scores on a measure might reflect differences in the measurement properties of the self-report instrument used, such as a change in the meaning of the items and their relations, rather than true differences in the latent variable. Consequently, it is appropriate to re-examine the measurement properties of WEMWBS in the context of the veterinary profession, to inform judgements of the utility of the instrument in this occupational group.

To provide context for the psychometric analyses presented in subsequent chapters, Chapter 3 outlines the theory which underpins methods for evaluating the properties of psychological measurement instruments and explains the quality criteria against which instruments can be assessed.

3.1 Psychometric theory: traditional and modern

Psychometric methods are concerned with the construction and evaluation of rating scales in order to determine their success at making the variable operational and in locating people on the scale. The theories which underpin psychometric methods can be divided into two categories: traditional (classical test theory, CTT) and modern (item response theory, IRT; and Rasch measurement theory, RMT) (Hobart and Cano 2009, Andrich 2011).4 Classical test theory is currently the predominant paradigm (Cano and Hobart 2011).

3.2 Traditional psychometric methods

The central tenet of CTT on which traditional psychometric methods are based is that the score that a person obtains on a given instrument, which is symbolised by (X), can be deconstructed into the person’s true score (T) and a random error component (E) (DeVellis 2006, Hobart and Cano 2009). The CTT model can be expressed as:

Xip = Tip + Eip

Where Xip is an observed score for test i and person p, Tip is a true score of test i and person p, and Eip is an error score of test i and person p.

4 The term modern test theory is not used in reference to an absolute period in the development of test theory, but to contrast it to classical test theory. The basics of modern test theory were established by Thurstone in the 1920s (Thurstone 1928).

56 Chapter 3: Measurement properties of health rating scales

The person’s true score, T, is defined as the expected value of the observed score over an infinite number of repeat administrations with the same instrument.

Classical test theory is grounded in the definition of measurement as proposed by Stevens (1946), i.e. the assignment of numerals to objects or events according to some rule. If the assumption is made that measurement is quantification of a variable, and that variables can go from ‘less of’ to ‘more of’, then it is reasonable to consider the total score generated by summing a set of item scores is a measure of that variable provided it can be demonstrated that the items address the same underlying construct (Hobart 2003).

3.2.1 Limitations of traditional psychometric methods

The assumptions that underpin CTT are actually broadly unfounded, and, in fact, CTT is a theory that cannot be tested, verified or more importantly, falsified in any data set, because T and E cannot be determined in a way that enables evaluation of their accuracy (Hobart and Cano 2009). This has four important implications (Hobart et al. 2007). First, measurement theories that cannot be tested are, by definition, weak theories that lead to only weak inferences about the performance of a rating scale and what it measures. Second, theories that cannot be challenged are easily satisfied by datasets. Third, because the parameters can not be estimated with confidence, only the ordinal raw scores can be analysed. Finally, the equation derived from CTT for calculating the confidence intervals around the scores for individuals gives large values that indicate a lack of confidence when comparing changes and differences among individuals. This leads to several limitations of rating scales constructed and evaluated by traditional psychometric methods (DeVellis 2006, Hobart and Cano 2009, Cano and Hobart 2011). These include:

1. Multidimensionality. The rating scales have potential to measure more than one underlying construct.

2. Ordered counts are not interval measures. The unit of measurement implied by the data is not equal across the whole range of the continuum (Figure 3.3).

3. Scores are sample dependent. An item’s characteristics depend on the true level of that characteristic in the person being tested. Consequently the level of a characteristic expressed by a test item will vary substantially if the true score distributions vary between populations. Chapter 3: Measurement properties of health rating scales 57

4. Scores are item dependent. A person’s true score depends on the level of a characteristic expressed by the test item. Consequently scores may not be comparable between test items which express low levels of a characteristic and test items which express higher levels of a characteristic.

5. Comparison of test results between different populations. Without evidence of measurement invariance, differences between groups might merely reflect differences in the measurement properties of the instrument used.

6. Handling of missing data. A widely-used approach is to impute values for missing data with the person-specific mean score (the mean score of the items that have been answered by that individual), but this assumes that all the items express the same level of a characteristic.

7. Data are not suitable for the assessment of individuals. Ordinal scores are only suitable for group-level comparisons because the 95% confidence intervals around the ordinal score of an individual are wide (computed as ± 1.96 × standard error of measurement. See Section 4.4.3.2)

Consider measurement in the physical sciences (referred to as ‘fundamental measurement’), for example, a ruler for measuring height. The ruler describes only one attribute (unidimensionality), which it measures on a linear continuum (that is, the differences between the calibrations are equal). The ability of a ruler to measure height is not seriously affected by the people being measured (sample independent). It does not matter which ruler is used to measure height (scale independent). The process of measurement remains the same across different populations (invariance).

3.3 Modern psychometric methods

Modern psychometric methods represent a logical progression from traditional methods because they attempt to improve the scientific quality of the theory underpinning rating scales. Because traditional psychometric methods are based on a weak measurement theory (Hobart et al. 2007, Hobart and Cano 2009), they are limited in their ability to adequately provide data to support the use of health measures. In view of these shortcomings, endeavours were made to improve the theory and, therefore, the scientific quality of the conclusions that could be drawn. Modern psychometric methods attempt to ensure that scales have the

58 Chapter 3: Measurement properties of health rating scales

characteristics of fundamental measurement as used in the physical sciences: unidimensionality, linearity, sample independence, scale independence, and invariance (Hobart 2003).

The modern psychometric methods are called Rasch analysis and Item Response Theory (IRT). Fundamentally, these methods differ from classical test theory (CTT) because their focus is the relationship between a person’s measurement and his or her probability of responding to an item, rather than the relationship between a person’s measurement and his or her observed scale total score (Cano and Hobart 2011). They are considered to be conceptually and theoretically superior to traditional methods, especially in the context of health measurement, and are now widely used for the construction of new instruments and the refinement of existing instruments (Hobart and Cano 2009). Although there are similarities between these two item-based approaches, there are also fundamental differences. Rasch measurement is concerned with how well the data fit the model which prescribes the item responses expected for interval scale measurement. By contrast, IRT is descriptive, aiming to find the item response model that explains the data (Andrich 2004, Hobart and Cano 2009, Massof 2011, Andrich 2011). There is debate around the relative merits of each approach and the circumstances in which they should be used but, generally, Rasch models are employed to validate measurements, and the more complex IRT models can be used to explore the data further if required (Massof 2011).

3.4 Rasch analysis

There are two main components to Rasch measurement theory (Rasch 1960):

(1) Conjoint measurement. A person’s response to an item is governed by only two factors: the location of the person and the location of the item on the shared continuum of the latent variable being measured (mental well-being in the case of WEMWBS). When a set of items is used as a measurement instrument, the aim of the items is to map out a continuum on which an aspect of people (the latent variable) can be measured (located). Because the items mark out the measurement continuum, their locations define it. Thus, there is a common continuum on which the aspect of the people that is to be measured and the items for doing the measuring are both located. Consequently, when a set of items is used to measure people there is a simultaneous interaction between people and items. Chapter 3: Measurement properties of health rating scales 59

(2) Invariant comparison. The relative location of any two persons on the continuum should be independent of the items used to make that comparison, provided those items belong to the calibrated set of items which define the variable under study (Wright and Linacre 1989). The symmetrical aspect of this criterion is that the relative location of any two items on the continuum should be independent of the persons used to make that comparison. The invariance criterion means that an instrument, in principle, is required to work in the same way for all individuals. This implies invariant functioning across any partitioning of the persons, e.g. across sample groups such as age, gender or different levels of the latent variable being measured.

Rasch (1960) developed a logistic model (now known as the Rasch unidimensional measurement model) and through applications, initially in the field of educational assessment, was able to demonstrate that this approach met the stringent criteria for measurement used in the physical sciences. Application of the Rasch model to item response data is known as Rasch analysis. Conventionally, because early applications of Rasch analysis were in the field of educational assessment, the position that each item and person occupies on the common continuum of the latent variable are termed item ‘difficulty’ and person ‘ability’. In the context of the health sciences, ‘ability’ may be understood to represent the amount the person has of a given symptom, trait or feeling and ‘difficulty’ may be understood to represent the magnitude of the symptom, trait or feeling represented by the item. For example, in a scale to measure depression, an item that reflected the sentiment that life was no longer worth living would be expected to represent a high level of depressive illness when affirmed, i.e. the item has a high level of item ‘difficulty’ and is likely to be affirmed only by people with a high level of ‘ability’.

There are two key features of the relationship between a person’s ‘ability’ and that expressed by an item in a questionnaire (‘difficulty’). First, the observed response is dependent only on the difference between a person’s ‘ability’ and the ‘difficulty’ expressed by the item. Second, the model is stochastic (i.e. based on the theory of probability) because uncertainty surrounds the expected response, which is consistent with the situation in real life (Borsboom 2005, Tesio et al. 2007). The Rasch equation is explained in Figure 3.1.

60 Chapter 3: Measurement properties of health rating scales

Fig. 3.1: Rasch equation

The Rasch model prescribes that the probability that a person will affirm an item is a logistic function of the difference between the person’s ability [θ] (in terms of the level of mental health and well-being of the person) and the difficulty of the question [β] (in terms of the level of mental health and well-being represented by the item), and only a function of that difference.

In the Rasch model, item and person estimates are measured in ‘logits’. A logit is the log odds of a person of a given ability (level of health), as assessed by their response to all the items combined, having a 50% chance of responding positively to that item (Pallant and Tennant 2007) (Figure 3.2). Chapter 3: Measurement properties of health rating scales 61

Fig. 3.2: Probability of a person affirming an item

The bottom of the scale displays the difference in location (in logits) between a person and an item. The top of the scale displays the corresponding probability of the person affirming the item. The probability of a person with ability of 0 logits affirming an item with a difficulty of 0 logits is 50%. The probability of a person affirming an item with a difficulty that is 2 logits higher than their ability is 12%.

The purpose of Rasch analysis is to maximise the homogeneity of the latent variable being measured (unidimensionality) and to reduce item redundancy without sacrificing measurement information, by removing items and/or collapsing scoring levels to yield a more valid and simple measure (Granger 2008).

In the application of Rasch analysis, response data for items intended to be summed into an overall ordinal score for a specific scale are tested against the expectations of the Rasch model (Rasch 1960). These expectations are a probabilistic form of Guttman scaling, which means that for the same person ability, the probability to endorse an easy item has to be higher than the probability to endorse a more difficult item and vice versa (Karabatsos 2001). Comparing persons with different abilities, a person with a higher ability is expected to affirm all items endorsed by a person with lower ability and additionally one or more difficult items. That is, responses are contingent on the level of the underlying construct (Hagquist et al. 2009). Hence, only certain response patterns are in accordance with the Guttman structure. Since the responses are not required to be deterministic but probabilistic, there is room for random variation implying that the response patterns for two individuals with the same total score do not have to be exactly the same (Hagquist et al. 2009).

The expectations of the model must be met for rating scale data to generate internally valid, interval-level measurement estimates that are stable (invariant) across items and people (Wright and Linacre 1989). Thus, the difference between

62 Chapter 3: Measurement properties of health rating scales

what should happen (expected) and what does happen (observed) indicates the extent to which rigorous measurement is achieved. If the expectations of the model are not met, this finding prompts an examination of the data to determine why a set of items hypothesised to be a measurement instrument are not performing as such. Statistical and graphical tests are used to evaluate the extent to which observed data fit predictions of those responses from the Rasch model (Smith 2000).

When the data adequately fit Rasch model expectations, the raw score from summed item responses can be transformed into interval-level measurements (Tennant and Conaghan 2007). The difference between an ordinal score (data do not fit Rasch model) and interval-level measurement is illustrated in Figure 3.3.

Fig. 3.3: Interval between adjacent scores on an ordinal scale is not equal

Ordinal scale +1

Linear scale +2.7

Ordinal scale +3

Linear scale +1.3

The interval between the raw scores of 3 and 4 on the ordinal scale corresponds to 2.7 units of measurement on the interval scale. The interval between the raw scores of 4 and 7 on the same ordinal scale corresponds to 1.3 units on the interval scale.

The interval scaling properties of these Rasch-transformed measurements enables the legitimate application of parametric statistical analyses, given appropriate distributions. Interval scaling is also a requirement for meaningful interpretation of changes in scores over time, which may occur, for example, following remedial interventions (Norquist et al. 2007). An interval scale is able to measure how much more or less of the latent variable of interest is present with much greater precision than a raw ordinal scale. In the context of health rating scales, McHorney et al. Chapter 3: Measurement properties of health rating scales 63

(1997) defined ‘precision’ in terms of its attributes: usefulness in making comparisons across groups and between test administrations.

3.4.1 Limitations of the Rasch model

Limitations of the Rasch model include it being overly restrictive because it does not permit each item to have a different discrimination, and there being no provision in the model for other parameters such as guessing. Some suggest that the model is also limited by its need for unidimensional data and claim that it is too simple to match the complexity of human behaviour. Further, Rasch analysis is complex, and traditional psychometric methods are easier to compute (Cano and Hobart 2011).

3.5 Quality criteria for the measurement properties of health rating scales

Psychometric methods are employed to determine the extent to which a quantitative conceptualisation of a latent variable has been operationalised successfully. Typically, traditional psychometric methods consider the evaluation of rating scales in terms of three main properties: reliability, validity and responsiveness. There are guidelines as to how psychometric properties should be evaluated and reported (Lohr 2002, Terwee et al. 2007, Mokkink et al. 2010a, b, c; Cano and Hobart 2011) and recent published literature indicates that various different approaches are adopted by researchers (Hobart and Cano 2009). Several authors have recommended that a comprehensive scale evaluation should involve the examination of six psychometric properties: data quality, scaling assumptions, targeting, reliability, validity and responsiveness (McHorney et al. 1994, Ware and Gandek 1998, Hobart et al. 2001a, Lohr 2002, Food and Drug Administration 2009, Hobart and Cano 2009, Cano and Hobart 2011). Several psychometric analyses provide evidence for more than one property, and there is overlap between properties. Nevertheless, this classification provides a useful framework to describe the methods and results of rating scale evaluations (Hobart and Cano 2009). The approach taken to evaluate measurement properties in the present study aims to be consistent with these contemporary recommendations.

This section describes the key measurement properties of instruments using the framework and terminology advocated by Hobart and Cano (2009) and Mokkink et al. (2010a, b, c).

64 Chapter 3: Measurement properties of health rating scales

3.5.1 Six psychometric properties: data quality, scaling assumptions, targeting, reliability, validity and responsiveness

3.5.1.1 Data quality

Indicators of data quality reflect respondents’ understanding and acceptance of a rating scale and help to identify items that may be irrelevant, confusing, or upsetting to the target population (McHorney et al. 1994). A scale score cannot be confidently estimated if item-level missing data are high. Indicators of data quality are the percentage of missing item responses and the percentage of the sample for whom total scores can be calculated.

3.5.1.2 Scaling assumptions

It is generally accepted that a series of criteria should be satisfied for a set of items to be summed legitimately to form a single overall score for a scale (Hobart and Cano 2009):

Items should be roughly parallel, i.e. measure at the same point on the scale and have similar variance, otherwise they do not contribute equally to the variance of the total score and should be standardised before combination. A set of items is considered parallel when their item response option frequency distributions and their item mean scores and standard deviations are roughly similar. When items have similar variances it has been argued that these do not need to be standardised before items are combined (Hobart and Cano 2009).

Items should measure the same underlying construct (latent variable), otherwise it is not appropriate to combine them to generate a total score. That is, the items of a scale should be internally consistent. A set of items is considered to be measuring the same construct when each item’s corrected item-total correlation, which is the correlation between each item and the total score computed from the remaining items in that scale, exceeds a recommended criterion (Hobart and Cano 2009).

Items in the scale should contain a similar proportion of information concerning the construct being measured. Then there is no need to weight the items for differences in factor content before they are summed. This Chapter 3: Measurement properties of health rating scales 65

criterion is considered satisfied when the corrected item-total correlations are roughly similar (Hobart and Cano 2009).

For rating scales which contain sub-scales, such as the Hospital Anxiety and Depression Scale (Zigmond and Snaith 1983) which has separate sub- scales for anxiety and depressive symptoms, items should be correctly grouped into each sub-scale. That is, items should correlate higher with the total score of their own sub-scale (item own-scale correlation) than with the total score of the other sub-scales (item other-scale correlations) (Cano et al. 2008).

3.5.1.3 Targeting

Targeting is the extent to which the spectrum of health measured by a health rating scale matches the distribution of health in the study sample (McHorney and Tarlov 1995). Targeting can be studied at both the item and the scale level and is evaluated by examining score distribution, score means, standard deviation (SD), skewness statistics, and floor and ceiling effects (Hobart and Cano 2009).

3.5.1.4 Reliability

The reliability domain contains three measurement properties: internal consistency, reliability and measurement error. The reliability domain refers to the degree to which the measurement instrument is free from measurement error (Mokkink et al. 2010a, b, c) and is an estimate of the extent to which scores for people whose condition of interest has not changed are the same for repeated measurements:

1. using different sets of items from the same instrument (internal consistency);

2. over time (test-retest);

3. by different persons (i.e. raters) on the same occasion (inter-rater);

4. by the same person (i.e. raters or responders) on different occasions (intrarater).

The term reliability is derived from CCT. As detailed in Section 3.2, according to CTT, any ‘observed’ score on a measurement instrument consists of two components: the ‘true’ score plus the ‘error’ (Hobart and Cano 2009). The ‘error’ is thought to occur during each measurement. There are different types of

66 Chapter 3: Measurement properties of health rating scales

measurement error which relate to the measurement instrument itself, the measurement situation, the person administering the test, or the person being tested. Reliability is the proportion of variability in observed measurements that is due to the true variability between subjects, rather than some kind of error. This is calculated statistically and formally defined as:

Reliability = true subject variability / (true subject variability + measurement error).

The general principle is that the lower the ‘measurement error’, the higher the reliability and thus the quality of the measurement instrument. Reliability is a relative parameter, and will always have a value between 0 and 1. CCT does not specify how high reliability is supposed to be, however, a measure that has no error (i.e. all variation is true score variation) has a reliability of 1 and is perfectly reliable; a measure that has no true score (i.e. all variation is error) has zero reliability. It has been proposed that a reliability coefficient of at least 0.70 indicates appropriate reliability (Streiner and Norman 2008). Internal consistency, reliability and measurement error estimate reliability in different ways.

Internal consistency

Internal consistency evaluation differs from other reliability estimations, because administration of the measurement instrument is not repeated. It is measured by estimating the degree of inter-relatedness among individual items (Mokkink et al. 2010a, b, c). This means that if items in a scale are summarised into a total score, it should be known that the items are sufficiently correlated. Internal consistency establishes this correlation between a respondent's item-responses and suggests whether or not these items seem to measure the same construct. This method assumes that all items in a scale are part of one underlying construct (i.e. ‘unidimensional’). To check this assumption, the unidimensionality of the instrument must be assessed by applying a technique such as factor analysis before establishing the internal consistency. Internal consistency can be estimated by calculating the Cronbach's alpha (Cronbach 1951). Values of > 0.70 are required to demonstrate that the items are sufficiently correlated (Reeve et al. 2007). However, values > 0.95 can indicate that the instrument contains too many items that are assessing the same underlying construct (i.e. there is a high level of item redundancy because the same item is rephrased in several different ways).

Chapter 3: Measurement properties of health rating scales 67

Reliability

Test-retest reliability refers to the consistency of respondents’ scores over time. It is estimated by administering the instrument in a longitudinal study design on two or more different occasions to the same group of people (Rousson et al. 2002). This type of reliability depends on the stability of the outcome to fluctuations. It is based on the assumption that no real change occurs in what is measured between the time points so the timing of the consecutive measurements is important. Ideally, the time interval between administering the instrument should be long enough so the last score will not be affected by the respondents’ recollections of the first. An interval that is too short will tend to overestimate reliability. However, the interval should also not be so long that the subject has actually changed in-between, which may underestimate reliability. A typical interval is around 2 to 4 weeks (Scholtes et al. 2011). The traditional psychometric approach to assessing the reproducibility of item-level scores on a rating scale is the computation of intra-class correlation coefficients for each item. However, Rasch-based methods provide more rigorous and comprehensive assessments (Hobart and Cano 2009).

Inter-rater reliability is an estimate of the degree of consensus between the scores of several raters using the same measurement instrument. Inter-rater reliability can be estimated when the raters administer the same instrument to the same person or group of people simultaneously.

Intra-rater reliability refers to the consistency of raters’ scores over time. It is estimated when one rater scores the same instrument on two or more different occasions. Both intra- and inter-rater reliability depend primarily on adequate training of raters and standardisation of the test (Rousson et al. 2002).

Measurement error

Measurement error addresses the absolute amount of measurement error. In contrast to the relative reliability measures, the variation between the subjects does not affect the measurement error (de Vet et al. 2006). The preferred statistics to express measurement error are the standard error of measurement (SEM) and the smallest detectable change (SDC) (Mokkink et al. 2010b). Because of the absolute character of measurement error, both statistics are measured on the same scale as the instrument itself (Bartlett and Frost 2008). SEM is an estimate of the variability expected for observed scores when the true score is held constant (Dudek 1979).

68 Chapter 3: Measurement properties of health rating scales

SDC represents the minimum change that must be overcome to ensure real change. An observed change must, therefore, be higher than the threshold of the SDC to ensure real change (i.e. a change in excess of measurement error).

3.5.1.5 Validity

Validity refers to the degree to which the instrument measures the construct(s) it is purported to measure (Mokkink et al. 2010a, b, c). Validity includes three measurement properties: content validity, construct validity and criterion validity.

Content validity

Content validity refers to the degree to which the content (i.e. items) of the measurement instrument adequately reflects the construct to be measured (Mokkink et al. 2010a, b, c). This includes judgements on the relevance and comprehensiveness of the items in the instrument and the extent to which the content of the instrument is matched to the target population. Content validity is established by a systematic, qualitative approach including focus groups and expert panels. Face validity, which might seem to have a superficial resemblance to content validity, is a judgement of the appearance of the instrument indicating that the item set looks appropriate (Küçükdeveci et al. 2011).

Construct validity

Construct validity refers to the degree to which the scores of the rating scale are consistent with a priori hypotheses (e.g. with regard to internal relationships, relationships to scores on other instruments, or differences between relevant groups) based on the assumption that the rating scale validly measures the construct to be measured (Mokkink et al. 2010a, b, c).

A range of within-scale traditional psychometric analyses has been used to provide evidence of internal construct validity (Hobart and Cano 2009, Küçükdeveci et al. 2011), also known as structural validity (Mokkink et al. 2010a). These include item- total correlations and Cronbach’s alpha coefficients which indicate the degree of interrelatedness among the items, a necessary but not sufficient condition for unidimensionality (Cortina 1993).

The current literature offers two different approaches to the assessment of instrument unidimensionality: Chapter 3: Measurement properties of health rating scales 69

The traditional psychometric approach is exploratory factor analysis or confirmatory factor analysis. The aim of factor analysis is to identify clusters of items that intercorrelate but do not correlate with other clusters of items, thus forming internally consistent sub-scales or ‘components’ which explain the majority of the variance in the scale (Waugh and Chapman 2005). In general, exploratory factor analysis is used when no clear hypotheses exist about the underlying dimensions, whereas confirmatory factor analysis is used to test whether the data fit an a priori hypothesised factor structure (Floyd and Widaman 1995). The traditional psychometric approaches of evaluating the internal construct validity of a scale are based on parametric statistics. It has been argued that parametric statistics are inappropriate in the context of evaluating scales which are based on ordinal data because the assumption of interval scaling (i.e. a scale with known equal intervals between each score across the full range of scores) is not fulfilled (Pallant 2007).

The modern psychometric approach uses Rasch analysis to test how well the items in an instrument conform to a unidimensional model by checking if all of the items in the scale work together to measure a single underlying construct. Rasch analysis is increasingly adopted to provide a more robust interpretation of the internal construct validity of ordinal scales (Hobart and Cano 2009).

These two different methodological approaches are sometimes used in combination to examine instrument unidimensionality and may produce conflicting results (e.g. Waugh and Chapman 2005, Gyagenda and Engelhard 2009, de Morton and Nolan 2011). The aim of this research was to assess the unidimensionality of the WEMWBS using both traditional factor analysis and Rasch analysis.

External construct validity is the association between the rating scale and external criteria such as other measures of similar or different dimensions of health (Strauss and Smith 2009, Küçükdeveci et al. 2011): convergent construct validity is the extent to which scores on a particular instrument are associated with other measures of the same or theoretically similar construct (Campbell and Fiske 1959); discriminant construct validity is the extent to which scores are not associated with theoretically unrelated constructs or variables (Campbell and Fiske 1959, Strauss and Smith 2009).

70 Chapter 3: Measurement properties of health rating scales

Criterion validity

Criterion validity is an estimate of the degree to which the scores of a measurement instrument are an adequate reflection of a ‘gold standard’ (Mokkink et al. 2010a, b, c). Validation of self-report health rating scales is a challenge because the current prevailing opinion is that no gold standard exists for these instruments (Mokkink et al. 2010a). Consequently, it is more appropriate to assess construct validity by testing predefined hypotheses about expected associations with the scores of comparator instruments which measure theoretically related constructs (Mokkink et al. 2010a).

3.5.1.6 Responsiveness

Responsiveness is defined as the ability of an instrument to detect change over time in the construct to be measured (Mokkink et al. 2010a, b, c). An alternative terminology for this property is ‘sensitivity to change’, however ‘responsiveness’ is preferred because concerns have been expressed that use of the former might lead to confusion with the concept of sensitivity used in diagnostic research (Mokkink et al. 2010c). Recommended methods for determining responsiveness and minimally important differences are reviewed by Revicki et al. (2008). Appropriate methods to evaluate responsiveness include calculation of an effect size statistic and the testing of a priori hypotheses concerning the extent and direction of change in relation to comparator instruments.

3.6 Conclusion

Comprehensive evaluation of a rating scale should involve the examination of six psychometric properties: data quality, scaling assumptions, targeting, reliability, validity and responsiveness (McHorney et al. 1994, Ware and Gandek 1998, Hobart et al. 2001a, Hobart and Cano 2009, Cano and Hobart 2011). Although several analyses provide evidence for more than one property, and there is overlap among properties, this list of properties provides a useful framework to describe the methods and results of rating scale evaluations (Hobart and Cano 2009).

Rasch analysis offers significant advantages over traditional psychometric approaches to the assessment of internal construct validity, enabling examination of the contribution of items individually as they are added and removed from the item set, and facilitating the selection of items with the best measurement properties. The application of Rasch analysis to instrument response data from different population Chapter 3: Measurement properties of health rating scales 71

samples, for example occupational groups, provides a mechanism for testing invariance between those groups. This is an important feature where comparisons are to be made. Consequently, both Rasch analysis and traditional methods were applied to increase the comprehensiveness of the evaluation of the psychometric properties of WEMWBS.

3.7 Summary

Rating scales are used to measure unobservable (latent) variables known as theoretical constructs, which are abstract. Latent variables can be measured indirectly by asking questions intended to capture empirically the essential meaning of a construct. The simplest way to do this is to ask a single, straightforward question, or item. However, single items are limited because they are: unlikely to represent the broad scope of a complex theoretical construct; likely to be interpreted in many different ways by respondents; imprecise because they cannot discriminate, to a fine degree, between different levels of an attribute; and unreliable (prone to random error) because they do not produce consistent answers over time (Cano and Hobart 2011). Consequently, rating scales are usually made up of multiple items, in which each item addresses a different aspect of the same underlying construct. Using multiple items overcomes the scientific limitations of single items because: more items increase the scope of a scale; are less open to variable interpretation; enable better precision; and improve reliability by allowing random errors of measurement to average out. Responses to items that are believed to represent a single shared attribute are usually summed to form a composite score to indicate the magnitude of that attribute.

A comprehensive evaluation of a rating scale should involve the examination of six psychometric properties: – data quality, scaling assumptions, targeting, reliability, validity and responsiveness (McHorney et al. 1994, Ware and Gandek 1998, Hobart et al. 2001a, Hobart and Cano 2009, Mokkink et al. 2010a, b, c; Cano and Hobart 2011).

Traditional psychometric methods are based on CTT and have several limitations (Cano and Hobart 2011). These include:

Instrument scores are ordinal rather than interval, the invariance of which is unknown.

72 Chapter 3: Measurement properties of health rating scales

Scores for persons and samples are scale dependent because they lack the provision for varying item parameters, resulting in item parameters that must be regarded as fixed. That is, it is not possible to omit several test questions at different levels of the scale without affecting the individual score (measure).

Scale properties, such as reliability and validity, are sample dependent, varying between sample populations and population subgroups.

These limitations greatly restrict meaningful examination of differences in the severity of the latent variable between populations and population subgroups, and in the interpretation of change scores. The use of modern psychometric methods such as Rasch analysis enables these limitations to be overcome (Cano et al. 2011). This research aims to present a comprehensive approach in examining the psychometric properties of WEMWBS in the context of the veterinary profession by the complementary application of both traditional and modern psychometric methods.

. 73

Chapter 4:

Materials and methods

74 Chapter 4: Materials and methods

4.0 Introduction

The study aimed to examine the measurement properties of WEMWBS in the context of the veterinary profession to inform judgements of the utility of the instrument in this particular occupational group.

The objectives were to:

[a] Explore the properties of WEMWBS using traditional psychometric methods based on classical test theory (CTT).

[c] Re-examine the external construct validity of WEMWBS using the interval- level measurements derived from the Rasch analysis.

This chapter describes in detail the methods used to examine the psychometric properties of WEMWBS in the context of the veterinary profession. Both traditional and modern psychometric methods were applied.

The analyses were based on datasets derived from two independent surveys:

1. Survey 1. A postal questionnaire survey (mailed in October/November 2007) of a large stratified random sample of 3200 veterinary surgeons practising in the UK: A cross-sectional study of mental health and well-being and their associations in the UK veterinary profession. The questionnaire included WEMWBS and several other psychometric instruments relating to mental health and well-being.

2. Survey 2. The Royal College of Veterinary Surgeons (RCVS) Survey of the Profession 2010 administered during January/February 2010 by the Institute of Employment Studies on behalf of the RCVS. The questionnaire was mailed to all of the 23,500 veterinary surgeons registered to practise in the UK. WEMWBS was embedded in this principally employment-related questionnaire. Chapter 4: Materials and methods 75

The methodology followed three chronologically distinct stages:

I. Examination of the psychometric properties of WEMWBS in both datasets using traditional psychometric methods. The convergent construct validity of WEMWBS was examined using the dataset from Survey 1.

II. Rasch analysis to examine the internal construct validity of WEMWBS using subsets of data from Survey 2. The results of the Rasch analysis were then confirmed using a subset of data from Survey 1. A raw score to interval scale transformation was derived for the item set which fit Rasch model expectations.

III. The tests applied in stage I to examine the convergent construct validity of WEMWBS raw scores were repeated using the interval scale measures derived for the item set which fit Rasch model expectations. The purpose of this was to test whether refinement of the scale had impacted on its utility as an overall indicator of mental health and well-being.

4.1 Source of the datasets

Survey 1: A cross-sectional study of mental health and well-being and their associations in the UK veterinary profession

This dataset was derived from a postal questionnaire survey (mailed in October/November 2007) of a stratified random sample of 3200 veterinary surgeons practising in the UK. This number represents approximately 20% of the membership of the Royal College of Veterinary Surgeons (RCVS), excluding those practising overseas. Veterinary surgeons listed in the sampling frame (Vetfile®, Veterinary Business Development Ltd) were stratified according to type of work within the profession and selected at random within each stratum in proportion to the number of veterinary surgeons in each type of work practising in the UK. The questionnaire yielded a response rate of 56.1% (1796/3200). The demographic and occupational profile of respondents was fairly representative of the UK veterinary profession.

The overarching objective of the study was to explore the mental health and well- being of the UK veterinary profession to provide empirical evidence to assess their contribution to the elevated suicide risk and inform the development of strategies to improve psychological health and attenuate suicide risk. The methodology and

76 Chapter 4: Materials and methods

results are reported in detail elsewhere (Bartram 2009a, b, c; Bartram et al. 2009a, b, c; Bartram 2010; Bartram et al. 2010).

A copy of the questionnaire is included in Appendix 2. A summary of the results is presented in Appendix 3.

Survey 2: RCVS Survey of the Profession 2010

This dataset was derived from a postal questionnaire survey administered during January/February 2010 by the Institute of Employment Studies on behalf of the RCVS. The questionnaire, which was mailed to all veterinary surgeons registered to practise in the UK, yielded a response rate of 37.4% (8829/23,594). The aim of the survey was to provide RCVS and other interested parties with contemporary information on the structure, activities and changes within the profession to inform effective governance. Consequently, the content of the questionnaire was mainly employment-related.

Similar surveys have been conducted in 1998, 2000, 2002 and 2006. However, for the first time, the WEMWBS was embedded into the 2010 questionnaire. The decision to include WEMWBS was in response to discussions between DJB and the RCVS which followed the reporting of results from Survey 1 (summarised in Appendix 3). The RCVS recognised the potential value of routine inclusion of an instrument in the Survey of the Profession to measure population-level trends in mental health and well-being over time.

The methodology and descriptive statistics of the results of the survey have been reported by Robertson-Smith et al. (2010). Inferential statistics examining the associations between the WEMWBS score and certain demographic variables have been reported by Bartram (2010a).

A copy of the questionnaire is included in Appendix 4.

4.2 Ethical considerations

Survey 1

The study was approved by Southampton and South West Hampshire Research Ethics Committee (B) in August 2007, Ref: 07/H0504/122. A copy of the letter confirming ethical approval is included in Appendix 5. The work was covered by the public liability insurance of Veterinary Business Development Ltd., the organisation Chapter 4: Materials and methods 77

which provided use of the Vetfile® database as the sampling frame and mailed the questionnaires.

Survey 2

Permission was sought from the RCVS to use the WEMWBS data and relevant associated demographic data from this survey for the following study: Internal construct validity of the Warwick-Edinburgh Mental Well-being Scale (WEMWBS): a Rasch analysis using data from the RCVS Survey of the Profession. The University of Southampton School of Medicine Ethics Committee granted approval for the study in May 2010 (SOMSEC057.10), and the University Research Governance Office confirmed that the University was prepared to act as Research Sponsor in July 2010 (RGO ref 7415). The Royal College of Veterinary Surgeons duly supplied WEMWBS and associated demographic data in a suitable format for the analysis. Copies of the letters confirming ethical approval and sponsorship are included in Appendix 6.

4.3 Instruments

4.3.1 The Warwick-Edinburgh Mental Well-being Scale (WEMWBS) (Tennant et al. 2007).

WEMWBS is a 14-item measure for assessing population positive mental health, which operationalises a wide conception of mental well-being including positive affect (i.e. hedonic aspects of well-being: feelings of optimism, cheerfulness, relaxation), psychological functioning (i.e. eudaimonic aspects of well-being: energy, clear thinking, self-acceptance, personal development, competence and autonomy) and interpersonal relationships (Tennant et al. 2007). Each item is answered on a 5- point Likert scale (1 to 5), based on the respondent’s feelings over the previous two weeks, giving a minimum total score of 14 and a maximum total score of 70; a higher score indicates a higher level of mental well-being.

Evidence to support the acceptability of the psychometric properties of WEMWBS has been derived from UK adult general population (n=1749), university student (n=348) and teenage school student (n=1650) samples (Tennant et al. 2007, Clarke et al. 2011). The scale fulfilled traditional psychometric criteria (including Cronbach’s alphas: 0.87 to 0.91; one-week test-retest reliabilities [intra-class correlation coefficient]: 0.66 to 0.83; and moderate to high correlations with several comparator scales measuring well-being, positive affect or mental health). Rasch analyses

78 Chapter 4: Materials and methods

applied to datasets derived from a subset of the adult general population sample (n=779) resolved a 7-item scale which conformed to Rasch model expectations (other parameters included person separation indices: 0.837 to 0.854; item mean: 0.065 to 0.126; person mean: -0.475 to -0.437) (Stewart-Brown et al. 2009).

WEMWBS was included in Survey 1 and Survey 2.

The section of the Survey 1 questionnaire which includes the WEMWBS items is copied below (Figure 4.1).

Fig. 4.1: WEMWBS items

The Warwick-Edinburgh Mental Well-being Scale was funded by the Scottish Executive National Programme for improving mental health and well-being, commissioned by NHS Health Scotland, developed by the University of Warwick and the University of Edinburgh, and is jointly owned by NHS Health Scotland, the University of Warwick and the University of Edinburgh.

The following instruments were included in the Survey 1 questionnaire only.

4.3.2 The Hospital Anxiety and Depression Scale (HADS) (Zigmond and Snaith 1983).

HADS is a 14-item self-report measure of the prevalence and severity of both anxiety (HADS-A; 7 items) and depressive (HADS-D; 7 items) symptoms, developed for use in general medical out-patient clinics and now extensively validated and widely used in clinical practice and research (Zigmond and Snaith 1983, Herrmann 1997, Snaith 2003). Items in both sub-scales are scored on a 4-point Likert scale (0 to 3), resulting in a range of 0 to 21. A cut-off score of ≥ 8 score is used as an Chapter 4: Materials and methods 79

indicator of ‘caseness’, with score 8 to10 indicating possible case and score ≥ 11 indicating probable case (Bjelland et al. 2002, Snaith 2003). Evidence to support the adequacy of a range of different traditional and modern psychometric properties of HADS has been derived from multiple samples and settings (Pallant and Tennant 2007, Lambert et al. 2011).

The section of the questionnaire including the HADS scales is copied below (Figure 4.2).

Fig. 4.2: HADS items

HADS-A comprises 7 questions: 11, 13, 15, 17, 19, 21, 23; HADS-D comprises 7 questions: 12, 14, 16, 18, 20, 22, 24. HADS copyright © R.P. Snaith and A.S. Zigmond, 1983, 1992, 1994. Record form items originally published in Acta Psychiatrica Scandinavica, 67, 361–70, copyright © Munksgaard International Publishers Ltd, Copenhagen, 1983. Published by GL Assessment Limited, The Chiswick Centre, 414 Chiswick High Road, London W4 5TF, UK. All rights reserved. GL Assessment is part of the Granada Learning Group.

80 Chapter 4: Materials and methods

4.3.3 The Alcohol Use Disorders Identification Test alcohol consumption questions (AUDIT-C) (Bush et al. 1998).

AUDIT-C grades reported alcohol consumption using 3 items which measure frequency of drinking, typical quantity consumed and frequency of heavy drinking, respectively. The scale is approximately equal in accuracy to the full AUDIT (Reinert and Allen 2007). Each question is scored on a 5-point Likert scale (0 to 4), resulting in a range of 0 to 12. Cut-off scores of ≥ 4 for women and ≥ 5 for men are used as an indicator of ‘at-risk’ drinking (Gual et al. 2002). The AUDIT-C has been recommended as a simple and reliable tool for routine assessment of risky drinking (Wallace 2001). ‘At-risk’ drinkers have an increased risk of harmful consequences for their physical and mental health; and dependence. The adequacy of the traditional psychometric properties of AUDIT-C across multiple samples and settings is well-documented in the literature (reviewed by de Meneses-Gaya et al. 2009).

The section of the questionnaire including the AUDIT-C scale is copied below (Figure 4.3).

Fig. 4.3: AUDIT-C items

4.3.4 Three items on suicidal ideation sourced from the second National Survey of Psychiatric Morbidity (Singleton et al. 2001).

The three suicidal ideation items addressed a spectrum of suicidality in the previous 12 months from ‘tiredness of life’, to ‘death wishes’, through to ‘suicidal thoughts’ and were originally sourced from the 5-item questionnaire developed by Paykel et al. (1974). Evidence to support the adequacy of several traditional psychometric properties of the questions has been derived from samples of the general population, medical students and physicians (Tyssen et al. 2001b).

The section of the questionnaire including the suicidal ideation items is copied below (Figure 4.4). Chapter 4: Materials and methods 81

Fig. 4.4: Suicidal ideation items

4.3.5 The Health and Safety Executive Management Standards Indicator Tool (HSE MSIT) (Cousins et al. 2004).

HSE MSIT was developed as a work-related stress risk assessment tool to help reduce stress in the UK working population (MacKay et al. 2004). It is derived from the Job Content Questionnaire (Karasek et al. 1998) and the Whitehall II study questionnaire (Stansfeld et al. 2002) and has acceptable traditional psychometric properties in UK working population samples (Cousins et al. 2004, Edwards et al. 2008, Bevan et al. 2010).

The instrument measures perceptions of psychosocial work characteristics and comprises 35 questions grouped into seven key stressor domains which have the potential to have a negative impact on employee mental health and well-being – demands (workload and working patterns) (8 items), control (the extent to which individuals can control the way they do their work) (6 items), managerial support (the level of support from the organisation and line management in terms of encouragement, sponsorship and resources) (5 items), peer support (level of encouragement and support from peers) (4 items), relationships (the quality of relationships between colleagues) (4 items), role (understanding of duties and responsibilities) (5 items), and change (management and communication of organisational changes) (3 items).

Each question scores 1 to 5 from the least favourable work characteristics (high risk of stress at work) to the most favourable work characteristics (low risk of stress at work) respectively. The overall score for each of the seven stressor domain sub- scales is calculated for each respondent by adding the item scores for each question answered in that scale and dividing by the total number of questions answered in that scale.

The section of the questionnaire including the HSE MSIT items is copied below (Figure 4.5).

82 Chapter 4: Materials and methods

Fig. 4.5: HSE MSIT items

4.3.6 Negative and positive work-home interaction (WHI_N and WHI_P) sub- scales of the SWING instrument (Geurts et al. 2005).

Work and home roles can interact to have both negative and positive effects on mental health, depending on the demands an individual encounters in their work and family situation. Negative effects can arise through competing demands for time and energy; positive effects can arise through the provision of a greater number of resources (e.g. social support, higher self-esteem and financial income) which may buffer against distress in another role (Voydanoff 2004, Oomens et al. 2007).

The WHI_N and WHI_P sub-scales of the SWING instrument comprise a total of 13 items (WHI_N: 8 items; WHI_P: 5 items) for the measurement of time-based and strain-based elements of negative and positive work-home interaction. Each item is scored on a 4-point Likert scale from 0 to 3. The overall score for each sub-scale is Chapter 4: Materials and methods 83

calculated for each respondent by summing the item scores for each of the items answered and dividing by the total number of items. Several psychometric properties of the SWING have been examined in Netherlands working population samples, with satisfactory results (Geurts et al. 2005).

The section of the questionnaire including the WHI_N and WHI_P sub-scales of the SWING instrument is copied below (Figure 4.6).

Fig. 4.6: WHI_N and WHI_P sub-scale items

WHI_N comprises questions 80 to 87; WHI_P comprises questions 88 to 92.

4.4 Data analysis

4.4.1 Direction of scoring

The direction of scale varies between questions within HADS and HSE MSIT. Direction of scale refers to whether the response options to a particular question are scored in the direction of, for example, from 0 to 3 or in the reverse direction, from 3 to 0, as the response options are read from left to right across the page so that, for all questions within a section of the questionnaire, 0 always represents the least severe response and 3 always represents the most severe response. Care was taken to ensure the response options recorded in the dataset were scored in the appropriate direction for each question.

4.4.2 Treatment of missing data

Missing data occur in cross-sectional studies when a respondent accidentally or deliberately omits to respond to an item in a questionnaire (item non-response). In the surveys from which the datasets used in the present study were derived,

84 Chapter 4: Materials and methods

respondents were advised that participation was voluntary and that they could choose to omit answers to any questions. Missing data are problematic because incomplete datasets may lead to results that are different from those that would have been obtained if the dataset was complete. Missing values can be replaced by imputation using the technique which best minimises these differences. Alternatively, cases with missing data can be deleted from the analysis, but this reduces the sample size. Hawthorne and Elliott (2005) examined several procedures for handling missing data in the context of research in psychiatry and advocated the technique of person mean substitution (whereby the imputed value for a variable with missing data is derived from the non-missing items for that person) when data are missing from instrument scales.

The ‘randomness’ of missing data was checked to ensure that certain items were not being systematically missed by respondents or certain demographic groups of respondents.

4.4.2.1 The Warwick-Edinburgh Mental Well-being Scale (WEMWBS)

Survey 1

For up to 3 missing responses, the missing scores were imputed using the mean of the remaining items. If more than 3 of the 14 possible responses were missing, the scale was judged as invalid for that individual and the case was excluded from the analysis. This is the preferred method for the treatment of missing values in WEMWBS (Stewart-Brown and Janmohamed 2008).

Survey 2

No missing data were imputed because this would invalidate the results of the Rasch analysis (Hardouin et al. 2011).

4.4.2.2 The Hospital Anxiety and Depression Scale (HADS)

If a score for a single item within a sub-scale was missing then this was substituted by imputation using the mean of the remaining six items. If more than one item from a sub-scale was missing then the sub-scale was judged as invalid and the case was not included in the analysis. This is a more conservative approach to missing data than is advocated in a number of recent publications where missing values were estimated by assuming that the missing item(s) had a value equal to the average of those in existence (person mean substitution), provided that no more than two Chapter 4: Materials and methods 85

(Kleppa et al. 2008, Stordal et al. 2008) or three items (Jörngården et al. 2006) of a HADS sub-scale were missing. There is less likelihood of a significant p-value with the more conservative approach because fewer cases are included in the analysis (n is lower than if cases with >1 missing value were also included) and there is less attenuation of dispersion (measures of variance, such as SD, are likely to be higher than if cases with >1 missing value were also included as each of these cases would have >1 missing value replaced by the mean of the non-missing values).

4.4.2.3 The Alcohol Use Disorders Identification Test alcohol consumption questions (AUDIT-C)

If Q25 (frequency of drinking) was answered ‘never’ (score 0), a score of 0 was substituted for any missing response(s) to Q26 (typical quantity consumed) and/or Q27 (frequency of heavy drinking). Otherwise, if an item from the scale was missing, the scale was judged as invalid and the case was excluded from the analysis.

4.4.2.4 Suicidal ideation

No missing data were imputed.

4.4.2.5 The Health and Safety Executive Management Standards Indicator Tool (HSE MSIT)

The instrument comprises 35 questions grouped into seven key stressor domains – demands (8 items), control (6 items), managerial support (5 items), peer support (4 items), relationships (4 items), role (5 items), and change (3 items). A sub-scale was considered invalid for a respondent if two or more responses were missing from that scale, except for demands and control sub-scales, for which a scale was considered invalid for a respondent if three or more responses were missing from the scale concerned.

4.4.2.6 Negative and positive work-home interaction (WHI_N and WHI_P)

If a respondent answered fewer than 5 out of 8 items in the WHI_N scale, the scale was considered invalid for that individual; if fewer than 3 out of 5 items were answered in the WHI_P scale, the scale was considered invalid for that individual. Cases with scales defined in this way as being invalid were not included in the analysis.

86 Chapter 4: Materials and methods

4.4.3 Evaluation of psychometric properties

Statistical tests applied to the two datasets are summarised in Table 4.1.

Table 4.1: Summary of psychometric assessments applied to two samples

Dataset from Survey 1 Dataset from Survey 2 Psychometric property Statistical test n n Ordinal scale Interval scale Ordinal scale Interval scale

Data quality Responder bias: chi-square tests 1796 - 8829 - Missing and popular responses† 1796 - 8829 -

Scaling assumptions Corrected item-total correlations 1735 1764 7837 - Item mean and SD equivalence 1735 1764 7837 -

Targeting Floor/ceiling effects (individual items)† 1796 - 8829 - Floor/ceiling effects (total score) 1735 1764 7837 - Skewness, score range 1735 1764 7837 -

Internal construct validity Exploratory factor analysis 1735 1764 7837 7990 Confirmatory factor analysis 1735 1764 7837 7990 Rasch analysis 500 - 500 × 4 -

External construct validity Pearson correlation coefficient 1747-1757 1747-1759 - Multiple linear regression* 1398 1398 - - Multiple logistic regression* 1398-1449 1391-1449 - Group difference validity (gender) 1735 - 7837 - Discriminant validity (vet school) 1735 - 7837

Reliability Internal consistency: - Cronbach’s alpha 1735 1764 7837 - Mean inter-item correlation 1735 1764 7837 - Measurement error Standard error of measurement 1735 1764 7837 -

* Multiple regression after adjustment for: age, gender and other scales (AUDIT-C; HADS-A; HADS-D; HSE MSIT psychosocial work characteristics; suicidal ideation; WHI_N; WHI_P). With imputation of missing values for Survey 1 ordinal scale only. † All items are included for all cases even if there are missing items for a particular case. Otherwise, only cases with no missing WEMWBS values are included in the analyses. 87 88 Chapter 4: Materials and methods

4.4.3.1 Statistical notes, software and sample size issues

Statistical analysis for all traditional psychometric methods was performed using IBM SPSS® Statistics version 17.0 for Windows (SPSS Inc.) except for the following. Parallel analysis to identify the correct number of components to retain during the exploratory factor analytic procedure was performed using Monte Carlo PCA for Parallel Analysis software.5 Confirmatory factor analysis with maximum likelihood estimation was applied to the data using the structural equation modelling software IBM SPSS Amos™ Graphics version 17.0 for Windows (SPSS Inc.).

The data values for several key study variables are either categorical (e.g. suicidal ideation) or ordinal (e.g. Likert-type scales such as WEMWBS raw scores prior to transformation to a linear scale), i.e. the response categories have a rank order but the intervals between values are not necessarily equal. Consequently non- parametric statistical methods may be more appropriate, but the subject is contentious (Jamieson 2004, Pell 2005). However, parametric statistics were applied for traditional psychometric methods to enable adjustment for confounding factors when evaluating associations between the scales. Histograms were plotted to check that the assumption of normality of distribution was met. A review of the literature on the use of ordinal scales in statistical procedures concluded that departure from the interval-level measurement assumption does not necessarily increase Type I and Type II errors (Jaccard and Wan 1996). However, in the present analyses the results of parametric tests with ordinal data should be interpreted with appropriate caution.

The 95% confidence intervals (CIs) were computed for several key study variables to provide a measure of the precision of the estimates. Some of the confidence intervals were computed using Confidence Interval Analysis (CIA) software version 2.1.2.6

5 WATKINS, M.W. (2000) Monte Carlo PCA for parallel analysis [computer software], Ed & Psych Associates, State College, PA. Available at: http://www.allenandunwin.com/spss3/Files/MonteCarloPA.zip Accessed: 03 September 2011

6 BRYANT, T.N. (2000) Computer software for calculating confidence intervals (CIA). In Statistics with Confidence. Eds D.G. Altman, D. Machin, T.N. Bryant, M.J. Gardner. 2nd edition. London, BMJ Publishing Group. pp 208-213 [CIA software accompanies the book and is also available at http://som.soton.ac.uk/cia/] Chapter 4: Materials and methods 89

Confidence intervals computations for Pearson correlation coefficients were based on the Fisher r-to-z transformation, using software available online.7

Analysis of each dataset involved multiple comparisons which increases the likelihood of finding statistically significant associations by chance (Type I error). Bonferroni adjustment was made to p-values to account for the multiplicity (Bland and Altman 1995). Statistical significance was defined as a two-tailed p-value of < 0.05.

Based on the recommendations of Pallant (2007), both Survey 1 and Survey 2 sample sizes were deemed adequate for exploratory factor analysis. Data were initially inspected to ensure correlations of at least 0.3 existed between items. The Kaiser-Meyer-Olkin (KMO) Measure of Sampling Adequacy and Bartlett’s Test of Sphericity were inspected to check for a value of ≥ 0.6 or above and a p-value of < 0.05, respectively.

Rasch analysis was performed using Rasch Unidimensional Measurement Model software (RUMM2030 Standard Edition) (Andrich et al. 2010). Within the context of Rasch analysis, robust estimation of item parameters (within ± 0.5 logits [contraction of log-odds probability units] at α = 0.01) is considered achievable with a sample size of n = ≥ 250, irrespective of the targeting of items to persons (Linacre 1994).

The psychometric tests applied to the datasets and the criteria used as indicators of acceptable performance are summarised in Table 4.2.

7 LOWRY, R. Vassar Stats: web site for statistical computation. Available at: http://faculty.vassar.edu/lowry/VassarStats.html Accessed: 22 September 2011

Table 4.2: Psychometric properties, their definitions and criteria used to indicate acceptable performance

Psychometric property Definition Test(s) Criteria for acceptability

Data quality The extent to which WEMWBS 1. Percentage of missing data for each 1. Item-level missing data is < 10% (Cano et items are scored and WEMWBS item. al. 2008). total scores can be computed. 2. Percentage of cases for which a total 2. Total scores are computable for ≥ 95% of score can be computed (≥ 11 respondents (Smith et al. 2005). WEMWBS items completed) (Stewart- Brown and Janmohamed 2008).

Scaling assumptions The extent to which it is legitimate Summing of WEMWBS item scores is to sum a set of component considered legitimate, when the items: scores, without weighting or standardisation, to produce a 1. Are approximately parallel ( i.e. measure 1. Satisfied when items have similar mean single total score (Hobart and at the same point on the scale). scores and item response option Cano 2009). frequency distributions (McHorney et al. 2. Contribute similarly to the variation of 1994). the total score (i.e. they have similar variances), otherwise these should be 2. Satisfied when item scores have similar standardised. SDs (McHorney et al. 1994).

3. Measure a common underlying 3. Satisfied when corrected item-total score construct (i.e. mental well-being). correlations are in the range of 0.2 to 0.8 Otherwise it is not appropriate to (Cheung et al. 2000). combine them to generate a total score. 4. Satisfied when corrected item-total 4. Contain a similar proportion of correlations are roughly similar and information about the construct. exceed 0.30 (Hobart and Cano 2009). Otherwise components should be given different weights. Continued overleaf

Psychometric property Definition Test(s) Criteria for acceptability

Targeting The extent to which the range of Score distributions were examined at both No item has an aggregate endorsement the variable measured by the the item and the scale level. frequency of < 10% for two or more scale (here mental well-being) adjacent response categories (Smith et al. matches the range of that 2005). variable in the study sample. Maximum endorsement frequency for any item response category is < 80% (Cano et al. 2004).

Scores should span the entire scale range.

Floor and ceiling effects are < 15% (McHorney and Tarlov 1995).

Skewness statistics range from –1 to +1 (Cano et al. 2008).

Reliability Reliability is the extent to which Two types of reliability were examined to 1. Adequate scale internal consistency is scale scores are associated with quantify random error: indicated by a Cronbach’s alpha random error. High reliability coefficient of 0.70 to 0.95 (Terwee et al. indicates that scores are 1. Internal consistency reliability estimates 2007); mean inter-item correlation of 0.40 associated with little random the random error associated with total to 0.50 (Clark and Watson 1995). error, i.e. scores are consistent. scores from the intercorrelations among the components. 2. SEM was considered adequate if it represented < 10% of the total possible 2. Standard error of measurement (SEM) score range (van Baalen et al. 2006), i.e. gives an indication of the precision of < 5 in the case of WEMWBS (score range the total score. 14 to 70).

Continued overleaf 91

Psychometric property Definition Test(s) Criteria for acceptability

Validity The extent to which a scale Several aspects of construct validity were measures what it is intended to examined: 1. It was hypothesised that the WEMWBS measure. This is essential for the would show negative associations with meaningful interpretation of 1. Convergent construct validity was HADS, suicidal ideation and WHI_N and scores (Mokkink et al. 2010a, b, examined by computing associations positive associations with HSE MSIT sub- c). between WEMWBS and HADS, HSE scales. Evidence of construct validity MSIT psychosocial work characteristics, demonstrated if ≥ 75% of hypotheses suicidal ideation and WHI_N. confirmed (Terwee et al. 2007).

2. Discriminant construct validity was 2. It was hypothesised that WEMWBS examined by computing the association scores would not be influenced by this between WEMWBS score and variable and, therefore, the association veterinary school of graduation. would not be statistically significant.

3. Group difference validity was examined 3. It was hypothesised that the mean score by comparing the mean WEMWBS for men would be significantly higher than scores for men and women. for women.

4. Exploratory factor analysis (EFA) 4. Evidence from factor analysis that the followed by confirmatory factor analysis scale measures a single construct (CFA). (unidimensionality).

5. Rasch analysis. 5. Fit of an item set to the Rasch model, i.e. person abilty estimate for the two most different subsets of items differs significantly in ≤ 5% of respondents (Tennant and Conaghan 2007).

Chapter 4: Materials and methods 93

4.4.3.2 Traditional psychometric methods

The traditional psychometric paradigm is based upon correlational or descriptive analyses to evaluate data quality, scaling assumptions, targeting, reliability, validity and responsiveness (Hobart and Cano 2009, Cano and Hobart 2011).

WEMWBS data were examined for the following psychometric properties:

Data quality

Indicators of data quality reflect respondents’ understanding and acceptance of a rating scale and help to identify items that may be irrelevant, confusing or upsetting to the target population (McHorney et al. 1994). A scale score cannot be confidently estimated if item-level missing data are high. The percentage of missing data for each item and the percentage of total scores that could be computed were calculated. The criterion for acceptable item-level missing data was < 10% (Cano et al. 2008). It has been suggested that when responses for a case are missing, a total score can be calculated if at least 50% of the items (i.e. ≥ 7 WEMWBS items) have been completed (Cano et al. 2008). However, the more stringent criterion of ≥ 11 completed WEMWBS items was applied in the present study (Stewart-Brown and Janmohamed 2008). The criterion applied for the acceptable proportion of respondents with computable total scores was ≥ 95% (Smith et al. 2005). All analyses of percentage missing items did not include imputed items. The impact of age and gender on data quality was assessed by the use of chi-square tests to compare the age and gender of complete responders with incomplete or non- responders to the WEMWBS scale.

Scaling assumptions

Tests of scaling assumptions determine whether it is legitimate to generate scores for a rating scale using the algorithms proposed by the developers of the scale (Hobart and Cano 2009). In the case of WEMWBS these tests assess the legitimacy of summing individual item scores to generate an overall score.

Items in the scale should be roughly parallel, i.e. measure at the same point on the scale and have similar variance, otherwise they do not contribute equally to the variance of the total score and should be standardised before combination (Hobart and Cano 2009). This criterion was evaluated by

94 Chapter 4: Materials and methods

examining the symmetry of item response distributions and the equivalence of item mean scores and standard deviations (SDs) (McHorney et al. 1994).

Items in the scale should measure a common underlying construct, otherwise it is not appropriate to combine them to generate a total score. This criterion was evaluated by examining the correlation between each item and the total score computed from the remaining items in the scale (corrected item-total correlation). Corrected item-total score correlations (the correlation of an item score with the summed score for all other items) of 0.2 to 0.8 were sought (Cheung et al. 2000).

Items in the scale should contain a similar proportion of information concerning the construct being measured. This criterion was considered satisfied if the corrected item-total correlations were roughly similar and exceeded 0.30 (Hobart and Cano 2009).

Targeting

Targeting is the extent to which the spectrum of health measured by a health rating scale matches the distribution of health in the study sample, and indicates whether the scale is an acceptable instrument for that sample (McHorney and Tarlov 1995). The better the match, the greater the potential for precise measurement. Targeting was studied at both the item and the scale level. At the item level, it is recommended that scores are evenly distributed across the item response categories, with maximum endorsement frequencies (i.e. the proportion of respondents who endorse each response category) for any item response category < 80% (Cano et al. 2004). No item should have aggregate adjacent endorsement frequencies (i.e. the sum of the endorsement frequencies for any two or more adjacent response categories) of < 10% (Smith et al. 2005). At the scale level, targeting of WEMWBS was evaluated by examining score distribution, score means, standard deviation (SD), skewness statistics, and floor and ceiling effects. It is recommended that scale scores should span the entire range of the scale, the mean score be near the scale midpoint, and skewness statistics should range from –1 to +1 (Cano et al. 2008). Floor effect is the percentage of respondents scoring 14 (lowest level of mental well-being) and ceiling effect is the percentage of respondents scoring 70 (highest level of mental well-being). An upper limit of 15% was sought for floor or ceiling effects (McHorney and Tarlov 1995). Chapter 4: Materials and methods 95

Internal consistency reliability

Internal consistency is a function of the number of items and their covariation within a scale which measures a single construct, such as mental well-being. Two types were examined.

Cronbach’s alpha (Cronbach 1951) estimates the random error associated with scores from the intercorrelations among the items. A Cronbach’s alpha estimate of 0.70 to 0.95 (Terwee et al. 2007) was sought.

One well-known limitation of Cronbach’s alpha is that it is dependent on the number of items in a scale: the larger the number of items, the higher the alpha (Cortina 1993). The implication of this is that the relationships among items might be ‘hidden’ by a high alpha coefficient when the number of items is relatively large. One way to address this problem is to also report the mean inter-item correlation (the homogeneity coefficient) (Hobart and Cano 2009). As the WEMWBS has 14 items, reporting of the homogeneity coefficient was considered to be potentially informative. A mean inter-item correlation of 0.40 to 0.50 was sought (Clark and Watson 1995).

Standard error of measurement (SEM)

Another potentially useful reliability index is the standard error of measurement (SEM). It is computed as SEM = SD × √ (1 – reliability) (Dudek 1979, Hobart and Cano 2009) and is an indicator of scale precision (Lohr 2002). Suitable reliability parameters on which the calculation may be based include the test-retest reliability intra-class correlation coefficient (Terwee et al. 2007) and Cronbach’s alpha coefficient of internal consistency (Hobart and Cano 2009, Tighe et al. 2010). The SEM computation was based on Cronbach’s alpha in the present study. The SEM makes the reliability of a scale more tangible and clinically meaningful by enabling computation of the 95% confidence intervals around an individual person’s scores, computed as ± 1.96 SEM (assumes normal distribution) (Hobart and Cano 2009). However, the index has several limitations, including its dependence on the distribution of the sample and the implication that the standard error is constant across the scale range (Hobart and Cano 2009).

SEM was considered adequate if it represented < 10% of the total possible score range (van Baalen et al. 2006, van der Zee et al. 2010), i.e. < 5 in the case of the

96 Chapter 4: Materials and methods

WEMWBS (score range 14 to 70). The units of SEM are the same as for the ordinal scale from which the statistic is derived (Bartlett and Frost 2008).

Convergent construct validity

Validity is the extent to which an instrument measures what it is intended to measure. Convergent construct validity is the extent to which scores derived from a particular instrument are associated with different measures of the same or theoretically similar construct (Strauss and Smith 2009). Ideally, for the results of associations between a scale and comparator scales to be fully interpretable, evidence should exist of the reliability and validity of the comparator scales in similar population samples. However, the extent to which supporting evidence for the reliability and validity of these external measures will vary. This is a common limitation of external validity testing and the findings should be interpreted with this in mind (Cano et al. 2008).

To assess convergent construct validity (Campbell and Fiske 1959), Pearson correlation coefficients were computed to explore possible pairwise correlations between scores on the WEMWBS and scores on HADS-A, HADS-D and HSE MSIT psychosocial work characteristics sub-scales, using the dataset from Survey 1. In addition, using the same dataset, multiple regression was applied to estimate any associations between WEMWBS and AUDIT-C; HADS-A; HADS-D; HSE MSIT sub- scales; suicidal ideation; and WHI_N after adjustment for age and gender and each of the other scales. These instruments were considered to be appropriate as comparators to evaluate convergent construct validity because collectively they measure a range of psychological dimensions consistent with the research question (Section 1.5) concerning the validity of WEMWBS as an overall indicator of population mental health and well-being. Moreover, there is evidence to support the adequacy of several psychometric properties of the comparator instruments, which has been derived from multiple samples and settings, albeit not within the context of the veterinary profession (see Section 4.3).

The following a priori hypotheses were developed based on the item content of each instrument and expectations regarding the proximity of the constructs that they are intended to measure:

WEMWBS would show moderate negative correlations (r = -0.3 to -0.7) with HADS, and moderate positive correlations with several HSE MIST sub- Chapter 4: Materials and methods 97

scales including demands. The directions of correlation are different because an increase in HADS score signifies higher levels of anxiety and depressive symptoms; an increase in HSE MIST score signifies more favourable psychosocial work characteristics.

In multiple linear regression (WEMWBS as the dependent variable), suicidal ideation, high WHI_N,8 and one unit increases in the following scores: HADS-A; HADS-D; HSE MIST demands sub-scale; would each be associated with a decrease in WEMWBS score.

In multiple logistic regression (WEMWBS as the independent variable), a one unit increase in WEMWBS score would be associated with decreased odds of reporting HADS-A caseness (HADS-A score ≥ 8), HADS-D caseness (HADS-D score ≥ 8), and suicidal ideation.

Adjustment was made for age, gender and each of the other scales in the multiple regressions. It was considered that the results would demonstrate evidence of construct validity if ≥ 75% of these hypotheses were confirmed (Terwee et al. 2007). The tests of associations between scales in the Survey 1 dataset that are described above were repeated after transforming the raw WEMWBS scores to interval scale measurements for the refined version of WEMWBS which fitted the Rasch model. The objective was to check whether utility as an overall measure of mental health and well-being was retained with the refined measure. A score transformation table derived from the Rasch analysis was used for this purpose.

Discriminant construct validity

Discriminant construct validity is the extent to which scores are not associated with theoretically unrelated constructs or variables in the same conceptual domain (Strauss and Smith 2009), i.e. the extent to which a scale ‘does not measure what it is not designed to measure’ (Riazi et al. 2006, p. 1398). To assess discriminant construct validity, an a priori hypothesis was developed about the expected association between WEMWBS score and the country/region from which the respondents graduated. It was hypothesised that WEMWBS score would be unrelated to country/region of graduation and therefore the difference in mean scores across this variable would not be statistically significant. Country/region of

8 High WHI_N was defined as a score of > 1 SD above the mean score on the WHI_N scale, i.e. > 1.76.

98 Chapter 4: Materials and methods

graduation was categorised into UK, overseas (EU) and overseas (non-EU). This hypothesis was tested by applying one-way ANOVA tests.

Group difference validity

Group difference validity is the extent to which scores can discriminate between groups which are predicted to have differences in the level of the underlying construct (Hobart et al. 2001b, Hobart and Cano 2009). This property is also referred to as ‘known groups validity’ (e.g. Food and Drug Administration 2009, Leentjens et al. 2011).

To assess group difference validity, an a priori hypothesis was developed about the expected association between WEMWBS score and gender, based on the findings of recent UK psychiatric morbidity studies (Singleton et al. 2001, Braunholtz et al. 2007, McManus et al. 2009) and the results of WEMWBS in a UK general population sample (Tennant et al. 2007). It was hypothesised that the mean score for men would be higher than for women. The significance of the difference was evaluated using independent t-tests after checking for homogeneity of variance (Levene’s test).

Unidimensionality

Factor analytic techniques such as principal components analysis (PCA), based on a classical test theory approach, and Rasch analysis, based on a modern test theory approach, are two possible methods for assessing instrument unidimensionality (Hobart and Cano 2009). In the present study, the unidimensionality of the WEMWBS instrument was assessed by the application of both PCA and Rasch analysis (see Section 3.4.2.2 below). Evidence to support the internal construct validity of the WEMWBS would be provided by results which indicate that the 14 items are best considered as a single construct.

Exploratory factor analysis (EFA) was applied to assess the nature of the constructs influencing the response data. EFA was performed using the principal component analysis (PCA) steps described by Pallant (2007) to extract factors. Only cases with no missing WEMWBS values were used in the analysis. The data were assessed for suitability for PCA by checking that the following criteria were met: correlations between some items exceeded 0.3; the Kaiser-Meyer-Olkin Measure of Sampling Adequacy was ≥ 0.6; and the Bartlett Test of Sphericity yielded a significant value (p<0.05). An unrotated analysis was then performed and the results examined to Chapter 4: Materials and methods 99

estimate the number of components that should be extracted. The appropriate number of components was determined using the Kaiser criterion (eigenvalue ≥ 1) and visual examination of the scree plot (Cattell 1966). The scree plot is also based on eigenvalues but uses their relative rather than absolute values as a criterion. The point of abrupt transition from vertical to horizontal is identified (the ‘elbow’). The components proximal to the elbow are deemed potentially suitable to be retained. The number of components was confirmed by a parallel analysis (Horn 1965). Parallel analysis involves comparing the size of the eigenvalues with those obtained from a Monte Carlo procedure to randomly generate 100 data sets of the same size. Only those eigenvalues that exceed the corresponding values from the random data set are retained. This approach to identifying the correct number of components to retain has been shown to be the most accurate, with both Kaiser’s criterion and Cattell’s scree test tending to overestimate the number of components (Zwick and Velicer 1986). PCA analysis was performed for WEMWBS and the Rasch-derived refined WEMWBS on Survey 1 and Survey 2 datasets.

Confirmatory factor analysis (CFA) was applied using the steps described by Arbuckle (2008) to test the fit of the factor structure indicated by the results of the EFA. Only cases with no missing WEMWBS values were used in the analysis. Since no single fit index is perfectly reliable, it has been suggested that several indices should be used to estimate fit (Sharma et al. 2005). The chi-square statistic assesses overall model fit in covariance structure models, but it is sensitive to sample size. A sufficiently large sample tends to yield a significant value of chi- square (Bentler and Bonett 1980, Cole 1987), which indicates lack of satisfactory fit. Hence, the chi-square statistic has limited value (Kaplan 1990). The statistic was reported in the present study but was not used as a criterion to assess fit. A comparative fit index (CFI) > 0.9 and a root mean square error of approximation (RMSEA) < 0.08 were the criteria used to indicate acceptable fit (McDonald and Ho 2002).

4.4.3.3 The application of Rasch analysis

Rasch analysis was applied to examine the internal construct validity of WEMWBS. The method was guided by the recommendations of Tennant and Conaghan (2007) concerning the steps to be followed and the results that should be reported.

Several imputation methods produce bias on psychometric indices in the context of Rasch analysis. Generally, imputation methods artificially improve the psychometric

100 Chapter 4: Materials and methods

qualities of the scale. This is especially the case with the person mean score method of imputation, which was applied to missing data in the Survey 1 dataset for several of the traditional psychometric tests described above (Hardouin et al. 2011). Consequently, Rasch analysis was applied to datasets without imputation of missing WEMWBS scores.

Cases with no missing WEMWBS scores were selected in the Survey 2 dataset (n=7837). Four random samples of n=500 cases were drawn, without replacement, from these selected cases for use in the Rasch analysis (identified as subsets [S2_500a], [S2_500b], [S2_500c] and [S2_500d]). Rasch analysis was applied to subsets of the full dataset because there is evidence that some Rasch fit statistics for scales with polytomous response categories (i.e. scales in which the items have > 2 response options, like WEMWBS) are highly dependent on sample size, which translates into a greater likelihood of Type I errors (falsely rejecting an item as not fitting the Rasch model) with increased sample size (Smith et al. 2008). Use of a smaller sample size helps to identify only those deviations between the model and the data that are of practical significance. In order to obtain robust estimates of the internal construct validity of the scale, the findings of Rasch analysis on subset [500a] were cross-validated on each of the other subsets from the Survey 2 dataset and a random sample of 500 cases with no missing WEMWBS values drawn from the Survey 1 dataset (identified as S1_500). The results should be consistent across each of the 5 subsets of data (Figure 4.7). Chapter 4: Materials and methods 101

Fig. 4.7: Samples used in the Rasch analysis and output validation

S1 and S2 are random samples of 500 complete cases drawn without replacement from the Survey 1 and Survey 2 datasets respectively.

The sequential steps of the Rasch analysis are illustrated in Figure 4.8. It is an iterative process with the researcher systematically accumulating evidence of the validity of the responses. No single statistic is generally sufficient to decide whether a set of data fits the model. Each analysis is a case study in determining the diagnostic evidence for the internal consistency and validity of the data. Both statistical and graphical evidence are used simultaneously to inform judgements such as whether to discard or modify an item.

102 Chapter 4: Materials and methods

Fig. 4.8: The Rasch analysis process

Decide which parameterisation Likelihood ratio test of the Rasch model to use: Rating Scale Model or Partial Credit Model.

Check all persons are within a fit residual range of ± 2.5. Individual person fit Consider removing any misfits from the analysis.

If any item pair correlations are Local response dependency > 0.3, either combine the items into a ‘super-item’ or remove one of the items.

Remove items that are Individual item fit significantly under- or over-discriminating.

Visual inspection of item threshold maps for each item. Item threshold ordering If thresholds overlap, either merge adjacent response Criteria categories or delete item. NOT met

Goodness of fit The lower bound of the 95% CI for the proportion of significant Unidimensionality test tests should be ≤ 5%.

Criteria Criteria NOT met Visual inspection of DIF maps met Differential item functioning for each item. Exclude items DIF demonstrating significant DIF unless DIF cancels at test level.

Criteria met Visual inspection of person- Targeting item threshold map.

For the refined scale, check Criteria each of above parameters in met Validation each of the four validation datasets. Are they acceptable in all datasets?

Pool the datasets: S2_500a; Transform raw score S2_500b; S2_500c; S2_500d; Conversion to interval measure S1_500. Then obtain raw score table to logit score transformation.

Chapter 4: Materials and methods 103

Selecting the correct Rasch model

For instruments such as the WEMWBS which have polytomous items (i.e. items have > 2 response options), there are two parameterisations of the Rasch model that can be assessed using the RUMM software: the rating scale model or the partial credit model. These two models differ slightly in their mathematics. The former expects the distances between thresholds to be equal across items (Tennant and Conaghan 2007). This means that the metric distance between, for example, the thresholds separating categories 1 and 2, is the same across all items, and that the metric distance separating categories 2 and 3 is the same across all items. However, the distance between categories 1 and 2 does not have to be the same as the distance between categories 2 and 3. The rating scale model provides a higher degree of specificity. However, it is not always possible to satisfy the assumptions of a rating scale model, in which case the partial credit model should be used. The likelihood ratio test provides a test of which version of the model should be used by comparing the different parameterisations of the model and providing a chi-square statistic and probability. The rating scale model should be adopted if the outcome of the test is not significant (p > 0.05) (Tennant and Conaghan 2007).

The steps described below were applied iteratively to data subset [S2_500a]. This enabled an examination of the contribution of items individually as they were added and removed from the item set and facilitated the selection of items that provided optimal fit to the Rasch model.

Class intervals

Respondents are automatically placed into groups called class intervals. Class intervals are defined by ordering all respondents in terms of ability (level of mental well-being), determined by the responses to all items combined, and then splitting them into groups of approximately equivalent size across the sample (Tennant and Conaghan 2007). Several Rasch fit statistics are applied at the class interval level. Nine class intervals were used in this research.

Response option thresholds ordering

Each of the 14 items had a polytomous response format, scored ranging from 1 (‘none of the time’) to 5 (‘all of the time’). Responses to these options need to follow a logical sequence. It would be expected that as mental well-being increases for a respondent, it becomes more likely to score a 0, then a 1, then a 2, then a 3 and so

104 Chapter 4: Materials and methods

up to 5 for any particular item. In Rasch analysis terms, this would be indicated by an ordered set of response option thresholds for each item (Pallant et al. 2006). A threshold refers to the point between two adjacent response categories (e.g. 1 ‘none of the time’ and 2 ‘rarely’) where the probability of scoring a 1 on the item or scoring a 2 is 50/50 (given that the person will only score a 1 or a 2) (Pallant and Tennant 2007) (see Figure 3.2). Disordered thresholds can occur when respondents have difficulty consistently discriminating between response options. Threshold ordering (i.e. transition between adjacent response categories) for each item was assessed by comparing the logit ability rating for each response category using each item’s threshold category curve. An example of ideal response option threshold ordering curves is illustrated in Figure 4.9.

Fig. 4.9: Ideal response option thresholds ordering

Ideal category probability curves demonstrating ordered response thresholds

(T1 < T2 < T3 < T4). The term threshold refers to the point between two adjacent response options where either response is equally probable.

Tests of individual item fit

Tests of individual item fit to the Rasch model reflect the differences between the observed responses and those expected by the model. The following three indicators were examined: (1) fit residuals (item-person interaction), (2) chi-square probability values (item-trait interaction), and (3) item characteristic curves (ICCs).

A residual is a summation of individual item (or person) deviations from model expectations, which are then standardised to form a z-score. Residual scores Chapter 4: Materials and methods 105

between ± 2·5 are considered to indicate adequate fit to the model (Pallant and Tennant 2007).

The chi-square statistic tests whether the differences between the observed values and expected values across the class intervals for each item are statistically significant. A Bonferroni adjusted, non-significant chi-square statistic, indicates good fit to the model (Pallant and Tennant 2007). Bonferroni adjustment is applied to the alpha level for the number of items because multiple tests are performed, one for each chi-square statistic for each item (Hagquist et al. 2009).

An ICC is a graphical plot of the expected response (predicted from the model) to an item at each level of the measurement continuum.

Items with poor fit to the model were considered for removal from the scale.

Tests of individual person fit

A few respondents with response patterns that deviate from model expectation (identified by individual person fit residuals outside the acceptable range of ± 2·5) may cause serious misfit at the item level. This has potential to lead to a scale being discarded due to inadequate fit to the Rasch model, when it may be more appropriate to explore why these individuals have responded differently from others in the sample. A high negative person fit residual is indicative of fitting the model ‘too well’, with a response pattern very close to the ideal Guttman pattern, and is generally regarded as being less problematic than a high positive person fit residual (Muller and Roddy 2009). The latter may be due, for example, to cognitive impairment (inability to understand the questions), guessing, or lack of concentration (Tennant and Conaghan 2009). Removal of these misfitting respondents from the analyses can make a major difference to the results of tests of internal construct validity (Pallant and Tennant 2007). Moreover, the existence of such aberrant response patterns implies that the external construct validity of the scale may be inadequate for the sample population concerned (Pallant and Tennant 2007).

Differential item functioning

Another potential source of misfit in the data is differential item functioning (DIF). In this form of bias, subgroups of the sample population respond in a different manner to an item despite equal levels of the underlying characteristic being measured (Teresi and Fleishman 2007). The type of variables tested for DIF depends on the

106 Chapter 4: Materials and methods

factors thought to have a potential impact on responses to the items. In the present analysis, gender and age were considered as potential factors. For this purpose, the age of respondents was categorised into 4 different groups based on age quartiles computed from the full Survey 2 dataset of respondents with no missing WEMWBS values (under 33 years; 33 to 42 years; 43 to 54 years; 55 years and over). DIF is tested using analysis of variance (ANOVA) and a Bonferroni adjusted statistically significant probability indicates significant DIF.

A distinction in DIF analyses is between uniform and non-uniform DIF (Teresi and Fleishman 2007). Uniform DIF is indicated by a consistent systematic difference in item response between groups across the entire spectrum of the construct to be measured (e.g. mental well-being). For example, where one group displays a consistently higher or lower score on a given item relative to the overall ability judged by the respondents’ responses to all of the items aggregated together. Non- uniform DIF occurs when the direction of the difference between groups varies across different levels of the underlying construct, i.e. across class intervals.

Differential item functioning should be accounted for in order to be able to make invariant comparisons of summed scale scores between groups. Uniform DIF can be adjusted for by splitting the item into two new items, one for each subgroup, thereby controlling the bias and retaining the information from the item (Lawton et al. 2006). Alternatively, items displaying significant DIF should be considered for as candidates for removal from the scale. In the present study, items with DIF of similar magnitude and opposite direction were grouped into a subtest to investigate whether the DIF cancelled out at the test level. If the DIF cancelled out at the test level then the items were retained.

Response dependency

In order to conform to the Rasch model, the items are required to be locally independent, i.e. the entire correlation between the items has to be accounted for by the latent variable (mental well-being in this research). If there are significant correlations between item pairs after the contribution of the latent variable is removed (i.e. among the residuals), then the items are locally dependent. Response dependency occurs when items are linked in some way, such that the response to one item will determine the response to another, either because the items duplicate some feature of each other or because they both incorporate some other shared dimension. An example of this is where several walking items are included in the Chapter 4: Materials and methods 107

same mobility scale. If a person can walk several miles without difficulty, then that person must also be able to walk 1 mile, or any shorter distance, without difficulty. Response dependency can be identified by inspecting the residual correlation matrix and dealt with by item deletion or combining the items into a subtest. In the example above, the latter would be the equivalent of creating one walking-related item with response options that concern how far a person can walk. In this research, a residual correlation value of > 0.3 between the response patterns of any two scale item pairs was considered to indicate response dependence (La Porta et al. 2011).

Fit of the remaining item set to the Rasch model

After any corrections to the ordering of response categories and removal of items on the basis of poor individual item fit to the model and/or DIF, tests were applied to the remaining item sets to examine how well they conformed to the Rasch model.

Estimates of the internal consistency reliability of the item set were based on the person separation index (PSI). The PSI is analogous to Cronbach’s alpha used in traditional psychometric methods, but has the linear person ability estimate (logit value) from the Rasch model substituted for the ordinal raw score in the same formula (Tennant and Conaghan 2007). Reliability for group-level measurement is indicated by a PSI value ≥ 0.7 (Reeve et al. 2007, Tennant and Conaghan 2007). This means that it is possible to statistically differentiate between two subgroups of respondents (Fisher 1992). However, there are fundamental differences between the PSI and Cronbach’s alpha. For example, the PSI is expressed entirely in terms of person locations and so meets the true definition of reliability, and it is computed from linear measurements rather than raw scores (Hobart and Cano 2009).

Confirmation of unidimensionality is a prerequisite for the summation of items within a scale. If a scale is unidimensional, no associations (factor structure) within the first principal component of the residuals should exist once the factor accounting for the principal associations (the ‘Rasch factor’) has been extracted. To test this formally, the items were divided into two subsets based on whether the item loaded positively or negatively on the first principal component of the residuals. Then, t-tests were applied to compare the person ability estimates derived using only the items in these subsets. For a unidimensional scale, the proportion of statistically significant tests is expected to be < 5% (Pallant et al. 2006). If > 5% of the person ability estimates is significantly different, a binomial test of proportions is used to calculate the 95% confidence interval around the estimate. Unidimensionality is said to be supported if

108 Chapter 4: Materials and methods

the value of 5% falls within the lower bound of the 95% confidence interval (Tennant and Pallant 2006, Tennant and Conaghan 2007).

Targeting

The person-item threshold distribution was inspected visually to assess how well the level of difficulty represented by the final set of items and their response options was matched to the ability of the respondents. This graph places ability of respondents and difficulty of item responses on the same logit scale. Comparison of the mean location score for persons with the mean location score for the items (centred at zero) provides an indication of how well targeted the items are for people in the sample. For a well-targeted measure, the mean location for persons should be around the value of zero (Tennant and Conaghan 2007).

Transformation from raw score to interval measurement

The transformation from raw score to interval measurement was computed using a dataset comprising all respondents with no missing item response data from Surveys 1 and Survey 2 combined. Use of the large combined dataset enhanced the precision of the estimates irrespective of the targeting of the scale (Lundgren- Nilsson and Tennant 2011).

The raw scores were mapped against the corresponding estimated Rasch- transformed logit scores for the set of fitting items. The raw scores with their corresponding logit scores were copied and pasted into Microsoft Office Excel 2003 and the logit scores were rescaled into the score range of the fitting item set using the formula (Cieza et al. 2009): y = m + (s × person location), where: s = (desired range) / (current range in logits); m = (desired minimum scale value) – (current minimum scale value in logits × s)

4.5 Summary

The psychometric properties of data quality, scaling assumptions, targeting, reliability and validity were rigorously evaluated by the complementary application of traditional and modern psychometric methods to data from two large cross-sectional samples of the veterinary profession (n = 1796 and n = 8829) derived from independent postal surveys. Where possible, the findings were cross-validated Chapter 4: Materials and methods 109

between samples. External construct validity was assessed by testing a priori hypothesised associations between WEMWBS scores and those on comparator instruments measuring theoretically related constructs. Internal construct validity was examined using traditional factor analytic techniques and Rasch analysis. The application of Rasch analysis to the data enabled the assessment of response option thresholds ordering, individual item fit, individual person fit, differential item functioning (DIF), response dependency, unidimensionality, targeting, and the derivation of a raw score to interval measurement transformation.

111

Chapter 5:

Results

112 Chapter 5: Results

5.0 Introduction

The study methodology followed three chronologically distinct stages:

The results of stage III are presented alongside the results of stage I to enable the reader to more readily compare and contrast the results for psychometric analyses of the WEMWBS raw 14-item scale and the Rasch-derived version of the scale more readily.

5.1 Demographic and occupational characteristics of study samples

A summary of sample demographic and occupational characteristics is presented in Table 5.1. These have been reported in full elsewhere (Bartram et al. 2009b, Robertson-Smith et al. 2010). Chapter 5: Results 113

Table 5.1: Summary of sample demographic and occupational characteristics

Survey 1 Survey 2

n=1796 n=8829

SD SD

Male gender 49.9% - 49.7% - Mean age, years 40.8 11.0 45.5 15.9 Qualified in the UK 83.2% - 78.5% - Working in practice 84.0% - 80.8% - Full time employment 82.0% - 64.2% - Fully retired 0.2% - 12.2% -

The sampling frame for Survey 1 included almost exclusively veterinary surgeons who were currently in or seeking employment. By contrast, questionnaires in Survey 2 were sent to all registered veterinary surgeons, including those in retirement. This accounts for the higher mean age (mean difference = 4.7 years, 95% CI = 3.9 to 5.4 years, p < 0.001) of respondents to Survey 2 and the lower proportion in full time employment (difference = -18%, 95% CI = -20% to -16%, p < 0.001).

5.2 Traditional psychometric analyses

The results of traditional psychometric analysis provide evidence in support of the 14-item WEMWBS scale and the Rasch-derived interval scale as reliable and valid measures of overall population-level mental health and well-being among the study sample. The criteria were satisfied for all psychometric properties evaluated. The results of the traditional psychometric analyses for data quality, scaling assumptions, targeting, and reliability are presented in Table 5.2.

Table 5.2: Traditional psychometric methods – data quality, scaling assumptions, targeting, reliability

Dataset from Survey 1 Dataset from Survey 2 Property Ordinal scale Interval scale (7 items) Ordinal scale

Data quality Item missing data (%) 0.0 to 0.5 0.1 to 0.4 7.9 to 8.8 Proportion of cases with no missing data (%) 96.6 (1735/1796) 98.2 (1764/1796) 88.8 (7837/8829) …Computable scale scores (%) 99.9 (1795/1796) 99.7 (1791/1796) 92.0 (8124/8829) …Responder vs. non-responder: gender (chi-square) chi-square=4.9, df=1, p=0.027a - chi-square=143.4, df=1, p<0.001a …Responder vs. non-responder: mean age (t-test) t(1780)=1.5, p=0.128 - t(8587)=31.9, p<0.001b Scaling assumptions Item mean scores: range 2.73 to 3.90 3.09 to 3.90 2.86 to 3.88 Item SD: range 0.73 to 1.01 0.73 to 0.96 0.71 to 0.96 Item variance: range 0.53 to 1.02 0.53 to 0.91 0.50 to 0.92 Corrected item-total correlations (r): range 0.61 to 0.81 0.60 to 0.70 0.55 to 0.79 Targeting Mean score (SD) [95% CI] 48.77 (9.11) [48.34 to 49.20] 22.68 (3.83) [22.50 to 22.86] 49.05 (8.27) [48.87 to 49.23] Possible score range† 14 to 70 7 to 35 14 to 70 Observed score range 14 to 70 7 to 35 14 to 70 Floor/ceiling effect* (%) 0.06 (1/1735) / 0.92 (16/1735) 0.06 (1/1764) / 1.36 (24/1764) 0.08 (6/7837) / 0.60 (47/7837) Skewness -0.29 0.60 -0.32 Reliability Cronbach’s alpha (95% CI) 0.94 (0.93 to 0.94) 0.87 (0.86 to 0.88) 0.92 (0.92 to 0.93) Mean inter-item correlation 0.52 0.50 0.46 Standard error of measurement§ (SEM) 2.30 1.57 2.31

† High scores indicate higher mental well-being * Floor effect = % scoring the lowest possible score (lowest mental well-being); ceiling effect = % scoring the highest possible score (highest mental well- being). Responder vs. non-responder: a A higher proportion of non-responders is male; b The mean age of non-responders is higher. § SD × √(1 – Cronbach’s alpha); ‡ ≥ 11 completed items for the 14-item WEMWBS; ≥ 6 completed items for the 7-item scale. No missing values have been imputed for the purposes of the analyses reported in this table. Values that do not meet the a priori criteria for acceptability are highlighted in blue. n = refer to Table 4.1. Chapter 5: Results 115

5.2.1 Data quality

Data quality was high. WEMWBS scale raw scores were computable for 99% of respondents to Survey 1 and 88% of respondents in Survey 2. The proportion of item-level missing data was low (≤ 5% for Survey 1 and < 9% for Survey 2) (Figure 5.1). However, there was some responder bias: a higher proportion of incomplete or non-responders was male in both surveys and the mean age of incomplete or non- responders was higher than that of complete responders in Survey 2.

The quantity of missing data was higher in Survey 2 because the RCVS emphasised to survey recipients in the questionnaire and covering letter that respondents could omit the section of the questionnaire which included the WEMWBS items if they had any concerns about reporting such information regarding their mental health to their governing professional body. In an attempt to account for this, the missing data descriptive statistics for Survey 2 were adjusted by repeating the computations after selecting only those cases who responded to ≥ 1 WEMWBS item (93% [8178/8829]), i.e. the cases who did not respond to any WEMWBS items were omitted. Following this adjustment, the proportion of cases with no missing data was 96% (7837/8178) and the proportion of cases with computable scale scores (≥11 completed items) was 99% (8124/8178).

Fig. 5.1: Proportion of item-level missing data [a] Survey 1; and [b] Survey 2

[a] 0.55 0.50 0.45 0.40 0.35 0.30 0.25 0.20 Percentage 0.15 0.10 0.05 0.00 Item Item Item Item Item Item Item Item Item Item Item Item Item Item 1 2 3 4 5 6 7 8 9 10 11 12 13 14 n =1796 † [b] 0.90 0.80 0.70 0.60 0.50 0.40

Percentage 0.30 0.20 0.10 0.00 Item Item Item Item Item Item Item Item Item Item Item Item Item Item 1 2 3 4 5 6 7 8 9 10 11 12 13 14 n=7837

† After adjustment for cases who did not respond to any WEMWBS items

116 Chapter 5: Results

5.2.2 Scaling assumptions

The scale passed tests for scaling assumptions. Item response option frequency distributions and item mean scores and standard deviations were roughly similar. Item scores do not need to be standardised before items are summed into an overall score if the item scores have similar variances (Hobart and Cano 2009). The lowest corrected item-total correlation was 0.55, satisfying the recommended criterion of > 0.2 (Cheung et al. 2000). This provides evidence which suggests that items in the scale measured a common underlying construct. The range of corrected item-total correlations for each sample indicated that items in each of the sub-scales contained a similar proportion of information.

5.2.3 Targeting

Scale-to-sample targeting was good. Scale scores spanned the scale range and were not particularly skewed (-0.29 to +0.60). Mean scores were near the scale midpoint, and floor/ceiling effects were negligible. Score distributions are illustrated in Figure 5.2.

The endorsement frequency for two or more adjacent response categories was < 10% (Smith et al. 2005) for the following WEMWBS items in Survey 1: item 2 (category 1+2: 7.2%); item 6 (category 1+2: 6.9%); item 7 (category 1+2: 3.6%); item 11 (category 1+2: 4.7%); item 14 (category 1+2: 9.4%). The endorsement frequencies for all response categories for all items met the criterion for acceptability of < 80% (Cano et al. 2004) (Figure 5.3).

The endorsement frequency for two or more adjacent response categories was < 10% (Smith et al. 2005) for the following WEMWBS items in Survey 2: item 2 (category 1+2: 7.1%); item 4 (category 1+2: 8.0%); item 6 (category 1+2: 6.5%); item 11 (category 1+2: 4.7%); item 14 (category 1+2: 9.4%). The endorsement frequencies for all response categories for all items met the criterion for acceptability of < 80% (Cano et al. 2004) (Figure 5.4).

This low endorsement frequency for response categories 1 and 2 in items 2, 6, 11 and 14 in both surveys suggests that these categories could potentially be collapsed into a single category for the items concerned. It also suggests that consideration should be given to whether the targeting of these items is appropriate for the target population because the level of item difficulty may be too low. Chapter 5: Results 117

Fig. 5.2: Score distributions: [a] WEMWBS raw score, Survey 1; [b] WEMWBS raw score, Survey 2; and [c] Rasch-derived interval scale, Survey 1

[a]

(n=1735, no imputed values)

[b]

(n=7837, no imputed values)

[c]

(n=1764, no imputed values)

118 Chapter 5: Results

5.2.4 Reliability

Internal consistency reliability was high. The range of Cronbach’s alpha across the different samples was 0.87 to 0.94, satisfying the recommended criterion (i.e. within the range of 0.70 to 0.95) (Terwee et al. 2007). The mean inter-item correlation across each of the samples was 0.46 to 0.52. This value for the full scale in Survey 1 exceeded the recommended maximum of 0.5 which was sought (Clark and Watson 1995). This may suggest some item redundancy.

The SEM was considered adequate because the value was well within the criterion for acceptability of < 10% of the total possible score range (van Baalen et al. 2006, van der Zee et al. 2010), i.e. < 5 in the case of the WEMWBS (score range 14 to 70) and < 2 in the case of the Rasch-transformed 7-item scale (score range 7 to 35).

I've been feeling optimistic about the future (n=1795)

Fig. 5.3: Proportion of respondentsI've endorsing been feeling usefuleach (n=1789) WEMWBS item response category – Survey 1

I've been feeling relaxed (n=1788)

I've been feeling interested in other people (n=1788)

I've had energy to spare (n=1787)

I've been dealing with problems well (n=1793)

I've been thinking clearly (n=1790)

I've been feeling good about myself (n=1791)

I've been feeling close to other people (n=1788)

I've been feeling confident (n=1791)

I've been able to make up my own mind about things (n=1789)

I've been feeling loved (n=1790)

I've been interested in new things (n=1793)

I've been feeling cheerful (n=1796)

0% 20% 40% 60% 80% 100%

None of the time Rarely Some of the time Often All of the time

No imputed values

119

I've been feeling optimistic about the future (n=8130)

Fig. 5.4: Proportion of respondentsI've endorsing been feeling usefuleach (n=8124) WEMWBS item response category – Survey 2

I've been feeling relaxed (n=8127)

I've been feeling interested in other people (n=8104)

I've had energy to spare (n=8125)

I've been dealing with problems well (n=8126)

I've been thinking clearly (n=8121)

I've been feeling good about myself (n=8123)

I've been feeling close to other people (n=8106)

I've been feeling confident (n=8113)

I've been able to make up my own mind about things (n=8120)

I've been feeling loved (n=8049)

I've been interested in new things (n=8126)

I've been feeling cheerful (n=8129)

0% 20% 40% 60% 80% 100%

None of the time Rarely Some of the time Often All of the time

No imputed values

Chapter 5: Results 121

5.2.5 Convergent construct validity

Tests of convergent construct validity supported the validity of WEMWBS and the Rasch-derived 7-item interval scale. The direction, magnitude and pattern of associations with comparator scales were consistent with ≥ 75% of a priori hypotheses (Terwee et al. 2007) (Section 4.4.3.2).

WEMWBS would show moderate negative correlations (r = -0.3 to -0.7) with HADS which captures anxiety and depressive symptoms and moderate positive correlations with several HSE MIST psychosocial work characteristics, including the demands sub-scale.

The full WEMWBS and the interval scale showed negative correlations with HADS-A (r = -0.69 and r = -0.66 respectively) and HADS-D (r = -0.76 and r = -0.69 respectively). The correlation between the full WEMWBS and HADS-D was marginally higher than expected. Both versions of the WEMWBS scale were positively correlated with more favourable psychosocial work characteristics (HSE MSIT) in terms of demands, managerial support and control (range: r = 0.32 to 0.48 and r = 0.30 to 0.46 for the full WEMWBS and the interval scale respectively) (Table 5.3).

In multiple linear regression (WEMWBS as the dependent variable), suicidal ideation, high WHI_N,9 and one unit increases in the following scores: HADS-A; HADS-D; HSE MIST demands sub-scale, would each be associated with a decrease in WEMWBS score.

These criteria were fulfilled for both the full WEMWBS scale and the interval scale for suicidal ideation, HADS-A and HADS-D, but not for HSE MIST demands sub- scale or high WHI_N (Table 5.3). For example, after adjustment for age, gender and each of the other scales: reporting having experienced suicidal thoughts in the previous 12 months was associated with a 0.40 unit reduction in the score for the interval scale, (Coef. -0.40, 95% CI: -0.77 to -0.01, p=0.042); and a one unit increase in HADS-D score was associated with a 0.43 unit reduction in the score for the interval scale (Coef. -0.43, 95% CI: -0.48 to -0.38, p<0.001).

In multiple logistic regression (WEMWBS as the independent variable), a one unit increase in WEMWBS score would be associated with decreased

9 High WHI_N was defined as a score of > 1 SD above the mean score on the WHI_N scale, i.e. > 1.76

122 Chapter 5: Results

odds of reporting HADS-A caseness (HADS-A score ≥ 8), HADS-D caseness (HADS-D score ≥ 8), and suicidal ideation.

These criteria were fulfilled for both the full WEMWBS scale and the interval scale (Table 5.3). For example, after adjustment for age, gender and each of the other scales: a one unit increase in the interval scale was associated with a 52% reduction in the odds of reporting HADS-D probable caseness (HADS-D score ≥11) (OR: 0.48, 95% CI: 0.41 to 0.56, p<0.001); and a 13% reduction in the odds of reporting having experienced suicidal thoughts in the previous 12 months (OR: 0.87, 95% CI: 0.81 to 0.92, p<0.001),

Only the results of analyses to test the a priori hypothesised associations that were used to as criteria to assess the convergent construct validity of the WEMWBS and the interval scale are reported here. The full results of the multiple regressions (i.e. associations between the 14-item WEMWBS scale or the Rasch-derived 7-item interval scale and all the other scales used in Survey 1) are reported in Appendix 7.

5.2.6 Discriminant construct validity

The tests of discriminant construct validity for the full WEMWBS and the Rasch- derived 7-item interval scale were consistent with the a priori hypothesis; there was no significant difference in scores across country/region of graduation categories (UK; overseas [EU]; and overseas [non-EU]) (Table 5.3).

5.2.7 Group difference validity

The mean scores for men were significantly higher than the mean scores for women for the full WEMWBS and the Rasch-derived 7-item interval scale, confirming the a priori hypothesised group difference (Table 5.3).

Table 5.3: Traditional psychometric methods – convergent and discriminant construct validity and group difference validity

Dataset from Survey 1 Dataset from Survey 2 Statistical method/Variable Ordinal scale* Interval scale (7 items) Ordinal scale

Convergent construct validity

Pearson correlation coefficient r, (95% CI), p-value HADS-A -0.69 (-0.71 to -0.67), p<0.001 -0.66 (-0.69 to -0.63), p<0.001 - HADS-D -0.76 (-0.78 to -0.74), p<0.001 -0.69 (-0.71 to -0.67), p<0.001 - HSE MSIT demands sub-scale 0.32 (0.28 to 0.36), p<0.001 0.30 (0.26 to 0.34), p<0.001 - HSE MSIT control sub-scale 0.45 (0.41 to 0.49), p<0.001 0.43 (0.39 to 0.48), p<0.001 HSE MSIT managerial support sub-scale 0.48 (0.44 to 0.52), p<0.001 0.46 (0.42 to 0.50), p<0.001 Linear regression† Coef. (95% CI), p-value - HADS-A -0.61 (-0.70 to -0.52), p<0.001 -0.29 (-0.33 to -0.24), p<0.001 - HADS-D -1.22 (-1.33 to -1.11), p<0.001 -0.43 (-0.48 to -0.38), p<0.001 - HSE MSIT demands sub-scale -0.35 (-0.81 to 0.12), p=0.140 -0.14 (-0.36 to 0.08), p=0.218 - HSE MSIT control sub-scale 1.12 (0.66 to 1.58), p<0.001 0.49 (0.27 to 0.71), p<0.001 - HSE MSIT managerial support sub-scale 0.72 (0.23 to 1.21), p=0.004 0.31 (0.08 to 0.54), p=0.010 - Suicidal thoughts in previous 12 months -1.62 (-2.32 to -0.92), p<0.001 -0.40 (-0.77 to -0.01), p=0.042 - WHI_N 0.57 (-0.24 to 1.38), p=0.164 0.33 (-0.14 to 0.78), p=0.164 - Logistic regression‡ OR, (95% CI), p-value HADS-A possible or probable caseness (score ≥ 8) 0.85 (0.83 to 0.87), p<0.001 0.68 (0.64 to 0.71), p<0.001 - HADS-A probable caseness (score ≥ 11) 0.85 (0.83 to 0.87), p<0.001 0.66 (0.62 to 0.70), p<0.001 - HADS-D possible or probable caseness (score ≥ 8) 0.78 (0.76 to 0.81), p<0.001 0.56 (0.52 to 0.61), p<0.001 - HADS-D probable caseness (score ≥ 11) 0.75 (0.71 to 0.79), p<0.001 0.48 (0.41 to 0.56), p<0.001 - Suicidal thoughts in previous 12 months 0.93 (0.90 to 0.96), p<0.001 0.87 (0.81 to 0.92), p<0.001 -

Discriminant construct validity

One-way ANOVA Country/region of graduation F(2, 1786)=0.2, p=0.784 F(2, 1786)=0.3, p=0.751 F(2, 7820)=2.6, p=0.078 - Continued overleaf

123

Dataset from Survey 1 Dataset from Survey 2 Statistical method/Variable Ordinal scale* Interval scale (7 items) Ordinal scale

Group difference validity

Independent samples t-test Gender group differences in mean score Male (mean) (95% CI) n=895, 49.81 (49.22 to 50.40) n=875, 23.10 (22.84 to 23.36) n=3709, 49.94 (49.67 to 50.21) Female (mean) (95% CI) n=895, 47.84 (47.25 to 48.43) n=884, 22.27 (22.02 to 22.52) n=4104, 48.23 (47.98 to 48.48) Mean difference§ (95% CI), p-value -1.97 (-2.81 to -1.13), p<0.001 -0.84 (-1.19 to -0.48) p<0.001 -1.71 (-2.08 to -1.35), p<0.001

* Includes imputed values; † WEMWBS as the dependent variable; ‡ WEMWBS as the independent variable; § Independent samples t-tests, equal variances assumed (Levene’s test p>0.05). Values that are inconsistent with a priori hypotheses are highlighted in blue. Chapter 5: Results 125

5.2.8 Exploratory factor analysis (EFA)

The results of EFA provided evidence to support the unidimensionality of the full WEMWBS and the 7-item interval scale.

The data for the full WEMWBS and the 7-item interval scale were inspected for suitability for principal components analysis (PCA). Each sample met the required criteria: PCA correlation coefficients were > 0.3, the Kaiser-Meyer-Olkin (KMO) Measure of Sampling Adequacy was ≥ 0.6 and Bartlett's Test of Sphericity was statistically significant at the p < 0.05 level (Pallant 2007) (Table 5.4).

For the full WEMWBS, two factors were identified with an eigenvalue greater than 1.0 in both Survey 1 and Survey 2 datasets. However, the results of parallel analysis showed only one component in each dataset with an eigenvalue exceeding the corresponding criterion value for a randomly generated data matrix of the same size. Moreover, the scree plot line tests indicated deviation from the line after one factor only (Figure 5.5). For the Rasch-derived 7-item scale, only one factor was identified with an eigenvalue greater than 1.0 in both Survey 1 and Survey 2 datasets. This finding was supported by the scree plot line tests which indicated deviation from the line after one factor only (Figure 5.6).

Table 5.4: Summary of factor analytic assessments applied to two samples

Dataset from Survey 1 Dataset from Survey 2 Factor analytic technique/statistic Ordinal scale Interval scale (7 items) Ordinal scale Interval scale (7 items) n=1735 n=1764 n=7837 n=7990

Exploratory factor analysis Tests of suitability for PCA Correlation coefficients (range) 0.35 to 0.71 0.38 to 0.65 0.31 to 0.68 0.33 to 0.65 KMO Measure of Sampling Adequacy 0.95 0.90 0.94 0.87 2 2 2 2 Bartlett's Test of Sphericity χ =15494.4, χ =5275.4, χ =59233.4, χ =19844.1, df=91, p<0.001 df=21, p<0.001 df=91, p<0.001 df=21, p<0.001 PCA Factors with eigenvalue >0.1 2 factors: eigenvalues 1 factor: 2 factors: eigenvalues 1 factor: eigenvalue (% of variance explained) 7.77 (55.47%); 1.04 (7.40%) eigenvalue 4.00 (57.20%) 7.06 (50.41%); 1.05 (7.50%) eigenvalue 3.65 (52.12%) Next highest eigenvalue, 0.78 (5.57%) 0.75 (10.79%) 0.90 (6.42%) 0.85 (12.15%) (% of variance explained) Scree plot. Number of factors before 1 1 1 1 deviation of the line Parallel analysis criterion values: 1.15; 1.12 - 1.13; 1.10 - component 1; component 2

Confirmatory factor analysis† χ2 =1600.47, χ2 =274.67, χ2 =7232.85, χ2 =1661.45, Chi-square df = 77, p < 0.001 df = 14, p < 0.001 df = 77, p < 0.001 df = 14, p < 0.001 CFI 0.90 0.95 0.88 0.92 RMSEA (90% confidence interval) 0.10 (0.10 to 0.11) 0.10 (0.09 to 0.11) 0.11 (0.11 to 0.11) 0.12 (0.12 to 0.13) Item loading onto single factor (range) 0.62 to 0.83 0.64 to 0.77 0.56 to 0.81 0.60 to 0.77

†Standardised estimates are presented. CFI: comparative fit index. RMSEA: root mean square error of approximation. Values that do not meet the a priori criteria for acceptability are highlighted in blue. Chapter 5: Results 127

Fig. 5.5: Scree plot for 14-item WEMWBS [a] Survey 1, and [b] Survey 2

[a]

[b]

128 Chapter 5: Results

Fig. 5.6: Scree plot for Rasch-derived 7-item scale [a] Survey 1, and [b] Survey 2

[a]

[b]

Chapter 5: Results 129

5.2.9 Confirmatory factor analysis (CFA)

Confirmatory factor analysis (CFA) was used to test the fit of the hypothesised one- factor scale structure (Table 5.4). The chi-square value was significant for all samples, which is probably attributable to the large sample sizes (Bentler and Bonett 1980, Cole 1987). A comparative fit index (CFI) > 0.9 and a root mean square error of approximation (RMSEA) < 0.08 were the criteria used to indicate acceptable fit (McDonald and Ho 2002). The CFI was < 0.9 for the full WEMWBS in the Survey 2 dataset and the RMSEA was ≥ 1.0 for the full WEMWBS and the Rasch-derived 7-item scale in the datasets from Survey 1 and Survey 2. All items had significant standardised factor loadings which exceeded 0.50 on the single factor in the model.

5.3 Modern psychometric methods

5.3.1 Rasch analysis

5.3.1.1 Selection of Rasch model

The likelihood ratio test provides a test of which of the two versions of the Rasch model should be used by comparing the different parameterisations of the model and providing a chi-square statistic and probability. The outcome of the test was significant (chi-square=600.37, df =38, p < 0.001), indicating that the distance between response thresholds was inconsistent and that the partial credit version of the Rasch model should be adopted for the analyses.

5.3.1.2 Class interval structure

There were between 51 and 66 persons in each of the nine class intervals.

5.3.1.3 Response option thresholds ordering

Threshold ordering (i.e. the transition between adjacent response categories) for each item was assessed by comparing the logit ability rating for each response category using each item’s threshold category probability curve. All item response option thresholds were found to be ordered (Figure 5.7, Figure 5.8). That is, within each item, the transition from one category to the next represents an increase in the underlying trait of mental well-being.

130 Chapter 5: Results

Fig. 5.7: Response option thresholds ordering for item 9

Category probability curves for item 9 ‘I’ve been feeling close to other people’ showing ordered response category thresholds. Each curve represents a response category. The x- axis symbolises the construct, with the level of mental well-being increasing to the right. The y-axis shows the probability of endorsing the response categories at different levels of mental well-being: 0 ‘none of the time’, 1 ‘rarely’, 2 ‘some of the time’, 3 ‘often’, and 4 ‘all of the time’. It can be seen that the responses to this item fall in a logical progressive order – category responses represent an increase in mental well-being (logits) for each category. For example, at the lowest level of mental well-being (-6 logits) the most probable score is 0 ‘none of the time’, and at the highest level of mental well-being (+5 logits) the most probable score is 4 ‘all of the time’.

Fig. 5.8: Threshold map for the 14-item WEMWBS

Items are listed in order of increasing item number (item 1 to item 14) on the y-axis. Person location (logits) is on the x-axis. Chapter 5: Results 131

5.3.1.4 Individual item fit

The following three indicators were examined: (1) fit residuals (item-person interaction), (2) chi-square probability values (item-trait interaction), and (3) item characteristic curves (ICCs).

Seven of the 14 items had fit residuals outside the recommended range of ± 2.5 (if the data fit the Rasch model, the mean value is expected to be approximately zero). High positive fit residual values indicate misfit to model expectations and high negative fit residuals indicate item redundancy. Three items with high fit residuals and significant chi-square p-values were removed from the scale: item 8, ‘I’ve been feeling good about myself’; item 12, ‘I’ve been feeling loved’; and item 14, ‘I’ve been feeling cheerful’ (Table 5.5).

Table 5.5: Rasch fit statistics for each of the 14 items

Location Item SE Fit residualChi-square p-value (logits)

1 0.39 0.03 2.99 21.22 0.006579 2 0.09 0.04 2.25 16.95 0.030633 3 0.99 0.03 1.36 17.27 0.027462 4 -0.27 0.04 1.98 12.61 0.126131 5 1.44 0.03 5.81 50.48 0.000000 6 -0.01 0.04 -2.84 30.90 0.000146 7 -0.78 0.04 -2.12 14.57 0.068094 8 0.15 0.04 -10.59 117.82 0.000000 9 0.00 0.04 -0.68 10.88 0.208825 10 0.14 0.04 -6.22 41.37 0.000002 11 -1.25 0.04 -0.03 25.37 0.001346 12 -0.32 0.03 7.94 78.24 0.000000 13 -0.34 0.03 1.74 21.97 0.004977 14 -0.04 0.04 -9.61 98.51 0.000000

Reliability (PSI) = 0.92; SE = standard error. Bonferroni adjustment to the alpha level for 14 items p = 0.003571. Bold signifies misfit to the Rasch model.

The item characteristic curve (ICC) is a graphical indicator of item fit. The ICCs for the worst fitting (item 8 ‘I’ve been feeling good about myself’ and item 12 ‘I’ve been feeling loved’) and best fitting (item 9 ‘I’ve been feeling close to other people’) items are illustrated in Figure 5.9.

132 Chapter 5: Results

Fig. 5.9: Item characteristic curves for: [a] items 8, [b] item 12, and [c] item 9

[a]

[b]

[c]

The ICC is a graph for an individual item. It plots the expected response (predicted from the model) to an item at each level of the measurement continuum. The available responses for WEMWBS items are: 0 ‘none of the time’, 1 ‘rarely’, 2 ‘some of the time’, 3 ‘often’, and 4 ‘all of the time’. [a] The fit residual for item 8 is negative, indicating an over-discriminating item Chapter 5: Results 133

relative to the model. The dots represent the observed scores for the class intervals at the different ability levels. The chi-square probability (p<0.001, displayed in the line above the graph) indicates a statistically significant difference between expected and observed scores for this item. The class interval dots tend to be under the line at the left-hand end and above the line at the right-hand end (i.e. the observed scores form a steeper curve than the expected scores). Therefore, among people with higher levels of mental well-being, more than predicted report feeling good about themselves ‘often’ or ‘all of the time’, and among people with lower levels of mental well-being, fewer than predicted report feeling good about themselves ‘none of the time’ or ‘rarely’. [b] By contrast, the fit residual for item 12 is positive, indicating an under-discriminating item relative to the model. [c] The fit residual for item 9 is almost zero, indicating good fit to the model. The chi-square probability (p=0.209) indicates that the difference between expected and observed scores for this item is not statistically significant.

5.3.1.5 Individual person fit

Examination of individual person fit statistics showed that 14.4% (284/1971) of participants had fit residuals outside the acceptable range of ± 2.5. The range of individual person fit residuals was -5.88 to +4.95. A high negative person fit residual is indicative of fitting the model ‘too well’, with a response pattern very close to the ideal Guttman pattern (Figure 5.10). This is generally regarded as being less problematic to the analyses than a high positive person fit residual (Muller and Roddy 2009). Examination of the response patterns of individuals with high positive fit residuals identified several who had affirmed response categories which indicated opposite extremes of the continuum of mental well-being (Figure 5.10). However, no respondents were removed from the analyses.

134 Chapter 5: Results

Fig. 5.10: Item scoring for respondents with [a] a high negative person fit residual, and [b] a high positive person fit residual

[a]

[b]

¶ marks the item response categories affirmed by the respondent.

[a] The item response categories affirmed by this respondent (case 454) coincide exactly with those that the model would expect them to affirm at a certain level of mental well-being (-0.52 logits). Consequently there is a high negative person fit residual (-5.88). [b] This respondent (case 839) affirmed item response categories which indicate opposite extremes of mental well-being: the lowest response categories were affirmed for items 1 to 3 and the highest response categories for items 4 to 14. Consequently there is a high positive person fit residual (+4.94). Chapter 5: Results 135

5.3.1.6 Differential item functioning (DIF)

Several items showed uniform DIF for gender (Table 5.6; Figure 5.11). No items showed non-uniform DIF for gender, i.e. the difference between genders was consistent across the different levels of mental well-being (class intervals).

Table 5.6: Differential item functioning for gender

Main effect for gender Gender-trait interaction

Item (Uniform DIF) (Non-uniform DIF) MS F df p-value MS F df p-value

1 30.95 30.75 1 < 0.001 0.95 0.95 8 0.478 2 2.71 2.70 1 0.101 0.73 0.73 8 0.671 3 15.99 16.67 1 < 0.001 0.58 0.61 8 0.771 4 28.67 29.29 1 < 0.001 1.03 1.05 8 0.396 5 0.71 0.64 1 0.422 1.10 1.00 8 0.436 6 1.75 2.12 1 0.145 0.40 0.48 8 0.869 7 1.60 1.87 1 0.172 0.81 0.94 8 0.480 8 12.78 21.96 1 < 0.001 -0.27 -0.47 8 1.000 9 5.46 6.05 1 0.014 0.96 1.06 8 0.389 10 41.83 59.28 1 < 0.001 -0.12 -0.17 8 1.000 11 25.95 28.67 1 < 0.001 0.23 0.26 8 0.979 12 38.93 31.81 1 < 0.001 -0.36 -0.30 8 1.000 13 0.00 0.001 1 0.983 0.87 0.88 8 0.531 14 0.00 0.00 1 0.983 0.50 0.82 8 0.585

MS = mean square; F = F-ratio; df = degrees of freedom. Bonferroni adjustment to the alpha level p = 0.001 [Given that there are 14 items, and that for each item there are three probabilities – one for each of two main effects (a class interval effect and a person factor [gender] effect) and a person factor by class interval interaction – the criterion level for misfit for each statistic was 0.05/42 = 0.001]. MS is the chi-square statistic divided by the degrees of freedom. MS statistics have expectation 1.0, and range from 0 to infinity. MS > 1.0 indicates underfit to the Rasch model, i.e. the data are less predictable than the model expects; MS < 1.0 indicates overfit to the Rasch model, i.e. the data are more predictable than the model expects (Wright and Linacre 1994). Bold signifies misfit to the Rasch model.

136 Chapter 5: Results

Fig. 5.11: Item characteristic curves by gender for [a] item 7, and [b] item 10

[a]

[b]

[a] The item characteristic curves for males and females for item 7 are almost superimposed on each other. Males and females at the same level of mental well-being affirm the same response category, i.e. there is no differential item functioning for gender. [b] By contrast, the item characteristic curves for males and females for item 10 are significantly different. Females affirm a lower response category than males at the same level of mental well-being, i.e. there is differential item functioning for gender. The lines for males and females do not cross, indicating that the difference between genders is consistent across the different levels of mental well-being (the lines would cross if non-uniform DIF was apparent). Chapter 5: Results 137

Several items showed uniform DIF for age group (Table 5.7, Figure 5.12). No items showed non-uniform DIF for age group, i.e. the differences between age groups were consistent across the different levels of mental well-being (class intervals).

Table 5.7: Differential item functioning for age group

Main effect for age group Age group-trait interaction Item (Uniform DIF) (Non-uniform DIF) MS F df p-value MS F df p-value

1 32.53 33.34 3 < 0.001 0.66 0.67 24 0.883 2 7.19 7.28 3 < 0.001 1.19 1.21 24 0.223 3 27.57 29.93 3 < 0.001 1.12 1.23 24 0.216 4 2.49 2.52 3 0.056 1.23 1.25 24 0.187 5 1.01 0.92 3 0.432 1.53 1.39 24 0.099 6 0.28 0.34 3 0.796 1.12 1.36 24 0.113 7 0.69 0.81 3 0.489 0.85 0.99 24 0.475 8 1.45 2.47 3 0.060 0.50 0.85 24 0.668 9 1.79 1.97 3 0.116 0.70 0.78 24 0.770 10 5.93 8.27 3 < 0.001 0.72 1.01 24 0.452 11 11.77 13.17 3 < 0.001 1.38 1.55 24 0.044 12 15.42 12.66 3 < 0.001 0.87 0.71 24 0.846 13 0.92 0.93 3 0.423 0.87 0.88 24 0.630 14 3.23 5.32 3 0.001 0.46 0.76 24 0.790

MS = mean square; F = F-ratio; df = degrees of freedom. Bonferroni adjustment to the alpha level p = 0.001 (see legend to Fig. 5.6 for calculation). Bold signifies misfit to the Rasch model.

138 Chapter 5: Results

Fig. 5.12: Item characteristic curves by age group for [a] item 1, and [b] item 6

[a]

[b]

[a] The item characteristic curves for the different age group quartiles for item 1 are significantly different. Respondents ≥ 55 years of age affirm a lower response category than younger respondents at the same level of mental well-being, i.e. there is differential item functioning for age group. The lines for the different age groups do not cross, indicating that the difference between age groups is consistent across the different levels of mental well- being (lines would cross if non-uniform DIF was apparent). [b] By contrast, the item characteristic curves for the different age groups for item 6 are not significantly different. Different age groups at the same level of mental well-being affirm the same response category, i.e. there is no differential item functioning for age group. Chapter 5: Results 139

A degree of DIF cancellation can occur at the summed scale level when certain items in the scale have DIF of similar magnitude and opposite direction for a specific variable, thereby reducing the overall impact (Teresi and Fleishman 2007). Item 1 and item 3 displayed DIF for both gender and age group. These two items were grouped into a subtest to investigate whether the DIF cancelled out at the level of the summed scale. The DIF for both gender and age group cancelled and consequently both items were retained in the scale. (Figures 5.13 and 5.14 respectively).

Item 11 also displayed DIF for gender and age group. When item 11 was added to the subtest described in above (i.e. items 1, 3 and 11 combined), the DIF cancelled for both gender (MS=5.08, F=6.48, df=1, p=0.011) and age group (MS=2.33, F=2.97, df=3, p=0.031) [Bonferroni adjusted alpha level p=0.001].10 Consequently item 11 was also retained in the scale.

It is important to note that, even if DIF cancellation occurs at the level of the summed scale, the scores of specific individuals may still be affected by DIF. However, the purpose of the psychometric assessments in this research is to evaluate the potential utility of the scale for measurement at a population level. This involves the use of aggregate scores, so the possibility of DIF remaining at a person level was not a concern.

10 Given that there are 12 items (because items 1, 3 and 11 have been combined into a single item for the purpose of examining DIF cancellation) and that for each item there are three probabilities – one for each of two main effects (a class interval effect and a person factor [gender] effect) and a person factor by class interval interaction – the criterion level for misfit was 0.05/36 = 0.001.

140 Chapter 5: Results

Fig. 5.13: Item characteristic curves by gender for [a] item 1, [b] item 3, and [c] items 1 and 3 combined as a subset

[a]

[b]

[c]

Item 1 and item 3 showed uniform DIF for gender of similar magnitude and opposite direction. For example: [a] At a person location of +1.14 logits (class interval 6), the mean expected response values for item 1 were 2.54 for females and 2.35 for males (i.e. females Chapter 5: Results 141

affirm a higher response category than males at the same level of mental well-being). [b] By contrast, at the same person location, the mean expected response values for item 3 were 1.92 for females and 2.08 for males (i.e. males affirm a higher response category than females at the same level of mental well-being). [c] When item 1 and item 3 were combined into a subtest, the DIF cancelled, indicating that their item-level DIF for gender should not be problematic at the level of the summed scale (Main effect for gender, item 1 + item 3: MS=1.32, F=1.35, df=1, p=0.245) [Bonferroni adjusted alpha level p=0.001].

142 Chapter 5: Results

Fig. 5.14: Item characteristic curves by age group for [a] item 1, [b] item 3, and [c] items 1 and 3 combined as a subset

[a]

[b]

[c]

Item 1 and item 3 showed uniform DIF for age group of similar magnitude and opposite direction. For example: [a] At a person location of +1.14 logits (class interval 6), the mean expected response values for item 1 were 2.68 for the ‘under 33’ age group and 2.14 for the Chapter 5: Results 143

‘55 plus’ age group (i.e. the ’55 plus’ age group affirms a lower response category than the ‘under 33’ age group at the same level of mental well-being). [b] By contrast, at the same person location, the mean expected response values for item 3 were 1.76 for the ‘under 33’ age group and 2.33 for the ‘55 plus’ age group (i.e. the ’55 plus’ age group affirms a higher response category than the ‘under 33’ age group at the same level of mental well-being). [c] When item 1 and item 3 were combined into a subtest, the DIF cancelled, indicating that their item-level DIF for age group should not be problematic at the level of the summed scale (Main effect for age group, item 1 + item 3: MS=1.71, F=1.76, df=3, p=0.153) [Bonferroni adjusted alpha level p=0.001].

5.3.1.7 Response dependency

Response dependency between items 6 and 7 of the WEMWBS was identified by inspection of the residual correlation matrix (residual correlation +0.312). No item pairs within the Rasch-derived 7-item scale had a residual correlation above the threshold value for acceptability of 0.3.

5.3.1.8 Fit of item sets to the Rasch model

Initial fit of the WEMWBS to model expectations was poor (Analysis 1, Table 5.8). Item 8 ‘I’ve been feeling good about myself’, item 10 ‘I’ve been feeling confident’ and item 12 ‘I’ve been feeling loved’ showed significant individual item misfit (Table 5.5) and were deleted. This led to a marginal improvement in fit (Analysis 2, Table 5.8). Item 14 ‘I’ve been feeling cheerful’ was also deleted due to significant individual item misfit (Table 5.5). This led to a further marginal improvement in fit (Analysis 3, Table 5.8). Item 4 ‘I’ve been feeling interested in other people’ was deleted due to significant DIF for gender (Table 5.6) and Item 5 ‘I’ve had energy to spare’ and item 13 ‘I’ve been interested in new things’ were deleted due to individual item misfit (Table 5.7). A series of t-tests performed on the person estimates from the two most different subsets of items identified from the factor loadings on the first principal component of the residuals (Table 5.9) estimated that only 5.6 % (95% CI: 4.5 to 6.5%) of cases had statistically significant p-values, which supported the scale’s unidimensionality (Analysis 4, Table 5.8). This Rasch-derived 7-item scale, comprising items 1, 2, 3, 6, 7, 9 and 11 only, is identical to the Rasch model-fitting Short Warwick-Edinburgh Mental Well-being Scale (SWEMWBS) reported by Stewart-Brown et al. (2009). As might be expected with a shorter scale, the person separation index (PSI), which provides an indication of the ability of the measure to discriminate among respondents with different levels of the latent variable being measured, had fallen from 0.918 (Analysis 1) to 0.832 (Analysis 4) but remained

144 Chapter 5: Results

acceptable for group-level measurement. Rasch fit statistics for the seven items are presented in Table 5.10.

Given the 7-item scale’s satisfactory fit to the Rasch model and the evidence supporting its unidimensionality, the robustness of the solution (Analysis 4) was tested on data subsets S2_500b to S2_500d and S1_500 for the purpose of cross- validation (Analyses 5 to 8 respectively, Table 5.8). These data subsets also showed good fit to model expectations.

Table 5.8: Summary of model fit statistics for original scale, successive revisions and cross-validation

Item fit residuals Person fit residuals Overall model fit§ Unidimensionality Sample/ Analysis PSI t-tests %† notes/action number Mean SD Mean SD Chi-square df p-value (95% CI) S2_500a 11.7 1 -0.574 5.341 -0.540 1.644 558.2 112 < 0.001 0.918 Full 14-item scale (10.7 to 12.6) S2_500a 9.2 2 -0.551 3.436 -0.532 1.504 241.6 88 < 0.001 0.892 Delete items: 8, 10, 12 (8.3 to 10.3) S2_500a 8.6 3 -0.335 1.795 -0.507 1.433 161.9 80 < 0.001 0.876 Delete item: 14 (7.7 to 9.6) S2_500a 5.6 4 -0.570 2.326 -0.520 1.295 58.8 56 0.104 0.832 Delete items: 4, 5, 13 (4.5 to 6.5) S2_500b 4.6 5 -0.065 1.923 -0.476 1.222 57.4 56 0.068 0.833 Analysis 4 repeated (3.4 to 5.5) S2_500c 5.9 6 0.112 1.473 -0.389 1.933 56.1 56 0.041 0.894 Analysis 4 repeated (4.9 to 7.0) S2_500d 4.7 7 0.511 2.852 -0.287 1.342 62.8 56 0.119 0.845 Analysis 4 repeated (3.7 to 5.6) S1_500 5.4 8 -0.054 2.777 -0.516 1.459 71.9 56 0.235 0.832 Analysis 4 repeated (4.3 to 6.4)

SD= standard deviation; PSI= person separation index; df= degrees of freedom; § overall item-trait interaction; † % of statistically significant t-tests. Bold signifies misfit to the Rasch model. Bonferroni adjustment not applied, p= 0.05 (it is more conservative to not apply Bonferroni adjustment for summary of model fit statistics, thereby reducing the chance of Type I error, i.e. incorrectly declaring fit to the Rasch model). 145

146 Chapter 5: Results

Table 5.9: Principal component analysis of Rasch item residuals for the derived 7-item scale

Positively correlated items Negatively correlated items

Item 7: ‘Think clearly’ 0.718 Item 2: ‘Feeling useful’ -0.382 Item 6: ‘Deal with problems well’ 0.635 Item 1: ‘Feeling optimistic’ -0.681 Item 11: ‘Make up own mind’ 0.515

Only items with correlation > 0.3 are listed. Dimensionality was tested using the most robust approach currently available; testing for differences in individual person estimates between two sets of items loading on the first residual principal component. It has been suggested that the absence of any discernable patterning in the principal components analysis of residuals indicates scale unidimensionality. As shown in the table above, the factor loadings on the first principal component of the residuals indicated three positively correlated items and two negatively correlated items. A clear pattern was identified in the residuals – the two negatively loaded items tended to relate to the cognitive-evaluative dimension of mental well-being and the three positively loaded items tended to relate the psychological functioning dimension of mental well-being. The difference in individual person estimates (i.e. level of mental well- being level) between the two sets of items was statistically tested. Despite the observed pattern, the difference between the estimates generated by the two sets of items did not reach statistical significance, thus providing evidence to support scale unidimensionality.

Table 5.10: Rasch fit statistics for the 7 retained items

Location Item SE Fit residualChi-square p-value (logits)

1 0.48 0.03 0.27 7.00 0.536 2 0.00 0.04 -0.79 10.86 0.210 3 1.06 0.03 1.74 14.54 0.069 6 0.08 0.04 2.34 19.31 0.010 7 -0.66 0.04 2.15 18.78 0.046 9 0.11 0.03 2.47 11.48 0.176 11 -1.07 0.04 -0.70 20.28 0.009

Reliability (PSI) = 0.832; SE = standard error; Bonferroni adjustment to the alpha level for the number of items p = 0.007. The location value for each item (i.e. item difficulty) is the logit – the log-odds of responding to the item associated with that descriptor. By convention, in Rasch analysis the item difficulty scale is centred on zero logits. The fit residuals indicate the degree of divergence between the expected value and the actual data value for each person-item. The difference provides the residual score which approximates a z- standardised normal distribution where a mean of 0 and a standard deviation of 1 would be Chapter 5: Results 147

expected. The cut-off for an acceptable residual is generally set at ± 2.5. Residuals outside this range indicate a source of misfit to the model. The chi-square statistic compares the difference between the observed values with expected values across class intervals for each item. A non-significant Bonferroni adjusted p-value indicates good fit to the model.

5.3.1.9 Targeting of the derived 7-item scale

Since the level of mental well-being in respondents and item responses can be placed on the same logit scale, it is possible to test how well the final set of items and their response options are targeted, i.e. how well the level of mental well-being expressed by the items and their response options is matched to the level of mental well-being in respondents (Figures 5.11 and 5.12).

Fig. 5.15: Person-item threshold distribution for the derived 7-item scale

The figure shows the targeting between the distribution of person ability (top histogram) and the distribution of item threshold locations (inverted histogram). There was an acceptable breadth of spread of item thresholds (-4.614 to +4.700) across the range of person estimates (-6.109 to +6.376) for the sample, with adequate coverage across the measurement continuum. There were no item thresholds at the extremes of the highest and lowest levels of mental well-being (i.e. items do not discriminate at these levels). However, floor or ceiling effects were minimal. There was minimal stacking of item thresholds, which provided evidence to support that all the items contributed to the explanatory power of the scale (low item redundancy). The mean person score of +1.15 logits (SD 1.56) indicated that, on average, the mean level of mental well-being of the sample was slightly higher than the mean level of mental well-being represented by the item set of 0 logits (SD 0.70) (by convention, the mean item location is centred on zero logits in the Rasch model).

148 Chapter 5: Results

Fig. 5.16: Item map for the derived 7-item scale

The item map is an alternative mechanism to display the person-item targeting distribution. Each item is displayed along the location continuum. The location estimate for an item is the mean of the set of uncentralised thresholds for that item. The item map provides an opportunity to examine the meaning of the variable as a construct in terms of item content in association with the item locations. Items 1, 6 and 9, and items 2 and 7 are stacked because they measure similar levels of mental well-being (see Table 5.10 for item locations).

5.3.1.10 Transformation from raw score to interval measurement

In order to maximise the stability (invariance) of the raw score to interval measurement estimates, the transformation was computed using a dataset comprising all respondents from Surveys 1 and Survey 2 combined, with no missing item responses to the derived 7-item scale (n=9754). The summed raw scores were mapped against the corresponding estimated Rasch-transformed logit scores for the set of fitting items. Figure 5.13 shows this transformation graphically for the derived 7-item scale. It is evident that the transformation from raw scores (vertical axis) to interval measurement on the continuum (horizontal axis) is not linear over the whole range – the curve is slightly S-shaped. To calibrate an interval-level scale, logit measures for the fitting items were rescaled into the score range of the ordinal scale (7 to 35) (Table 5.11). The change in measurement implied by a one unit change in Chapter 5: Results 149

raw score varied 3.4-fold across the range of the scale, from 0.29 logits (raw score 16 to 17) to 0.99 logits (raw score 1 to 2).

Fig. 5.17: Scale characteristic curve for the derived 7-item scale

Table 5.11: Raw score to interval measure conversion table for the Rasch-derived 7- item scale

Logit Transformed Logit Transformed Raw Raw measure logit measure measure logit measure score score (interval) (interval) (interval) (interval)

7 -6.11 7.00 22 0.11 20.94 8 -5.12 9.22 23 0.44 21.68 9 -4.38 10.87 24 0.78 22.45 10 -3.83 12.10 25 1.14 23.26 11 -3.39 13.11 26 1.52 24.12 12 -3.00 13.96 27 1.93 25.04 13 -2.66 14.73 28 2.37 26.01 14 -2.35 15.44 29 2.82 27.02 15 -2.04 16.12 30 3.27 28.04 16 -1.74 16.79 31 3.73 29.07 17 -1.45 17.45 32 4.22 30.16 18 -1.15 18.12 33 4.77 31.39 19 -0.84 18.81 34 5.47 32.97 20 -0.53 19.50 35 6.38 35.00 21 -0.22 20.21

This nomogram enables the translation of raw summed scores on the SWEMWBS to interval measures rescaled into the same score range.

150 Chapter 5: Results

5.3.1.11 Correlations between WEMWBS and the derived 7-item scale scores

The Pearson correlation between the raw scores on WEMWBS (n=1735) and the derived 7-item scale (n=1764) was 0.970 (95% CI: 0.968 to 0.972, p<0.001) (Survey 1 dataset). The Pearson correlation between the raw score of the derived 7-item scale (n=1764) and its Rasch-transformed interval scale was 0.984 (95% CI: 0.983 to 0.985, p<0.001) (Survey 1 dataset).

The Pearson correlation between the raw scores on WEMWBS (n=7837) and the derived 7-item scale (n=7990) was 0.962 (95% CI: 0.961 to 0.962, p<0.001) (Survey 2 dataset). The Pearson correlation between the raw score of the derived 7-item scale (n=7990) and its Rasch-transformed interval scale was 0.991 (95% CI: 0.991 to 0.991, p<0.001) (Survey 2 dataset).

To compare the raw score to interval measure transformation for the 7-item scale derived from this research with that derived for a UK general population sample (Stewart-Brown et al. 2009), the Pearson correlation was computed after transforming the raw scores accordingly in the Survey 2 dataset (n=7990). The correlation was 0.999 (95% CI: 0.999 to 1.000, p<0.001).

5.4 Summary

Traditional psychometric methods provided evidence to support the psychometric properties of data quality, scaling assumptions, targeting, reliability and validity:

Data quality – Missing data were low for all items; WEMWBS scale scores were computable for 99% of respondents to Survey 1 and 88% (or 96% after adjustment) of respondents to Survey 2.

Scaling assumptions – Item response option frequency distributions and item mean scores and SDs were roughly similar. Corrected item total correlations were ≥ 0.55, suggesting that the items measured a common underlying construct.

Targeting – Scale scores spanned the scale range and were not particularly skewed. Mean scores were near the mid-point, and floor/ceiling effects were negligible. An endorsement frequency of < 10% for response categories 1 and 2 combined in items 2, 6, 11 and 14 across both surveys suggests that these categories could potentially be collapsed into a single category for the items concerned. It also suggests that consideration should be given to Chapter 5: Results 151

whether the targeting of these items is appropriate for the target population because the level of item difficulty may be too low.

Reliability – Internal consistency reliability was high (Cronbach’s alpha: 0.87 to 0.94). Mean inter-item correlations suggest possible item redundancy. SEM was within the criterion for acceptability.

Validity – The direction, magnitude and pattern of associations with comparator scales measuring theoretically similar constructs were generally consistent with a priori hypotheses (convergent construct validity). The associations with HADS and suicidal ideation support the use of the scale as an overall indicator of mental health and well-being. Associations were retained when these inferential analyses were repeated using the Rasch- derived 7-item interval scale. The scales were able to discriminate between gender groups which were predicted to have differences in the level of mental well-being based on previous surveys of psychiatric morbidity in the general population (group difference validity). As predicted, there was no significant difference in scores across country/regions of graduation (discriminant construct validity). Factor analytic assessments of internal construct validity indicated a single underlying construct (unidimensionality).

Application of Rasch analysis provided evidence of:

Response option thresholds ordering – Response option thresholds were found to be ordered, i.e. within each item, the transition from one option to the next represents an increase in the underlying trait of mental well-being.

Individual item fit – Several items over-discriminated relative to the model. E.g. item 8: among people with higher levels of mental well-being, more than predicted reported feeling good about themselves ‘often’ or ‘all of the time’, and, among people with lower levels of mental well-being, fewer than predicted report feeling good about themselves ‘none of the time’ or ‘rarely’. Several items under-discriminated relative to the model, e.g. item 12.

Differential item functioning (DIF) – Several items showed uniform bias for gender (e.g. item 10) or age group (e.g. item 1) or both gender and age group (e.g. item 11). DIF cancellation occurred at the level of the summed scale between items 1, 3 and 11 for both gender and age group.

152 Chapter 5: Results

Response dependency – The residual correlation between items 6 and 7 was +0.312 (above the criterion for acceptability of 0.3) which suggests that, to an extent, the response to one item determines the response to the other for reasons other than the magnitude of the latent variable (mental well- being).

Fit of item sets to the Rasch model – Initial fit of the WEMWBS to model expectations was poor. After deletion of items for individual item misfit and/or DIF, a 7-item scale was derived (comprising items 1, 2, 3, 6, 7, 9 and 11) which fitted model expectations. The robustness of the solution was cross- validated on four data subsets. Raw scores were transformed to interval measures in the score range of the ordinal scale for the fitting items.

Targeting – The Rasch-derived 7-item scale showed an acceptable spread of item thresholds across the range of person estimates. Floor and ceiling effects were minimal. The mean person score of +1.15 logits (SD: 1.56) indicated that, on average, the mean level of mental well-being of the sample was slightly higher than the mean level of mental well-being represented by the item set (0 logits, SD: 0.70).

Person separation index (PSI) – The Rasch-derived 7-item scale has the ability to discriminate between two subgroups of respondents (PSI: 0.832 to 0.894). 153

Chapter 6:

Discussion

154 Chapter 6: Discussion

6.0 Introduction

This research aimed to examine the measurement properties of WEMWBS in the context of the veterinary profession to inform judgements concerning: [i] the defensibility of comparing WEMWBS scores for the veterinary profession with those for other groups, and [ii] the potential utility of the instrument for future research, including, for example, the monitoring of clinically meaningful trends in mental health and well-being in this occupational group over time.

Specifically, the objectives were to:

[a] Explore the properties of WEMWBS using traditional psychometric methods based on classical test theory (CTT).

[c] Re-examine the external construct validity of WEMWBS using the interval- level measurements derived from the Rasch analysis.

This chapter is structured around these study aims and objectives. It includes critical assessment of the strength of evidence, derived from both traditional and modern methods, in support of the acceptability of the psychometric properties of the instrument in the UK veterinary profession. The findings from traditional and modern psychometric paradigms are compared and contrasted. Gaps that remain in the evidence base are identified. The strengths and limitations of the study are considered. The chapter concludes with a discussion of the implications for research and the profession of both the psychometric analyses and the literature review of potential influences on the elevated suicide risk. Proposals for interventions which have potential to improve psychological health and attenuate suicide risk are included in the discussion of the implications of the study for the profession.

6.1 Psychometric properties of WEMWBS: traditional psychometric methods

Evidence in support of the psychometric properties of WEMWBS and the Rasch- derived 7-item interval scale was provided through traditional assessments of the psychometric properties of: data quality; scaling assumptions; targeting; internal consistency reliability; internal construct validity and external construct validity. The Chapter 6: Discussion 155

findings are broadly consistent with those reported for UK student and general population samples (Tennant et al. 2007).

WEMWBS raw scores were computable for 99% of respondents to Survey 1 and 88% of respondents to Survey 2. A probable explanation for the difference is that respondents to Survey 1, in which the questionnaire was focused on mental health, were inevitably more willing to volunteer to complete the WEMWBS items than respondents to Survey 2, in which the questionnaire was mainly employment-related and reported the information at a population level to their governing professional body. The proportion of respondents who did not complete any WEMWBS items in Survey 2 (18%) is similar to that reported for a UK general population sample (16%) (Tennant et al. 2007).

At the scale level, WEMWBS and the Rasch-derived 7-item scale met most accepted traditional psychometric criteria. Neither scale demonstrated floor or ceiling effects in either of the study populations, indicating that they may have potential for documenting overall improvements in population mental well-being. By contrast, several measures of mental health in non-clinical populations show floor or ceiling effects, with most respondents endorsing the item response categories which indicate the lowest level of symptom severity. For example, HADS-D demonstrated a floor effect in a sample of the veterinary profession (Sample 1 in the present study) (Figure 6.1) (Bartram 2009a).

Fig. 6.1: HADS-D score floor effect in Sample 1

n=1721. The floor effect (percent at minimum score, 0) represents a subsample for whom improvement in psychological health cannot not register as a change in score.

156 Chapter 6: Discussion

Close examination of the item level findings for WEMWBS revealed suboptimal scale-to-sample targeting for some items. The low endorsement frequency for response categories 1 and 2 in items 2, 6, 11 and 14 in both surveys suggests that the item difficulty may be too low for the population concerned (i.e. the response categories for these items are affirmed only at very low levels of mental well-being). Moreover, the high Cronbach’s alpha values (up to 0.94) for the full WEMWBS scale suggest some item redundancy.

Exploratory factor analysis (EFA) with parallel analysis for both WEMWBS and the Rasch-derived 7-item scale indicated a one-factor solution, suggesting that they measure a single underlying construct (unidimensionality). However, confirmatory factor analysis (CFA) did not support this model for the WEMWBS in Survey 2 and was borderline for the Rasch-derived 7-item scale in the same sample. Tests of convergent construct validity supported the validity of both WEMWBS and the Rasch-derived 7-item scale and the utility of the scales as an overall indicator of mental health and well-being.

6.2 Psychometric properties of WEMWBS: Rasch analysis

Initial estimation of model fit suggested that WEMWBS data showed deviation from Rasch model expectations and, therefore, cannot be assumed to be compatible with interval-level scaling when used in this population. On the basis of item-fit statistics and examination for differential item functioning (DIF), misfitting items were removed sequentially to derive a 7-item scale which satisfied the criteria for Rasch model fit.

There was little indication from the assessments of data quality that the 14-item WEMWBS was considered to be overly long when applied to the veterinary profession; thus at the outset of the Rasch analysis, there was no explicit intention to derive a new measure for the sake of shortening the instrument to minimise respondent burden. Removing items from a scale should be carefully considered because content validity may be compromised (Bohlig et al. 1998) and evidence for the external validity of an existing published scale is likely to have included all items, so the revised scale would need to be re-validated (Lundgren-Nilsson and Tennant 2011). Moreover, protecting the integrity of the item set would facilitate comparisons across studies in which the full scale was administered. However, brevity of assessments is often valued by respondents and, provided adequate reliability is preserved, removing items to improve measurement properties is considered to be a desirable goal (Heinemann and Deutsch 2011). Chapter 6: Discussion 157

The Rasch-derived 7-item scale had minimal item bias (differential item functioning, DIF), minimal response dependency, and its polytomous response structure worked as intended, with higher scores within an item reflecting greater overall mental well- being. Four of the seven excluded items showed bias for gender. It was perhaps because of this DIF (which can be a cause of multidimensionality) that it was not possible to construct a second meaningful scale from the seven deleted items. The person-item threshold distribution indicated that targeting of the Rasch-derived 7- item scale was good. The person separation index was acceptable across each of the samples tested (range 0.832 to 0.894), exceeding the minimum value of 0.7 for group level measurement (Reeve et al. 2007). This supports the ability of the measure to discriminate among groups of respondents with different levels of mental well-being.

It is interesting that the 7-item scale derived in this research is identical to the Short Warwick-Edinburgh Mental Well-being Scale (SWEMWBS), derived from the application of Rasch analysis to WEMWBS data from a UK general population sample and reported by Stewart-Brown et al. (2009). This will enable valid comparison of results between the two populations. Although the same items were deleted in both studies, the reasons for deletion were sometimes different. For example, item 1 ‘I’ve been feeling optimistic about the future’ displayed DIF for gender in the present study, with females affirming a higher response category than males at the same level of mental well-being. However, this item did not display DIF for gender in the UK general population sample (Stewart-Brown et al. 2009). A possible explanation for the difference is that women in the present study are members of an occupational group in which the proportion of women is increasing and prospects for suitable future employment are perceived more favourably.

6.3 Comparison and contrast of findings: traditional vs. modern paradigms

The findings of the examinations of psychometric properties by Rasch analysis were generally consistent with those derived by traditional methods. For example, the findings on targeting were similar (traditional methods include examination of item category endorsement frequencies and range of overall scores; the Rasch analysis method involves inspection of the person-item threshold distribution).

Conflicting results were obtained from the two approaches to the assessment of the unidimensionality of WEMWBS. Exploratory factor analysis (EFA) indicated a one- factor solution for the 14-item WEMWBS in Sample 1 and Sample 2, which was

158 Chapter 6: Discussion

confirmed by confirmatory factor analysis (CFA) for Sample 1. By contrast, the more sophisticated techniques involved in Rasch analysis indicated that the 14-item scale was multidimensional and item deletion was required to resolve a unidimensional scale. Moreover, when CFA was applied to the Rasch-derived 7-item scale for Sample 2, unidimensionality was not confirmed. Given the differing conceptual underpinnings of each approach, the paradoxical results are not surprising and support similar results that have been identified in studies of other instruments where both approaches have been applied independently to the same sample (e.g. Waugh and Chapman 2005, Gyagenda and Engelhard 2009, de Morton and Nolan 2011). Conflicting results might be due to limitations associated with using ordinal data in factor analysis (Stout 1990). In addition, there is evidence that person and item distributions can cause traditional factor analysis to report too many factors (Bond 1994, Wright 1996, van der Ven and Ellis 2000, Aryadoust 2009). Modern methods are considered to be conceptually and theoretically superior to traditional methods in this context (Hobart and Cano 2009).

6.4 Transformation of raw scores to interval measurements

The summed raw scores for the set of fitting items were mapped against the corresponding estimated Rasch-transformed logit scores.

The raw score to interval measure S-shaped curve (ogive) is almost linear for a section around the midpoint of the scale. This is reflected in the high Pearson correlation between the raw score of the derived 7-item scale and its Rasch transformed interval scale (r =0.984 in the Survey 1 dataset; r =0.991 in the Survey 2 dataset). However, near equal-spaced progression of central raw scores is not interval-level measurement. Furthermore, the approximate linearity between central scores and their corresponding measures breaks down as scores approach their extremes (Linacre 1998). The change in measurement implied by a one unit change in raw score of the derived 7-item scale varies 3.4-fold across the range of the scale, from 0.29 logits (raw score 16 to 17) to 0.99 logits (raw score 1 to 2). This has important implications for studies in which precise measurement of change underpins the inferences made. It also has implications for performing basic statistical tests on raw scores. It questions the legitimacy of addition (which underpins means and SDs), subtraction (which underpins change score analysis), division and multiplication and all the statistical tests that handle data in this way. By Chapter 6: Discussion 159

contrast, it is legitimate to undertake these statistical analyses on Rasch-derived measurements because they are on an interval scale (Hobart and Cano 2009).

Although the 7-item scale is identical to the SWEMWBS (Stewart-Brown et al. 2009), the raw score to interval measure transformations from the respective studies are inevitably not identical because they are derived from different populations. It is suggested that the transformation derived in the present study is used for future analyses in the context of the veterinary profession. Although future analyses in the context of the general population are likely to use the transformation published by Stewart-Brown et al. (2009), the differences are minimal and unlikely to affect the interpretation of comparisons between the two population samples. However, this assumption should be tested.

A limitation to the application of raw score to interval measure conversion tables such as Table 5.11 to other datasets is that they are valid only in the presence of complete data (Kersten et al. 2011). Consequently cases with missing item response values should be omitted when such transformations are applied.

6.5 Convergent construct validity

Convergent construct validity is the extent to which scores derived from a particular instrument are associated with different measures of the same or theoretically similar construct (Strauss and Smith 2009). Ideally, for the results of associations between a scale and comparator scales to be fully interpretable, evidence should exist of the reliability and validity of the comparator scales in similar population samples. However, such evidence is restricted to other population samples for the comparator scales used in the present study. This is a common limitation of external validity testing and the findings should be interpreted with this in mind (Cano et al. 2008).

The study extends support for the convergent construct validity of WEMWBS by demonstrating its associations with scales that differ from those used in previous assessments. Tennant et al. (2007) used several comparator scales which measured domains that were expected to be associated with mental well-being such as positive and negative affect, psychological functioning, life satisfaction and general health. Only one measure of psychiatric morbidity (GHQ-12, Goldberg and Williams 1988) was included. By contrast, the present study included comparator scales which measured specific domains of mental ill health such as anxiety and

160 Chapter 6: Discussion

depressive symptoms and suicidal ideation, and risk of work-related stress. The associations between WEMWBS and specific comparator measures of mental ill health were maintained when the analyses were repeated for the Rasch-derived interval scale. The convergent construct validity, or any other traditional psychometric properties, of this 7-item interval scale (SWEMWBS, Stewart-Brown et al. 2009) have not been reported previously. The scales’ associations with HADS-A and HADS-D caseness and with suicidal ideation are particularly relevant in the context of a profession with an elevated risk of suicide and support their use as overall indicators of mental health and well-being in this occupational group.

Notwithstanding the evidence presented above, which is generally considered to support external construct validity, several authors contend that current methodology is weak and our understanding of exactly what rating scales are measuring is therefore limited (Hobart et al. 2007, Cano and Hobart 2011). Statistical tests of external construct validity comprise a range of examinations to assess the extent to which scores derived from a particular instrument measure what the instrument purports to measure, based on the magnitude and direction of association with different measures of the same or theoretically related construct (Strauss and Smith 2009). A limitation of this approach is that showing that a scale does not correlate highly with measures of a dissimilar construct tells us nothing about what the scale actually measures. Similarly, showing that a scale is highly associated with a measure of a similar construct only tells us that the two are related (Hobart et al. 2007, Cano and Hobart 2011). A key problem with all statistical tests of validity is that they focus on person scores and between-person variation in these scores. They are weak because there is no independent means of assessing the extent to which the intention of the scale is attained (Stenner et al. 1983). Consequently these validation techniques entail circular reasoning (Stenner et al. 1983), generate only circumstantial evidence of validity (Massof 2002), enable limited development of construct theories, and result in ‘primitive’ understandings of exactly what is being measured (Stenner and Smith 1982). However, they have been essentially unchallenged for decades (Hobart et al. 2007).

A potential solution is an approach referred to as ‘theory-referenced measurement’ (Stenner et al. 2006) which suggests that investigators propose and test explicit construct theories about the factors that determine the location of items on the continuum on which people can be located. This process is made explicit by expressing the construct theory as a mathematical equation. The validity of a Chapter 6: Discussion 161

construct theory is the extent to which this equation predicts variation in item locations. When the equation predicts the vast majority of the variance in the item locations, it is possible to explain why items are located at different points on the continuum, i.e. the characteristics that determine item locations can be defined. Only at this point does it become acceptable to claim to have identified the construct that is being measured (Hobart and Cano 2009).

6.6 Remaining gaps in the evidence base

Several omissions remain in the evidence base to support the utility of WEMWBS and the Rasch-derived 7-item interval scale for monitoring overall mental health and well-being at a population level among the veterinary profession.

There is limited evidence to support the ability of WEMWBS to detect changes at a population level, for example after intervention. The scale was administered to parents (n=1071) before and after attendance at a parenting programme. The mean WEMWBS score increased by over 7 units, which represented a moderate to high effect size of 0.71 (Lindsay et al. 2011). However, there is currently no evidence to support the responsiveness of the Rasch-derived 7-item interval scale (SWEMWBS). Although no attempt was made to estimate responsiveness robustly using the two successive surveys employed in the present study, some small indication of the potential responsiveness to change might have been gleaned if the mean scores for the profession had changed in the intervening two years. However computations (not shown) demonstrated that this was not the case for either men or women. Appropriate methods to evaluate responsiveness include the testing of a priori hypotheses concerning the extent and direction of change in relation to comparator instruments (Revicki et al. 2008).

Test-retest reliability refers to the consistency of respondents’ scores over time. It is estimated by administering the measurement in a longitudinal study design on two or more different occasions to the same group of people (Rousson et al. 2002), typically around 2 to 4 weeks apart (Scholtes et al. 2011). The one-week test-retest reliability of WEMWBS has been assessed among UK university student (n=348) (Tennant et al. 2007) and teenage school student (n=1650) (Clarke et al. 2011) samples. Reliability was high in the university student sample (intra-class correlation coefficient, ICCC: 0.83) but lower in the teenage school student sample (ICCC: 0.66). No test-retest reliability evidence is available for SWEMWBS (although the 7- item scale is embedded within WEMWBS, it is possible that the test-retest reliability

162 Chapter 6: Discussion

of the common items is not representative of the other items in the full scale) or either scale in the context of the veterinary profession.

Social desirability response bias (a tendency to overestimate desirable traits and underestimate undesirable traits when using self-report measures) can distort the response patterns, although self-administered questionnaires are less prone to the latter than face-to-face interviews (Bowling 2005). The influence of social desirability bias on the summed total score of WEMWBS has been assessed using the Balanced Inventory of Desirable Responding (BIDR) (Paulhus and Reid 1991), in a sample of UK university students (n=116) (Tennant et al. 2007). The BIDR is a 40- item self-report instrument which includes sub-scales measuring impression management and self-deception and is administered concurrently with other instruments to identify individuals who distort their responses. Correlation between WEMWBS scores and the BIDR sub-scales suggested that the susceptibility of WEMWBS to social desirability bias was similar to or lower than that of other comparable scales (Tennant et al. 2007). However, it would be interesting to investigate the potential influence of socially desirable responding on WEMWBS scores in the context of the veterinary profession. The tendency to give socially desirable responses might be increased by a false perception that the RCVS could gain access to an identifiable individual’s results.

6.7 Validation is a process, not an outcome

The psychometric evaluations provide evidence in support of the validity of the WEMWBS and the derived 7-item scale as an overall indicator of population level mental health and well-being in the UK veterinary profession. Each new psychometric test provides additional information supporting or undermining the validation of an instrument; evidence for the validity of a measure accrues over time (Strauss and Smith 2009). Thus, validation is a process not an outcome. It would be inappropriate to claim that validity had been ‘demonstrated’, ‘proven’ or ‘confirmed’. Future studies of the use of WEMWBS and the derived 7-item scale in the UK veterinary profession and other population groups will define the instruments’ psychometric properties further. Although the validation process is ongoing, it is not necessarily infinite. If a well-validated measure does not behave as ‘expected’ in a study, the measure would not be abandoned; instead, other possible explanations for the outcome, such as deficient research design would be explored.

Chapter 6: Discussion 163

6.8 Strengths and limitations

6.8.1 Strengths

6.8.1.1 Sample and setting

Psychometric properties are not inherent to an instrument, but estimates that may vary across different samples and settings (Terwee et al. 2007). A strength of this research is that the sample and setting for administration of the instrument are appropriate to draw meaningful conclusions from the analyses. The study was based on datasets derived from two large nationwide samples of veterinary surgeons in a range of different types of employment. The samples are considered to be representative of the UK veterinary profession because the demographic and occupational profiles of respondents to the two surveys are similar and one of the surveys into which WEMWBS was embedded (Survey 2) was sent to all veterinary surgeons registered with the Royal College of Veterinary Surgeons (RCVS). Moreover, responses to the latter are used by the RCVS as the source of the most recent information on the structure of the veterinary profession. It is possible that some participants may have responded to both surveys. However, Survey 2 was administered over two years after Survey 1, during which time participant responses are likely to have changed, so there is limited potential for this to have impacted the results negatively. The response rates of 56.1% (1796/3200) (Survey 1) and 37.4% (8829/23,594) (Survey 2) are less than optimal but compare favourably with other postal surveys of the veterinary profession (cf. Hansez et al. 2008, Gardner and Hini 2006, Mellanby and Herrtage 2004), which may reflect a high level of personal salience of the subject of mental health to the sample population. The setting for administration of the instrument in Survey 2 (RCVS Survey of the Profession) is the same as that in which the instrument is likely to be used for the monitoring of trends in mental health and well-being in this occupational group over time.

Overall, the results of the psychometric analyses are considered to be generalisable to the wider population of veterinary surgeons registered in the UK.

6.8.1.2 Cross-validation of results between independently derived samples

The psychometric analyses were applied to two large, independently-derived datasets. This meant that, where possible, the results from one dataset could be cross-validated by reference to the results from the other and across subsamples of

164 Chapter 6: Discussion

the same dataset. The consistency of the results in these cross-validations contributes greatly to the robustness of the study findings and their interpretation.

6.8.1.3 Comprehensive approach: complementary application of traditional and modern methods

The study adopted a comprehensive and rigorous approach to the assessment of the psychometric properties of WEMWBS and the derived 7-item scale, by the complementary application of both traditional and modern methods, to enhance understanding of the performance of the scales in the context of the UK veterinary profession. While the traditional approach provides relevant information regarding, for example, missing values distributions and external (convergent) construct validity, Rasch analysis is a powerful tool for the assessment of differential item functioning (DIF), ordering of response category thresholds, dimensionality and the extent to which an instrument measures a continuum of severity.

Evaluation of the internal construct validity of WEMWBS and item reduction were primarily guided by Rasch analysis. The advantages of this method include the following: (1) it provides measurements of people that are independent of the sampling distribution of the items used and locates items in each scale independent of the sampling distribution of the people in whom they are derived; (2) as such, it allows for more accurate individual person measurements on ‘fixed’ rulers; and (3) therefore improves the potential for health rating scales to measure clinically meaningful change (Hobart and Cano 2009).

After Rasch analysis, the derived 7-item scale was evaluated using traditional psychometric methods to provide traditional psychometric evidence for the instrument (especially in relation to convergent construct validity) and to help bridge the knowledge gap between old and new psychometric methods. It is important to note that from both traditional and modern psychometric perspectives, the scale meets the accepted contemporary criteria for measurement.

The methodology adopted for the traditional and modern psychometric analyses and the reporting of results is consistent with current recommendations (Smith et al. 2003, Tennant and Conaghan 2007, Hobart and Cano 2009).

Chapter 6: Discussion 165

6.8.1.4 Extends evidence to support external validity

The research extends support for the convergent construct validity of WEMWBS by demonstrating its associations with scales that differ from those used in previous assessments. In addition, the convergent construct validity (or any other traditional psychometric properties) of the Rasch-derived 7-item interval scale (SWEMWBS, Stewart-Brown et al. 2009) has not been reported previously.

6.8.1.5 Objectives achieved

The study aimed to examine the measurement properties of WEMWBS in the context of the veterinary profession to inform judgements concerning: [i] the defensibility of comparing WEMWBS scores for the veterinary profession with those for other groups, and [ii] the potential utility of the instrument for future research, including, for example, the monitoring of clinically meaningful trends in mental health and well-being in this occupational group over time. The study achieved all the objectives described previously (Section 1.5 and Section 6.0).

6.8.2 Limitations

6.8.2.1 Cross-sectional study design

The interpretation of some of the results is limited by the cross-sectional design, which allows investigation of associations but does not allow conclusions to be drawn about the direction of causality. For example, convergent construct validity was examined by computing associations between WEMWBS or the derived 7-item interval scale and other scales including HADS. A one unit increase in the interval scale was associated with a 52% reduction in the odds of reporting HADS-D probable caseness (OR: 0.48, 95% CI: 0.41 to 0.56, p<0.001). While this association provides robust evidence to support the construct validity of the scale under investigation, care must be taken with interpretation because it is not possible to infer the direction of causality in this inverse relationship.

The cross-sectional study design also meant that test-re-test reliability and responsiveness of the instrument could not be assessed (see Section 6.6).

6.8.2.2 Extent of validity and reliability of comparator scales

The evidence to support the convergent construct validity of WEMWBS and the derived 7-item interval scale is limited by the assumption that the comparator scales

166 Chapter 6: Discussion

used are reliable and valid in the context of the UK veterinary profession. Evidence to support the validity and reliability of the comparator scales is restricted to samples of other populations. This is a common limitation of external validity testing and the findings should be interpreted with this in mind (Cano et al. 2008, Mokkink et al. 2010a). Moreover, several of these evaluations in other populations identified psychometric issues. For example, concerns regarding the psychometric properties of the HADS scale (Zigmond and Snaith 1983) are well documented (summarised by Lambert et al. 2011). In addition, the comparator scales were ordinal and not transformed to interval scales before testing the associations using parametric methods.

6.8.2.3 Absence of qualitative assessments

The application of qualitative methods in addition to psychometric evaluations can enrich understanding of the properties of health rating scales (Hobart et al. 2007, Persson et al. 2008, Clarke et al. 2011, Cano and Hobart 2011). For example, qualitative attributes such as acceptability, comprehensibility, ease of use, format, and the extent to which the items are considered to map out a comprehensive and meaningful continuum of mental health and well-being could have been assessed in focus groups of UK veterinarians. Debriefing interviews could be conducted to evaluate consistency in meaning. However, it was cautiously assumed that evidence to support the acceptability of these characteristics of the scale was adequately provided by previously reported qualitative contributions during scale development and validation among the UK general population (Tennant et al. 2007). Qualitative assessments would be more important if the instrument was translated into other languages to enable comparative studies of mental health and well-being among veterinary surgeons across different countries.

6.8.2.4 Potential for bias

The setting for administration of the instrument may have confused some respondents into contextualising their responses within the work setting rather than considering their lives as a whole – the instrument was embedded into questionnaires which were clearly identifiable as being specific to this particular occupational group. It would be interesting to discuss with a representative sample of respondents whether they consciously contextualised their responses in this way. Chapter 6: Discussion 167

Selection bias is unlikely to have influenced the results because the random sample selected for Survey 1 was stratified by type of work with proportions of each type of work that are representative of the wider population of veterinary surgeons practising in the UK. Representativeness was further confirmed by comparison of the gender and decade of qualification profile of the sample with RCVS membership data for veterinary surgeons practising in the UK. The questionnaire for Survey 2 was mailed to all veterinary surgeons registered with the Royal College of Veterinary Surgeons rather than a selected sample population.

Response bias due to respondents differing systematically from non-respondents had potential to influence the results. Veterinary surgeons with mental health problems may have been disinclined to respond due to their symptoms or concerns that they might be identified, or conversely those without mental health problems may have considered that the instrument was not relevant to them.

Respondents may have affirmed higher scoring categories if they were concerned that their individual results may have been identifiable. Such socially desirable responding was not assessed. Conversely, respondents who were concerned that working conditions were adversely affecting the mental health of the profession may have affirmed lower scoring categories in an attempt to deliberately skew the results to prompt the RCVS into remedial actions.

The sequential order of presentation of composite measurement scales within a questionnaire may influence the pattern of responses (Lucas 1992, Cheung et al. 2005, Kieffer et al. 2011). The possibility of an instrument order effect requires caution in the comparison of results across studies in which the scales may not have been presented in the same order or context. No attempt was made to control for any possible order or other contextual effects in the current study by, for example, systematically rotating the order of instruments within the questionnaire and counterbalancing the order of completion across participants.

6.8.2.5 Generalisability to other population groups

The psychometric properties of instruments are generally regarded as being population-specific (Meredith and Teresi 2006, Terwee et al. 2007). While the psychometric evaluations in this research are considered to be robust for the UK veterinary profession and are consistent with those reported for UK general

168 Chapter 6: Discussion

population and student samples, they are not necessarily generalisable to other occupational groups or to the veterinary profession in other countries.

6.9 Implications for research

6.9.1 Implications of psychometric evaluations

6.9.1.1 Health rating scales: the potential for misinterpretation

It is vital that researchers are aware of the key issues surrounding the use and analysis of rating scale data because rating scales have an increasingly crucial role in the determination of patient care, the guidance of clinical research directions, the evaluation of advances in basic science and the evaluation of clinician professionalism (Hobart et al. 2010). Each of these eventually impacts on patients, clinical practice and clinicians. There is considerable potential for misinterpretation if the analysis of scale scores is based on the assumption that ordinal scores are measurements, without recognising the limitations.

Health rating scales are instruments which are usually intended to quantify characteristics of people that cannot be measured directly (latent variables). They are constructed to map out variables as a line on which people can be located. Psychometric methods concern the construction and evaluation of rating scales in order to determine their success at operationalising the variable and identifying the location of people. Traditional psychometric methods are widely used for constructing and evaluating rating scales. Classical Test Theory (CTT), which underpins these methods, postulates the existence of a true score, that error scores are uncorrelated with each other and with the true scores, and that observed true and error scores are linearly related. However, because true scores and error scores cannot be determined, the appropriateness of the assumptions cannot be verified and it is only possible to surmise that they are met. Most health rating scales use Likert’s method of summated ratings. Here, ordered response options are allocated sequentially ordered integers, and item scores are summed to give total scores. Whilst this method is widespread in health measurement, the belief that ordinal level total scores approximate interval-level measurements is not well founded. Other problems arising from the use of total scores and traditional psychometric methods are that evaluations of scales are sample dependent and the measurement of people is scale dependent. Even at a purely logical level these undermine a belief in the interpretation of total scores as measurements. Chapter 6: Discussion 169

6.9.1.2 Rasch analysis: findings beyond the reach of traditional methods

Rasch analysis subjects rating scales to a level of psychometric scrutiny which is beyond the reach of traditional methods. The traditional psychometric paradigm is based upon correlational or descriptive analyses to evaluate scaling assumptions (legitimacy of summing items), reliability and validity. By contrast, Rasch analysis uses a mathematical model (the Rasch model) to evaluate the legitimacy of summing items to generate measurements (Cano et al. 2011). As traditional psychometric methods are based on a weak measurement theory (Hobart et al. 2007, Hobart and Cano 2009), the methods that are based upon it are limited in the extent to which they can adequately provide data to support the use of health rating scales. Importantly, a key limitation of traditional psychometric methods is that they mask individual item performance within the context of the scale. Rasch analysis, based on a strong measurement theory, (Hobart et al. 2007, Hobart and Cano 2009) adds sophistication and refinement to traditional psychometric methods, and provides detailed item-level data to diagnose measurement problems within the instrument.

The present study demonstrates the value of the complementary application of both traditional and modern methods to increase the comprehensiveness of the psychometric evaluation. The traditional approach provides relevant information regarding, for example, missing values distributions and external construct validity; Rasch analysis enables the examination of differential item functioning (DIF), ordering of response category thresholds, dimensionality and the extent to which an instrument measures a continuum of severity. The added value of Rasch analysis over traditional methods was clearly highlighted in the present study by the key limitations it uncovered relating to the validity of the WEMWBS: individual item misfit (over- or under-discrimination) and differential item functioning (DIF). A 7-item scale was derived which conformed to model expectations and enabled raw score to interval measure transformation. By highlighting and elaborating on the key problems in WEMWBS items, these analyses extend the limited existing evidence- base to inform further research aimed at improving the scale.

6.9.1.3 Need for further evaluation: responsiveness and test-retest reliability

The study was limited by the extent to which the important psychometric properties of test-re-test reliability (stability of responses over time) and responsiveness (sensitivity to change) were evaluated. Although there is some evidence to support

170 Chapter 6: Discussion

these properties for WEMWBS (Tennant et al. 2007, Lindsay et al. 2011), the evidence is limited to UK general population samples only and there is no published data on these properties for the Rasch-derived 7-item scale (Short Warwick- Edinburgh Mental Well-being Scale, SWEMWBS). The potential of the scales to detect change is indicated by the minimal item and total score floor and ceiling effects identified. However, test-re-test reliability and responsiveness should be evaluated further within the context of the veterinary profession before the scales can be advocated as being appropriate for use as tools for the purpose of monitoring population level changes in the mental health and well-being of this occupational group or as outcome measures for intervention studies.

Test-retest reliability could be evaluated by administering the measurement in a longitudinal study design on two or more different occasions to the same group of people (Rousson et al. 2002). Possible methods to evaluate responsiveness include the testing of a priori hypotheses concerning the extent and direction of change following interventions in relation to comparator instruments (Revicki et al. 2008). The application of Rasch analysis has potential to advance the examination of rating scale responsiveness over traditional methods by enabling, for example, analysis of responsiveness of individual items with respect to their calibrated positions within a measure, and estimates of responsiveness based on interval-level measurements (Granger 2008, Hobart et al. 2010, Batcho et al. 2011). Rasch analysis would also enable responsiveness to be examined at the individual person level, although this is less relevant if the scale is intended for the measurement of population-level changes.

6.9.1.4 Opportunity for future refinement of the scales

The 14-item WEMWBS scale captures a wide conception of mental well-being including affective-emotional aspects, cognitive-evaluative dimensions and psychological functioning. Other potential components of mental well-being such as spirituality and meaning in life are not included (Tennant et al. 2007). Furthermore, in terms of face validity, the Rasch-derived 7-item scale presents a more restricted view of mental well-being than the 14-item scale, with most items representing aspects of psychological and eudaimonic well-being, and few covering hedonic well- being or affect. The scale may benefit from modification in the future to accommodate other components which the veterinary profession consider to be related to mental well-being. These could be identified by the qualitative research Chapter 6: Discussion 171

outlined in Section 6.8.2.3. It would be beneficial to retain the items comprising the Rasch-derived 7-item scale (SWEMWBS) to enable comparison of measurements with those for the general population. The psychometric properties of the scale would need to be reassessed carefully if refinements were made.

6.9.1.5 Invariance of psychometric properties should not be taken for granted

An issue which has potential to confound the interpretation of health rating scale scores is the extent to which evidence for the psychometric properties of validity, reliability and responsiveness for an instrument can be generalised across different groups of people. While this research provided evidence to support the measurement invariance of SWEMWBS across two different sample populations (the UK general population and the veterinary profession), it is important to reinforce the need to test the psychometric properties of instruments in samples of the target population in which they are to be used. Psychometric properties are not inherent to an instrument, but estimates that may vary across different samples and settings (Terwee et al. 2007). Rating scales are not universally applicable – examples of measurement variance across sample populations are manifold (e.g. Gregorich 2006, Meredith and Teresi 2006, Garnaat and Norton 2010, Lambert et al. 2011). Comparison of psychometric properties across various samples and settings lends new insights into the possible uses and/or limitations of an instrument. To the extent to which it is possible, the psychometric properties of an instrument should be examined in every study in which the instrument is used. Using measures without critical review can have negative repercussions on the interpretation of study results, which in turn might compromise the extent to which research findings benefit the study population.

6.9.1.6 Scale selection: WEMWBS or the Rasch-derived 7-item version?

The 7-item scale offers superior measurement properties compared to the 14-item WEMWBS from which it is derived. These include:

Interval measurement: The summed raw scores for the derived 7-item scale can be transformed to interval-level measurements. An interval-level measurement system is one in which the unit of measurement implied by the data is equal across the whole range of the continuum. Interval-level measurement means that mental well-being can be quantified with greater precision. In addition, assessments of within-group changes and between-

172 Chapter 6: Discussion

group differences are strengthened by the compatibility of the interval measurements with parametric statistics applied to test cross-sectional or longitudinal differences in scores.

Unidimensionality: The items define a single underlying construct, i.e. mental well-being.

Invariant comparison: The invariance criterion means that an instrument, in principle, is required to work in the same way for all individuals. This implies invariant functioning across subgroups within the population, e.g. across gender, age group and different levels of mental well-being.

The brevity of the 7-item instrument is a further advantage if respondent burden is a concern or there are space constraints in the questionnaire. However, given that SWEMWBS is embedded within WEMWBS, it may be appropriate to continue to collect data on all 14 items to enable further investigation of psychometric properties such as dimensionality and DIF in other samples. This will also allow for very cautious comparisons with the results of studies in other populations which only report results for the full scale.

6.9.2 Implications of the wider findings of the literature review

6.9.2.1 Qualitative investigation of influences on suicide risk

Studies of potential influences on the elevated suicide risk among the UK veterinary profession are currently limited to quantitative investigations (Bartram et al. 2009a, b, c). Mixed methods research, integrating qualitative and quantitative data, holds potential for a more comprehensive understanding of mental health and well-being in the profession which may have implications for the provision of interventions. The data collection for two qualitative studies is completed and plans have been made for analysis and reporting. A synopsis of each of these studies is provided in Appendix 8.

(1) A semi-structured interview study aims to explore the causal attribution of suicidal thoughts among a purposive sample of respondents to Survey 1 – Understanding suicidal thoughts and help-seeking behaviour among UK veterinary surgeons: a semi-structured interview study (SOMSEC026.09).

(2) A systematic examination of data from Coroners’ inquest proceedings for deaths of veterinary surgeons which received a verdict of suicide seeks to Chapter 6: Discussion 173

identify factors which support or refute the hypothetical model for suicide risk (Figure 2.1) and explore the ways in which the inquest process elucidates the final pathway to suicide – Qualitative analysis of Coroners’ data for veterinary surgeon suicides within a region of England (SOMSEC051.10).

6.9.2.2 Longitudinal studies to explore causality of associations

Prospective longitudinal studies would help to determine whether the cross-sectional associations identified in a previous study by Bartram et al. (2009a, b, c) are causal. The mental health and well-being of the cohort of respondents who voluntarily identified themselves in that study could be followed up over time for this purpose. A multiphase design with cause and effect variables measured at each phase would help to determine the rate of change of possible causal relationships and allow for the examination of reciprocal effects (Taris and Kompier 2003).

6.9.2.3 Examination of each stage of the veterinary career path

The focus of the majority of studies to date is on practising veterinary surgeons (studies are summarised in Table 2.13). However, each stage of the veterinary career path – from the characteristics of applicants to veterinary schools, undergraduate training, subsequent employment, and through to retirement – requires examination to identify early predisposing factors and later triggers for suicidal behaviour in the profession. This ‘life course’ approach to studying suicide (Gunnell and Lewis 2005) will enable multiple points on the career continuum to be targeted with appropriate interventions. For example, research with veterinary students could help identify whether there is a predilection towards mental health problems in applicants to veterinary school, whether there are negative influences on mental health during undergraduate training, and whether individual-specific maladaptive coping strategies might play a role in the development of mental ill- health.

6.9.2.4 The role of attitudes towards suicide

The role of attitudes towards suicide should be considered when investigating differences in suicide rates between population groups (Etzersdorfer et al. 1998). It has been suggested that attitudes towards suicide (the degree to which an individual views suicide as an acceptable option under some circumstances) may moderate the link between hopelessness and depressive symptoms and levels of suicidal ideation (Gibb et al. 2006). Attitudes can also influence help-seeking behaviour. The

174 Chapter 6: Discussion

veterinary profession’s role in the provision of animal euthanasia and the facilitation of a ‘good death’ may normalise suicide, with death perceived as a rational solution to an intractable problem. There has been no rigorous examination of this hypothesis to date. Data on attitudes towards suicide could be collected using, for example, items from the Suicide Opinion Questionnaire (Domino et al. 1982).

6.9.2.5 International research collaboration

The development of international research collaborations with countries in which suicide risk among veterinarians does not appear to be elevated, such as Denmark (Hawton et al. 2011) and New Zealand (Skegg et al. 2010), may be of value. Comparative studies of well-being, lifestyles and access to support could inform the development of interventions for those countries in which risk is higher.

If the WEMWBS or SWEMWBS was used in international research the instrument would need to be translated into the appropriate language, care taken to ensure that the meaning of the items was consistent, and the psychometric properties reassessed in the new sample population.

6.9.2.6 Translating research into evidence-based interventions

Work is required to address the challenge of translating research findings into evidence-based interventions which will then need to be piloted and their effectiveness and utility evaluated across the range of different veterinary work settings.

6.10 Implications for the profession

6.10.1 Implications of psychometric evaluations

6.10.1.1 Monitoring of trends in mental health and well-being

This research examined the measurement properties of WEMWBS in the context of the UK veterinary profession and provided evidence to support the utility of the instrument for future research, including, for example, the monitoring of trends in mental health and well-being in this occupational group over time and changes in response to interventions. The study also provides a robust estimate of mental health and well-being against which the trends and changes can be monitored. Chapter 6: Discussion 175

The RCVS has confirmed its intention to embed the instrument into future editions of the RCVS Survey of the Profession. The date of the next survey is not yet established, although typically the surveys take place approximately every four years (L. Lockett, personal communication).11 This questionnaire provides a suitable medium for the monitoring because it is mailed to the entire membership, and the present examination of the psychometric properties of WEMWBS used data derived from this setting. WEMWBS would be an appropriate instrument for inclusion in the questionnaire as it is designed to measure population-level changes in mental well- being and the present study provides evidence that the score is associated with other measures including anxiety and depressive symptoms and suicidal ideation. The Rasch-derived 7-item version of the scale (SWEMWBS) has more robust measurement properties than the full scale while retaining external construct validity. However, given that SWEMWBS is embedded within the larger WEMWBS, unless greater brevity is required due to space constraints in the questionnaire, it may be appropriate to continue to collect data on the all 14 items to enable further investigation of psychometric properties such as dimensionality and DIF in other samples. This will also allow for comparisons, albeit at the ordinal level and with great caution in full knowledge of the limitations of such comparisons, with the results of studies in other populations which only report results for the 14-item scale.

6.10.2 Implications of the wider findings of the literature review

6.10.2.1 Towards enhanced mental well-being and suicide prevention

Interventions to improve psychological health have been implemented effectively in a variety of different work settings (Michie and Williams 2003). An occupation- specific suicide prevention programme implemented in the US Air Force – aimed at decreasing stigma, enhancing social networks, facilitating help seeking, and enhancing understanding of mental health – reported success in lowering suicide rates (Knox et al. 2003, Knox et al. 2010).

Studies suggest that poor mental health among doctors is associated with, for example, being irritable with patients, taking short-cuts and not following procedures, which may markedly reduce standards of patient care (Firth-Cozens 2001, Taylor et al. 2007). Given similarities between medical and veterinary practice, intervention is important, not only for the well-being of individual members

11 Lizzie Lockett, Head of Communications, RCVS, London. 30 September 2011 [E-mail].

176 Chapter 6: Discussion

of the veterinary profession, but also to safeguard veterinary public health and the health and welfare of animals committed to veterinary care.

The structure of the UK veterinary profession is highly fragmented. Most practices are small businesses without the benefit of dedicated human resource staff or occupational health professionals to advance improvements in working conditions and the management and health of employees. Employers have ethical and legal duties to assess the risk of stress-related ill-health arising from work activities and take measures to control that risk. However, a balance must be achieved between personal and organisational responsibility for mental health and well-being; a comprehensive programme of interventions including both individual-oriented and work environment-oriented elements delivers the most favourable outcomes (Bond 2004, LaMontagne et al. 2007).

Rose ([1992] 2008) observed that the prevalence of many diseases in a population is directly related to the population mean of the underlying risk factors: when there is a small shift in the mean of a normal distribution there is a large shift in the tails of the distribution (Figure 6.2).

Fig. 6.2: The mental health spectrum

Shifting the mean of a population may have a substantial impact on the tails, e.g. decreasing levels of mental disorder and increasing flourishing. Based on a figure in Huppert et al. (2005).

Consequently, shifting the mean of the risk factors in a population by population- level interventions leads to a decline in the prevalence of the disease. The Chapter 6: Discussion 177

prevalence of psychiatric disorder in a population is linearly related to the mean number of psychiatric symptoms in that population (Anderson et al. 1993, Whittington and Huppert 1996). Anderson et al. (1993) concluded:

Populations thus carry a collective responsibility for their own mental health and well-being. This implies that explanations for the differing prevalence rates of psychiatric morbidity must be sought in the characteristics of their parent populations; and control measures are unlikely to succeed if they do not involve population-wide changes. (p. 475)

It is important therefore that interventions within the veterinary profession are directed at the level of the entire occupational population, rather than merely targeting those in psychological distress.

The cross-sectional nature of studies of mental health in the profession (for example, Gardner and Hini 2006, Bartram et al. 2009a, b, c; Fritschi et al. 2009) precludes inference of causality but the results suggest a number of primary (preventive), secondary (ameliorative) and tertiary (reactive) interventions, for both organisations and individuals, which have the potential to reduce psychological distress in the profession. Consideration is given to the concept of ‘countervailing interventions’ – those focused on increasing the positive experience of work rather than decreasing the negative aspects – by incorporating some of the sources of satisfaction identified by Bartram et al. (2009c), as these may help counteract the effects of workplace stressors (Kelloway et al. 2008). Proposed interventions are summarised in Tables 6.1 and 6.2. Research is required to assess the effectiveness of these proposed interventions on the dual outcomes of veterinary surgeons’ mental health and work performance in terms of animal and client care (for example, reduction in complaints and litigation). This approach applies the precautionary principle in recognising the need for further research on interventions while simultaneously advocating that there is adequate evidence from outside the veterinary profession to justify prompt concerted action, and is in accordance with the assertions of Rose ([1992] 2008) on preventive medicine:

Certainty is not a prerequisite for action. …that action may then proceed alongside continuing research and evaluation, recognising that new evidence may mean a change of policy. (p. 145)

178 Chapter 6: Discussion

Employee participation in the process of introducing interventions can help to refine interventions to optimise their fit within the context of a specific workplace, and may in itself help to diminish some dimensions of work-related stress (LaMontagne et al. 2007).

The following are possible areas for intervention to improve the mental health and well-being of veterinary surgeons, based on evidence of their need within the profession and efficacy of such interventions in other fields.

6.10.2.2 Possible areas for intervention

6.10.2.2.1 Mental health promotion

A person’s psychological distress may be invisible to others, except members of their close social networks, including work colleagues, whose capacity to recognise, assess and respond to distress is therefore vitally important (Owens et al. 2009). Moreover, signs of distress are diverse (Rudd et al. 2006) and communications are often oblique and ambiguous (Owens et al. 2011). Raising awareness among the general public of potential warning signs can increase their ability to recognise that a person is suicidal (Van Orden et al. 2006). Educational initiatives to improve personal mental health and well-being and encourage effective identification and support of distress in others should be integrated into the early undergraduate curriculum and made accessible to those already working in the profession. Topics could include evidence-based techniques to: enhance intra- and inter-personal adaptive coping skills; enable early recognition of mental ill-health in oneself and others; recognise and engage suicidal colleagues to help them to stay safe and seek further assistance; increase awareness of the dangers of alcohol misuse; challenge stigma and discrimination; and encourage help-seeking behaviours. Peer- led tuition could be provided by a national network of suitably trained veterinary surgeons. Primary interventions at undergraduate level are likely to have the greatest long-term benefit and be among the easiest to implement (Gerrity 2001). Timely intervention with young people who show early signs of mental ill-health is important as younger age of onset of depression is associated with greater suicidal intent, irrespective of the age of onset of suicidal thoughts (Thompson 2008). A three-hour suicide awareness workshop (safeTALK) (McLean et al. 2007) was effective in improving suicide awareness and intervention skills among a small cohort of veterinary undergraduates, as part of their professional development curriculum (Mellanby et al. 2010). However, neither the durability of changes in Chapter 6: Discussion 179

awareness nor the ability of participants to offer improved detection and support for colleagues with suicidal thoughts or behaviours was assessed.

Consideration should be given to the production of a short educational video for use in veterinary schools and conferences as part of an outreach campaign to raise awareness of suggested measures to reduce the levels of psychological distress in the profession.12

6.10.2.2.2 Research and monitoring

Potential future research initiatives are discussed in Section 6.9.2. The potential use of WEMWBS and the Rasch-derived 7-item interval scale for monitoring of trends in mental health and well-being in the veterinary profession over time and changes in response to interventions are discussed in Section 6.10.1.1.

6.10.2.2.3 Accessible and appropriate support services

There is a need to increase awareness (Figure 6.3) and evaluate the effectiveness of existing initiatives to support mental health in the UK veterinary profession, such as Vet Helpline and Veterinary Surgeons’ Health Support Programme, to determine whether modification or additional provision is required. Consideration could be given to the introduction of a telephone counselling service, possibly as part of an employee assistance programme provided by individual veterinary employers or associated with membership of a veterinary professional organisation, such as the British Veterinary Association. Similarly, a service to facilitate employment dispute resolution and encourage better employment relations within the veterinary profession has potential to improve working life.

12 The American Foundation for Suicide Prevention developed a public television documentary (Struggling in Silence) (broadcast in May 2008) and two educational videos for use within the medical profession as part of an outreach campaign to help reduce physician depression and suicide. Available at: http://www.afsp.org/index.cfm?fuseaction=home.view Page&page_ID=A8913D2A-B133-78B0-E7ADEB9D84C8B34B. Accessed: 04 October 2011. This is one of a series of initiatives to implement the recommendations of a workshop convened to devise a strategy to address mental health issues within the medical profession (Center et al. 2003).

180 Chapter 6: Discussion

Fig. 6.3: An example of advertising to raise awareness of support services

Advertisements to raise awareness of Vet Helpline (24-hour peer-support telephone helpline), Veterinary Surgeons Health Support Programme (professional treatment and support for addiction and other mental health disorders) and Veterinary Benevolent Fund (financial support in necessitous circumstances for veterinary surgeons and their families).

6.10.2.2.4 Working conditions

Veterinary surgeons self-report less favourable psychosocial work characteristics than the working general population in relation to demands and managerial support (Bartram et al. 2009b, c). There is consistent evidence from several cross-sectional and longitudinal studies that high psychological demands, low decision latitude (control) and low social support at work are predictive of mental ill-health (Stansfeld and Candy 2006, Bonde 2008, Netterstrøm et al. 2008). Working conditions are potentially modifiable and improvements can enhance mental health and well-being. Successful interventions that improve psychological health use training and organisational approaches to increase participation in decision-making, increase support and feedback, and improve communication (Michie and Williams 2003).

Veterinary surgeons in the UK report relationships with colleagues and clients as important sources of work-related satisfaction (Bartram et al. 2009c). Consequently, Chapter 6: Discussion 181

encouraging both formal and informal supportive relationships between work colleagues and with clients has the potential to improve mental health and wellbeing. The development of a workplace culture in which there is regular constructive feedback and where problems are addressed sensitively could generate a more supportive work environment. Opportunities for clinical supervision, mentorship and review could be expanded and collaborative team-working encouraged. The importance of workplace relationships has been recently highlighted by research associating a poor team climate at work with depressive disorders and subsequent antidepressant use (Sinokki et al. 2009). Deficiencies in people-management skills, including supervision, team leadership and competence in strategies for the prevention and reduction of work-related stress among employees, could be identified and addressed in veterinary practices. Meetings to discuss stressful work experiences appear to protect against suicidal ideation among female physicians (Fridner et al. 2009), suggesting that regular implementation of such meetings may have similar value among veterinary surgeons.

High levels of job satisfaction have been shown to protect doctors’ mental health against the harmful effects of job stress (Taylor et al. 2005). Interpersonal relationships are a central theme of mental health promotion programmes (Stewart- Brown 2005). Durkheim ([1897] 2006) observed over a century ago that the frequency of suicide varied inversely with social cohesion:

…we might observe that since the strength of the collective is one of the obstacles that best limits suicide, this cannot weaken without suicide rising. (p. 225)

…in a coherent and vital community, there is a continual exchange of ideas and feelings from all to each and from each to all which is like mutual moral support, so that the individual, instead of being reduced to his resources only, participates in the collective energy and draws on it when his own is exhausted. (pp. 225-226)

Few studies have examined closely the social networks of those who take their own lives, the nature of the bonds that unite individuals and the ways in which these may promote or constrain both suicidal behaviour and attempts to intervene (Owens et al. 2009). While it may be facile to suggest that measures to improve social support

182 Chapter 6: Discussion

within the veterinary profession would make suicide less likely, the possibility should not be discounted.

The structure of the veterinary profession is presently undergoing considerable changes, including increasing numbers of women graduates, increasing numbers of practices in corporate ownership, a decline in traditional mixed practice in favour of specialisation, and a trend towards consolidation of farm animal practices into fewer units covering larger geographical regions. Higher perceived amounts of workplace change have negative effects on psychological health (e.g. Loretto et al. 2005). Perceived co-worker support and accurate information about impending change can reduce the impact on health (Platt et al. 1999). Organisations should provide employees with timely information to enable them to understand the reasons for proposed changes and ensure adequate employee consultation on changes and opportunities for employees to influence proposals.

All employers have a legal responsibility under existing UK health and safety at work legislation to ensure the health, safety and welfare of their employees and are required to undertake risk assessments and take reasonably practicable steps to address the risks identified, but failings in active health risk management systems in UK small-animal veterinary practices have been reported (D’Souza et al. 2009). The legal responsibility includes minimising the risk of stress-related illness. There is a need to ensure that attention to mental health, and not merely physical health, is recognised by veterinary employers as an integral requirement of the legislation. The recent incorporation of formalised procedures for surveillance of psychosocial risks, and the development and implementation of suitable plans for their mitigation, into the requirements for accreditation under the RCVS Practice Standards Scheme (RCVS 2010b) is creditable, especially in view of the potential effects of the mental health and well-being of veterinary surgeons on animal care. Practices rely heavily on health and safety guidance produced by veterinary professional bodies (D’Souza et al. 2009) and consequently, by the previous omission of any reference to work- related stress in the standards scheme requirements relating to health and safety of employees, the RCVS may have been inadvertently reinforcing any perception among employers that health and safety legislation only relates to physical risks. Bartram and Turley (2009) provided guidance on risk assessment for work-related stress in the context of veterinary practices using an approach advocated by the Health and Safety Executive. Further materials could be developed that present a compelling business case for psychological health and safety at work, beyond Chapter 6: Discussion 183

ethical and legal obligations, and inform veterinary businesses on how to create and sustain working environments that optimise the mental health of employees. Workplaces implementing specific measures to improve the mental health and wellbeing of veterinary surgeons should be identified and publicised as models of best practice, such as Mill House Veterinary Surgery and Hospital, Kings Lynn which became the first veterinary practice to be awarded Investors in People gold status and health and well-being good practice accreditation (Anon 2010).

The RCVS Professional Development Phase (PDP) aims to support new graduates during their attainment of clinical skills in their first year in practice. The incorporation of elements of emotional reflection into this phase would give recognition to the importance of psychological health and encourage new graduates to consider their emotional needs and development opportunities early in their careers.

6.10.2.2.5 Other personal and work-related stressors

The total number of hours worked and out-of-hours on-call duties are reported by study respondents to be major contributors to stress (Bartram et al. 2009c), although, in keeping with studies among doctors (Tyssen and Vaglum 2002), a recent study (Bartram et al. 2009c ) failed to show an association between working long hours and mental health and well-being. Nevertheless, the requirements of the Working Time Regulations (1998)13 – which prescribe maximum working hours and minimum rest periods – should be observed. The related issue of whether inactive periods during on-call duties contribute towards the total number of hours worked requires prompt resolution.

Making professional mistakes, the possibility of client complaints and/or litigation, and client expectations are reported to be major contributors to stress among veterinary surgeons in the UK (Bartram et al. 2009b, c). Consultation skills training focused on improving competencies in communication to manage client expectations and deal with complaints may help to mitigate these stressors. The curricula of some UK veterinary schools contain elements of professional communication skills training (Gray et al. 2006) but there is scope for this to be greatly expanded. The RCVS and professional indemnity insurance providers should give consideration to how to improve veterinary surgeons’ access to

13 The Working Time Regulations (1998) require that employees should not have to work more than an average of 48 hours per week unless they choose to opt out. Available at: http://www.opsi.gov.uk/si/si1998/19981833.htm. Accessed 04 October 2011

184 Chapter 6: Discussion

appropriate advice and support when cases are referred to them. There is a general consensus on a need to update the RCVS complaints and disciplinary procedures (House of Commons Environment, Food and Rural Affairs Committee [EFRACom] 2008) in the interests of both the public and the profession. Efforts to improve veterinary surgeons’ respect for and confidence in the procedures could relieve some of their associated concerns. The impact of the recent development and implementation of the RCVS Health Protocol (RCVS 2010a), which aims to protect animals and the interests of the public by helping veterinary surgeons whose fitness to practise may be impaired because of health, has not yet been evaluated.

In the current study, managing personal finances is self-reported as an important stressor for recent veterinary graduates (Bartram et al. 2009c). It is likely that concerns will rise among future graduates given the introduction of tuition fees in 2006 and the increases which have been recently confirmed. The provision of debt management guidance during undergraduate training has the potential to help reduce the stress associated with the repayment of student loans after graduation. Consideration should be given to lobbying government to allow students’ costs associated with compulsory extramural studies to be incorporated into the tuition fee loan structure.

6.10.2.2.6 Work-home interaction

Some veterinary surgeons experience high levels of negative work-home interaction (Bartram et al. 2009b). Renewed attention to existing policies and practices is recommended to address the extent to which work and home life are in conflict for both men and women in the profession. This may be achieved through flexible working hours, reduced working and on-call hours, or job sharing as negative work- home interaction is associated with the number of hours worked and on-call. The formation of out-of-hours co-operatives with local practices or the use of dedicated providers of out-of-hours veterinary care has potential to reduce negative work- home interaction.

6.10.2.2.7 Crisis intervention and return to work

Attempts should be made to facilitate the early identification of signs of psychological distress at work and enable appropriate and supportive intervention, which may involve making reasonable adjustments to the work environment to enable an individual to continue working. Screening workers for early signs of Chapter 6: Discussion 185

depression and encouraging effective treatment can improve clinical outcomes and work performance (Wang et al. 2007). Associated initiatives to help break the stigma of mental health as a ‘weakness of character’ would be complementary.

Employers should create a positive culture of acceptability of sickness absence which enables veterinary surgeons to take time off work without excessive feelings of guilt, encourages them to recover sufficiently before returning to work, and manages the consequences of individual absences for their colleagues in a fair and equitable way.

6.10.2.2.8 Limiting access to means

The elevated levels of psychological distress reported among UK veterinary surgeons suggest that the ready availability of lethal means is probably not operating in isolation to increase suicide risk within the profession (Bartram et al. 2009b). However, there is overwhelming evidence of an association between suicide rates and availability of lethal means (Hawton 2007) and of the effectiveness of restrictions on access to lethal means as an approach to suicide prevention due to limited method substitution (Daigle 2005, Mann et al. 2005, Hawton 2007, Florentine and Crane 2010, Thomas and Gunnell 2010). Consequently, consideration should be given to mechanisms to moderate the opportunity for veterinary surgeons to use certain medicines for suicide at times of known high risk, notwithstanding the immediate implications for dispensing and administration of medicines. In other professions where ready access to lethal means is inevitable, such as the use of firearms by the military, at times of known high risk, restrictions are placed on the opportunity to use lethal means for suicide (e.g. solitary armed duty), as distinct from imposing blanket restrictions on access (Mahon et al. 2005, McAllister et al. 2011). This model may have application in the veterinary profession: for example, restriction of unsupervised access to injectable barbiturates at times of high risk to minimise opportunity to remove them from the workplace for self- administration.

It is accepted standard practice for practising veterinary surgeons to store veterinary medicines, including injectable barbiturates,14 at home for on-call purposes. A potential access limitation strategy could involve the introduction into codes of

14 Most barbiturates used in veterinary medicine (e.g. pentobarbitone) are classified as Schedule 3 Controlled Drugs under the Misuse of Drugs Regulations 2001 (not quinalbarbitone – Schedule 2). A lockable bag is generally considered to be an acceptable locked receptacle for safe custody.

186 Chapter 6: Discussion

professional conduct, the requirement that injectable barbiturates can only be removed from the practice premises for the period that a veterinary surgeon is on- call and must be promptly returned to the practice afterwards. This would impede access to barbiturates for self-poisoning on impulse, in effect, buying time and potentially preventing an attempt, at least in the short term, until after a suicidal impulse has subsided. There is some support for this assumption. Interviews of suicide attempt survivors indicate that two-thirds of suicide attempts are contemplated for less than an hour beforehand (Williams and Wells 1991), with a recent study finding that almost 50% of patients reported an interval of no more than 10 minutes between their first current thought of suicide and their actual attempt (Diesenhammer et al. 2009). However, before any potential limitation strategy is introduced, it is important to explore the association between access to barbiturates and suicidal behaviour further. The qualitative studies outlined in Appendix 8 hold promise to be informative in this regard.

Suicide prevention by attempting to alter cultural attitudes within the profession towards the acceptability of using veterinary medicines as a means of suicide is likely to be more challenging than restricting access, but is nevertheless a potential approach worthy of consideration.

6.10.2.2.9 Signage

Placement of signage at geographical suicide ‘hot spots’, displaying details of accessible sources of emotional support, can be an effective suicide prevention measure (e.g. King and Frost 2005). Signage is a low cost intervention that capitalises on existing services, can be promptly implemented and is based on robust evidence of effectiveness. Signs encouraging individuals to seek help and displaying helpline telephone numbers, positioned in close proximity to practice medicine stores, might have a similar effect on individuals considering suicide and could also serve to increase awareness among all practice employees of sources of support (Figure 6.4).

Chapter 6: Discussion 187

Fig. 6.4: An example of signage (sticker)

The above signage was produced and distributed by the British Veterinary Association for display in 2010 following discussions with DJB.

Consideration should be given by veterinary employers and professional bodies to the development and implementation of policies to facilitate the reintegration of veterinary surgeons with respect and sensitivity back into the workplace following a mental health crisis.

Consideration should also be given to the development of a protocol for effective suicide postvention to provide guidance to veterinary employers on communication to employees of information relating to a colleague’s completed suicide and provision of support to all those affected.

188 Chapter 6: Discussion

6.10.2.2.10 Reporting of crisis mastery

It is well established that extensive and prurient reporting of suicide by the media can influence copycat acts – the so-called ‘Werther effect’15 (Phillips 1974) – and as a consequence, several countries have developed guidelines that promote responsible reporting of suicide (Pirkis 2009). However, a recent study has also described a ‘Papageno’16 protective effect of non-sensationalist media reporting, if the reports focus on individuals who demonstrate mastery of a crisis by adopting positive coping strategies in adverse circumstances (Niederkrotenthaler et al. 2010). The specific cause of death of veterinary surgeons is very rarely reported in the veterinary press so suicide reporting is unlikely to be contributing to the elevated suicide risk in this occupational group, but consideration should be given to inviting articles for publication from veterinary surgeons who adopted coping strategies other than suicidal behaviour in adverse circumstances.

6.11 Conclusions

The observations from recent research and the proposals for interventions described above have numerous implications for the veterinary profession, and ongoing advocacy will be required across a range of stakeholders within the profession to demonstrate the need for and encourage the development and implementation of suitable interventions. The establishment of a single collaborative body (a ‘veterinary support consortium’) with a mandate from key veterinary professional organisations to provide direction to, and coordination and integration of initiatives that attempt to improve psychological health within the veterinary profession could help to ensure a unified strategy and maximise effectiveness.

15 In honour of the character in J.W. von Goethe’s novel, The Sorrows of Young Werther (1774), who shot himself with a pistol after an ill-fated love-affair. Following its publication there was a series of reports of young men who took their own lives in the same way, which led to the book being banned by several authorities.

16 In honour of the character in Mozart’s opera,The Magic Flute (1791). When Papageno fears that he has lost his love, Papagena, he prepares to kill himself. But three boys save him at the last minute by reminding him of other alternatives to dying. Chapter 6: Discussion 189

Table 6.1: Proposed individual-oriented interventions with potential to help alleviate psychological distress among veterinary surgeons

Primary Secondary Tertiary

Enhanced assessments at Improving help-seeking Counselling and undergraduate level to inform the behaviour, including psychotherapy development, timing and targeting of encouragement to use educational interventions to improve existing support initiatives resilience during subsequent career Consider screening undergraduates to Support during identify at-risk individuals and provide complaints and appropriate levels of support disciplinary proceedings Mental health promotion: techniques to Provision of a service to improve mental well-being, enhance help resolve adaptive coping skills, increase employment-related awareness of the dangers of alcohol disputes misuse, and reduce perceived stigma Education to facilitate early recognition Litigation under of psychological problems in self and employment law others Consultation skills training to improve communication and thereby help to manage client expectations and handle complaints effectively Introduction of emotional reflection into the Professional Development Phase Increase awareness of existing support initiatives: Vet Helpline, Veterinary Surgeons’ Health Support Programme, Vetlife website Advice for veterinary surgeons considering alternative careers within or outside the profession

190 Chapter 6: Discussion

Table 6.2: Proposed organisation-oriented interventions with potential to help alleviate psychological distress among veterinary surgeons

Primary Secondary Tertiary

Regular population-level monitoring of Provision of accessible Policies to facilitate mental health and well-being of the and appropriate support rehabilitation and veterinary profession using valid and services, e.g. reintegration back into reliable instruments counselling, cognitive the workplace behavioural therapy, telephone helplines Regular monitoring of psychosocial Reduce discriminatory Consider controlling work characteristics of individual work-practices access to specific workplaces: iterative process of medicines measurement and modification Providing a safe means of recognising Adjustments to the work Signage positioned in and discussing concerns about environment to enable an close proximity to colleagues individual to continue medicine stores working Improving management skills for senior Intervention after a veterinary surgeons: better supervision suicide (postvention) skills, recognition of risk, team to support those leadership, regular staff appraisals, affected and reduce improved communication of change the risk of contagion Increase management support, including mentoring programmes to assist new graduates to gain confidence Increase participation in decision- making Changing work practices: reduction in working hours, use of dedicated providers of out-of-hours veterinary care, regular breaks, adequate clerical support Evaluation of the effectiveness of existing initiatives to support mental health Provide opportunities to enhance interpersonal relationships with colleagues and clients Opportunities for improved work-home interaction e.g. flexible work options Establish well-being as an essential value of the organisation to create the appropriate culture Restoration of confidence in RCVS complaints and disciplinary processes Reporting of crisis mastery in the veterinary press

191

Chapter 7:

Conclusions

192 Chapter 7: Conclusions

7.0 Towards meaningful measurement

Measurement requires the construction of an instrument for carrying out the practical process of measuring. Some variables, like height, can be measured directly and by relatively straightforward means. Other variables, like mental well-being, need to be approached indirectly through quantifying their manifestations. Indirectly measured variables are often called ‘latent’ variables to emphasise this fact.

Rating scales are constructed to measure latent variables. It is customary for a rating scale to consist of a set of items, each of which represents a different component of the latent variable. Every item is scored and item scores are summed to give a total score for each person. This value is a measure of the variable quantified by the set of items.

Whether a rating scale generates clinically meaningful and scientifically sound measurements depends on the decisions made during its construction and on its performance during testing. The decisions concern the components selected to form the set, their clinical grading and numerical scoring, and how components are combined to give a single overall value. Performance is tested against a number of predefined measurement (psychometric) criteria.

The original WEMWBS measures mental well-being by combining ratings of 14 components (feeling optimistic, useful, relaxed, interested in other people etc.) representing a wide conception of mental well-being including affective-emotional aspects, cognitive-evaluative dimensions and psychological functioning (Tennant et al. 2007). Scores for the 14 components are summed, without weighting, into a total WEMWBS score. Higher scores are purported to indicate higher levels of mental well-being. This process appears prima facie to be appropriate but requires empirical proof that it ‘works’ in the context of the population being measured. The psychometric properties of instruments are generally regarded as being population- specific (Meredith and Teresi 2006, Terwee et al. 2007). This means that evidence is required to support the choice of items forming the set, scoring of the individual items and appropriateness of combining item scores into a single score. Evidence is also required to support the validity and reliability of the item set as a measure of mental well-being. Traditional and modern psychometric methods provide formal frameworks for gathering and synthesising this evidence. Chapter 7: Conclusions 193

By virtue of the comprehensive evaluation of psychometric properties using both traditional and modern analytical methods applied to two large representative datasets, with validation of the results across datasets, the study has provided robust evidence to inform judgments concerning the suitability of WEMWBS (and a derived item set) as a meaningful measure of mental health and well-being in the UK veterinary profession.

7.1 Revisiting the study aims and objectives

The study fulfilled the objectives to examine comprehensively the measurement properties of WEMWBS in the context of the UK veterinary profession, by the complementary application of both traditional and modern psychometric methods.

The aim of these psychometric assessments was to inform judgements concerning:

[i] the defensibility of comparing WEMWBS scores for the veterinary profession with those for other population groups, and

7.1.1 Defensibility of comparing scores across other population groups

The WEMWBS score for the UK veterinary profession is reported to be statistically significantly lower than for general population samples (Bartram 2009a, Bartram et al. 2009b), indicating lower levels of mental well-being in this occupational group. However, evidence from the present study indicates that score comparisons for the 14-item WEMWBS across different population groups must be interpreted with great caution. For example, WEMWBS did not satisfy the strict unidimensionality

194 Chapter 7: Conclusions

expectations of the Rasch model, meaning that differences in WEMWBS scores might reflect differences in the level of a latent variable other than mental well-being. In addition, several items in the scale showed bias (DIF) for age and/or gender, and the items affected were not consistent with those reported following Rasch analysis for a UK general population sample (Stewart-Brown et al. 2009).

While there are limitations to the utility of the 14-item WEMWBS for comparisons across population groups, this research provided evidence to support the measurement invariance of the Rasch-derived 7-item interval scale, which is identical in item content to SWEMWBS (Stewart-Brown et al. 2009), across the UK general population and the veterinary profession. Use of this scale will enable valid score comparisons between samples of the respective populations, and the interval-level measurement properties enable any difference between scores to be interpreted meaningfully.17 In the event of any future scale modifications to accommodate additional components which the veterinary profession may consider to be related to mental well-being, the items comprising SWEMWBS should be retained to enable comparison of measurements with those for the general population.

7.1.2 Potential utility for research, including monitoring of trends

The study provided evidence to support the acceptability of data quality, targeting, external construct validity and internal consistency reliability of the 14-tem WEMWBS. However, scores obtained following administration of the instrument to a veterinary surgeon population sample have several limitations which increase the potential for misinference. The scores enable estimation of whether a subgroup of the population has a higher or lower level of mental well-being than another subgroup, and estimation of whether the level of mental well-being has increased or decreased over time. However, the scale is multidimensional so any difference between scores may be attributable to a second (or third) latent variable measured by the scale. Moreover, the scale is ordinal, not interval, so it is not possible to estimate by how much the level of mental well-being is higher or lower. In addition, several items in the scale showed bias (DIF) for age and/or gender so scores are influenced by differences in the demographic structure of the sample or subgroups of that sample.

17 Stewart-Brown et al. (2009) did not report summary descriptive statistics (mean, SD), calculated after transformation of SWEMWBS raw scores to interval measures, for their UK general population sample. Consequently, comparison with the measurement of mental well- being on the same scale for the veterinary profession samples in the present study is not yet possible. Chapter 7: Conclusions 195

The Rasch-derived 7-item scale offers superior measurement properties compared to the 14-item WEMWBS from which it is derived, and is consequently more suitable for research purposes. The ability to transform summed raw scores for the 7-item scale to interval-level measurements using the conversion table provided in Table 5.11, means that mental well-being can be quantified with greater precision across the continuum. In addition, parametric statistical analyses can be applied legitimately to the interval measures, given appropriate distributions. This means that the data can be explored with more sophisticated inferential procedures and greater statistical power – assessments of within-group changes and between-group differences are strengthened. The scale conforms to the rigorous unidimensionality expectations of the Rasch model so the scale measures a single latent variable, and the property of invariance means that the instrument works in the same way across subgroups within the population. The person-item threshold distribution for the scale in this research provided evidence that the range of level of mental well-being expressed by the item set is adequately targeted to the range of level of mental well-being within the veterinary profession. The brevity of the 7-item instrument is a further advantage if respondent burden is a concern or when there are space constraints in the questionnaire. The associations of interval measurements on the 7-item scale with HADS sub-scale scores and suicidal ideation provide evidence to support the external construct validity of the scale. These associations also indicate that the scale is clinically meaningful as an overall indicator of mental health and well-being in this occupational group – it is possible to estimate how much a one unit difference in the interval scale measurement is associated with the odds of anxiety or depression caseness or the odds of reporting having experienced suicidal thoughts in the previous 12 months. The person separation index (PSI) for the 7-item scale provided evidence to support its ability to discriminate between subgroups of respondents with different levels of mental well-being. This was consistent with the evidence from the test of group difference validity, which indicated the ability to discriminate between gender groups with a priori hypothesised differences in mental well-being.

The potential of the 7-item scale to detect change is indicated by the minimal item and total score floor and ceiling effects identified. However, the properties of test- retest reliability and responsiveness should be evaluated further within the context of the veterinary profession before the scales can be advocated as being appropriate for use as tools for the purpose of monitoring population level changes in the mental health and well-being of this occupational group or as outcome measures for intervention studies.

196 Chapter 7: Conclusions

7.2 A beginning rather than a conclusion

Recent research into the mental health and well-being of the UK veterinary profession has provided empirical insights into possible explanations for the elevated suicide risk in this occupational group. The levels of psychological distress reported suggest that ready access to and knowledge of lethal means is probably not operating in isolation to increase suicide risk within the profession (Bartram et al. 2009b). The research has raised awareness of mental health and well-being concerns within the profession and prompted several commendable initiatives with potential to improve the current situation, including: development and implementation of the RCVS Health Protocol18 (RCVS 2010a); emphasis in the accreditation requirements of the new RCVS Practice Standards Scheme19 (RCVS 2010b) that health and safety assessments must address both physical and psychological risks; the embedding of WEMWBS into the RCVS Survey of the Profession 2010 questionnaire; inclusion of mental health awareness in the professional key skills undergraduate curriculum of several veterinary schools and the compulsory modules of the postgraduate RCVS Certificate in Advanced Veterinary Practice; and an increased frequency of editorials and advertising in the veterinary press to raise awareness of existing sources of support such as Vet Helpline and the Veterinary Surgeons’ Health Support Programme.

This research provides evidence to support the acceptability of the measurement properties of a scale which can be used to examine differences between subgroups of the profession and, notwithstanding the need for prior evidence of the scale’s responsiveness to change, to monitor changes in mental well-being over time and assess the effectiveness of interventions. These contributions have the potential to advance the scope and quality of mental health and well-being research, and offer promise of the reward of informing effective targeted interventions which could make a difference to the everyday life of members of the veterinary profession, improve psychological health and attenuate suicide risk. The research community is

18 For vets among whom health-related problems may be a mitigating factor in complaints made against them which concern alleged professional misconduct, the RCVS Health Protocol provides a framework for effective rehabilitation and helps to prevent escalation of the complaint to the Disciplinary Committee. This is in the combined interest of the vet and the public whose animals are under his/her care.

19 UK health and safety legislation requires that employers must assess the nature and scale of physical and psychological risks and adjust the work environment accordingly. From the inception of the Practice Standards Scheme in 2005 until the introduction of the latest accreditation requirements in 2010, the four-yearly RCVS inspections checked practice compliance with legislative requirements for physical risks only. Chapter 7: Conclusions 197

encouraged to seize these opportunities; the veterinary profession must ensure the impact on mental health and well-being of research findings is maximised.

199

Appendices

201

Appendix 1

The Warwick-Edinburgh Mental Well-being Scale (WEMWBS)

203

The Warwick-Edinburgh Mental Well-being Scale (WEMWBS)

Below are some statements about feelings and thoughts. Please tick the box that best describes your experience of each over the last 2 weeks.

None Some All of Item Statement of the Rarely of the Often the time time time

I’ve been feeling optimistic 1 1 2 3 4 5 about the future

2 I’ve been feeling useful 1 2 3 4 5

3 I’ve been feeling relaxed 1 2 3 4 5

I’ve been feeling interested in 4 1 2 3 4 5 other people

5 I’ve had energy to spare 1 2 3 4 5

I’ve been dealing with problems 6 1 2 3 4 5 well

7 I’ve been thinking clearly 1 2 3 4 5

I’ve been feeling good about 8 1 2 3 4 5 myself I’ve been feeling close to other 9 1 2 3 4 5 people

10 I’ve been feeling confident 1 2 3 4 5

I’ve been able to make up my 11 1 2 3 4 5 own mind about things

12 I’ve been feeling loved 1 2 3 4 5

I’ve been interested in new 13 1 2 3 4 5 things

14 I’ve been feeling cheerful 1 2 3 4 5

The Short Warwick-Edinburgh Mental Well-being Scale (SWEMWBS) comprises the items in the shaded rows only.

205

Appendix 2

Survey 1 questionnaire

A cross-sectional study of mental health and well-being and their associations in the UK veterinary profession

207

David J. Bartram External supervisor: BVetMed MRCVS Dr David S. Baldwin 43 Leafy Lane MB BS DM FRCPsych Whiteley Reader in Psychiatry Fareham Division of Clinical Neurosciences Hants. PO15 7HL School of Medicine E-mail: [email protected] University of Southampton www,vetwellbeing.co.uk

October 2007

Dear Colleague

Survey of mental health and well-being in the UK veterinary profession

You are one of 3,200 veterinary surgeons in the UK selected at random and invited to participate in this survey; an independent, academic research initiative with external supervision through the School of Medicine, University of Southampton.

The study is prompted by the statistic that a disproportionately high number of veterinary surgeons die by suicide: one of the highest rates of any occupation.

The purpose of this survey is to identify some of the variables associated with work- related stress and inform the development of strategies to help improve mental health within the veterinary profession. The measured parameters will be compared against normative data for the UK population (and other professions where possible) and provide a baseline from which to monitor change and future trends. The results will be available to the British Veterinary Association and the Royal College of Veterinary Surgeons and submitted for publication in peer-reviewed veterinary and medical journals.

Participation is voluntary. You are encouraged to complete and return the questionnaire regardless of whether you have any past or present experience of stress or mental health problems. It takes around 20 minutes to complete. Your answers are needed to help ensure that the results are representative and can be generalised to the whole profession. A high response rate is critical to the success of the survey.

Your responses are anonymous. The source of individual data cannot be identified so confidentiality is assured. The study has been reviewed by an NHS Research Ethics Committee to help protect the interests of participants.

The survey-based research expertise of Veterinary Business Development Ltd. is used to distribute the questionnaire and receive the envelopes containing the responses. The study is otherwise completely independent of the company and is entirely non- commercial.

I hope that you will be able to assist in this key initiative and thank you in advance for your response. Please try to return the completed questionnaire as soon as possible, or before the end of November at the latest, in the enclosed reply-paid envelope. If you have any questions or would like to know more about the study please contact me, preferably by sending an e-mail to the following address: [email protected]

Yours faithfully

David J. Bartram BVetMed MRCVS

208

209

210

211

213

Appendix 3

Study synopsis – Survey 1

A cross-sectional study of mental health and well-being and their associations in the UK veterinary profession

215

A3.1 Background

The study was approved by Southampton and South West Hampshire Research Ethics Committee (B) in August 2007, Ref: 07/H0504/122. The work was covered by the public liability insurance of Veterinary Business Development Ltd, the organisation which provided use of the Vetfile® database as the sampling frame and mailed the questionnaires. Academic supervision for the study was provided by Professor D.S. Baldwin, Mental Health Group, Faculty of Medicine, University of Southampton and the initial work was reported and assessed for the award of Diploma of Fellowship of the Royal College of Veterinary Surgeons (Bartram 2009).

The overarching objective of this study was to explore the mental health and well- being of the UK veterinary profession to provide empirical evidence to assess their contribution to the elevated suicide risk and inform the development of strategies to improve psychological health and attenuate suicide risk.

The study aimed to estimate a range of psychological health parameters for the population of veterinary surgeons in the UK, investigate possible differences across demographic and occupational factors, compare population parameter values with published data for the general population and other normative groups, and propose potentially effective interventions.

The primary hypothesis was that the veterinary profession has higher levels of mental ill-health, lower levels of mental well-being and less favourable psychosocial work characteristics, when compared to the general population. The secondary hypothesis was that self-reported measures of mental ill-health, mental well-being and psychosocial work characteristics differ significantly across demographic and occupational factors within the profession.

A synopsis of the study follows. The methodology and results have been reported in detail elsewhere (Bartram 2009a, b, c; Bartram et al. 2009a, b, c; Bartram 2010; Bartram et al. 2010).

A3.2 Sample

A random stratified sample of 3200 veterinary surgeons practising in the UK was identified. This number represents approximately 20% of the membership of the Royal College of Veterinary Surgeons (RCVS), excluding those practising overseas

216

or retired. Veterinary surgeons listed in the sampling frame (Vetfile®, Veterinary Business Development Ltd) were stratified according to type of work within the profession and selected at random within each stratum in proportion to the number of veterinary surgeons in each type of work practising in the UK. The gender and decade of qualification profile of the entire study cohort was then compared with RCVS membership data for veterinary surgeons practising in the UK to assess the representativeness of the sample.

A3.3 Measures

The 4-page questionnaire comprised a total of 120 items which included:

• Background information consisting of brief demographic and occupational details, including age, sex, main type of work in the profession, position in the practice (if applicable), hours of work and hours on-call in a typical week. Several of these items and item options were sourced from RCVS Survey of the Profession 2006 (Robinson and Hooker 2006).

• The Hospital Anxiety and Depression Scale (HADS) (Zigmond and Snaith 1983).

• The Alcohol Use Disorders Identification Test alcohol consumption questions (AUDIT-C) (Bush et al. 1998).

• Three questions on suicidal ideation sourced from the second National Survey of Psychiatric Morbidity (Singleton et al. 2001).

• The Warwick-Edinburgh Mental Well-being Scale (WEMWBS) (Tennant et al. 2007).

• The Health and Safety Executive Management Standards Indicator Tool (HSE MSIT) (Cousins et al. 2004).

• Negative and positive work-home interaction (WHI_N and WHI_P) sub-scales of the SWING instrument (Geurts et al. 2005).

• A series of 27 original items focused specifically on potential sources of stress in the veterinary profession (one domain of 9 items referred to clinical work and were only completed by respondents to whom this domain was relevant). Examples include: client expectations; euthanasia of animals; dealing with client grief; administrative and clerical tasks. These were developed in collaboration 217

with informal focus groups and revised following pre- and pilot testing. Respondents scored on a 5-point Likert scale (0-4; from ‘not at all’ to ‘very much’) how much each item contributed to the stress they feel.

• An open question inviting respondents to identify in free text up to three main sources of pleasure and/or satisfaction in practice. Responses were grouped according to themes and a coding frame was developed to describe the thematic content.

A3.4 Treatment of missing data

Person mean score imputation was used to replace missing scale or sub-scale scores, provided that no more than one (HADS sub-scale; HSE MSIT sub-scales – except demands and control), two (demands and control sub-scales of HSE MSIT) or three (WEMWBS) values were missing. If greater than the specified number of values was missing then the scale or sub-scale was judged as invalid for that respondent and was not included in the analysis.

A3.5 Scope of statistical analysis

Statistical analysis was limited to within-sample comparison of outcome measures across a range of demographic and occupational factors, and comparison of estimated study population parameters of mental health and well-being with those reported for the general population. Examination of associations between measures on different scales and evaluation of the psychometric properties of the scales was outside the scope of this study.

A3.6 Results

Evaluable questionnaires were returned by 1796 participants, a response rate of 56.1%. The demographic and occupational profile of respondents was fairly representative of the UK veterinary profession.

The prevalence of anxiety, depression, and co-morbid anxiety and depression ‘probable cases’ (i.e. HADS sub-scale score ≥ 11) was 26.3% (95% CI: 24.3 to 28.4%), 5.8% (95% CI: 4.8 to 7.0%) and 4.5% (95% CI: 3.6 to 5.6%) respectively. The prevalence of HADS-A probable cases was 2.1 times higher, and the prevalence of HADS-D probable cases was 1.6 times higher than reported for the UK adult general population (HADS-A: 26.3%, 95% CI: 24.2 to 28.4%, vs. 12.6%,

218

95% CI: 11.2 to 14.2%, p<0.001; HADS-D: 5.8%, 95% CI: 4.7 to 6.9% vs. 3.6%, 95% CI: 2.7 to 4.5%, p=0.002) and the difference between the distribution of veterinary surgeons and the general population across HADS-A and HADS-D case categories was significant (p<0.001).

In comparison with alcohol consumption figures for the general population, veterinary surgeons in the sample were less likely to be non-drinkers (5.4%, 95% CI: 0.4 to 6.6% vs. 12.3%, 95% CI: 11.6 to 13.0%, p<0.001) and more likely to be ‘at- risk’ drinkers (62.6%, 95% CI: 60.3 to 64.8% vs. 47.7%, 95% CI: 46.6 to 48.8%, p<0.001). They drank more frequently than the general population, but consumed less on a typical drinking day and had a prevalence of daily and weekly binge- drinking that was similar to the general population.

The 12-month prevalence of suicidal thoughts was around 5.5 times higher than among the general population (21.3%, 95% CI: 21.0 to 23.6% vs. 3.9%, 95% CI: 3.4 to 4.3%, p<0.001).

The mean WEMWBS score was significantly lower than for the general population (48.85, 95% CI: 48.43 to 49.28 vs. 51.05, 95% CI: 50.51 to 51.59, p<0.001).

Veterinary surgeons self-reported significantly less favourable psychosocial work characteristics (higher risk of work-related stress) than the general population for the demands, managerial support and peer support stressor domains. The mean sub- scale scores for the sample of veterinary surgeons were in the following corresponding percentile scores for organisations in the general population sample: demands, 10th-25th percentile; control, 50th-75th percentile; managerial support, 5th- 10th percentile; peer support, 25th-50th percentile; relationships, 75th-90th percentile; role, 50th-75th percentile; change, 75th-90th percentile.

The sample of veterinary surgeons self-reported higher levels of negative work home interaction than the general population on the WHI_N scale (mean: 1.19, 95% CI: 1.16 to 1.21 vs. 0.86, 95% CI: 0.84 to 0.88, p<0.001) and higher levels of positive work home interaction than the general population on the WHI_P scale (mean: 0.97, 95% CI: 0.94 to 0.99 vs. 0.83, 95% CI: 0.80 to 0.86, p<0.001).

Number of hours worked; making professional mistakes; and the possibility of client complaints or litigation were the main reported contributors to stress. Good clinical outcomes; relationships with colleagues; and intellectual challenge/learning were the greatest sources of satisfaction. 219

Mental health and well-being were associated with a variety of demographic and occupational variables including age, gender, type of work, and employment status.

A3.7 Conclusions

Compared to the general population, veterinary surgeons were estimated to report higher levels of anxiety and depressive symptoms; higher 12-month prevalence of suicidal thoughts; less favourable psychosocial work characteristics in relation to demands and managerial support; lower levels of positive mental well-being; and higher levels of negative work-home interaction. They drank more frequently than the general population, but consumed less on a typical drinking day and the prevalence of daily and weekly binge-drinking was similar to that for the general population. The levels of psychological distress reported suggest ready access to and knowledge of lethal means is probably not operating in isolation to increase suicide risk within the profession.

221

Appendix 4

Survey 2 questionnaire

RCVS Survey of the Profession 2010

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

239

Appendix 5

Ethical approval and sponsorship – Survey 1

A cross-sectional study of mental health and well-being and their associations in the UK veterinary profession

241

242

243

245

Appendix 6

Ethical approval and sponsorship – Survey 2

Internal construct validity of the Warwick-Edinburgh Mental Well-being Scale (WEMWBS): a Rasch analysis using data from the RCVS Survey of the Profession 2010

247

248

249

Appendix 7

Further results of multiple regressions to estimate associations between scales administered in Survey 1

251

Table A7.1: Associations between scales and the estimated odds of at-risk drinking

[a] OR 95% CI p-value

Age 0.98 0.97 to 0.99 <0.001 Gender† 0.89 0.70 to 1.13 0.346 HADS-A 1.03 0.99 to 1.07 0.144 HADS-D 0.96 0.91 to 1.01 0.078 WEMWBS 0.99 0.97 to 1.02 0.525 Work characteristics Demands 0.75 0.63 to 0.90 0.002 Control 1.17 0.96 to 1.41 0.114 Management support 1.05 0.86 to 1.29 0.627 Peer support 1.19 0.94 to 1.52 0.144 Relationships 1.10 0.90 to 1.35 0.348 Role 0.93 0.75 to 1.15 0.475 Change 0.92 0.77 to 1.10 0.341 n=1449

[b] OR 95% CI p-value

Age 0.98 0.97 to 0.99 <0.001 Gender† 0.90 0.71 to 1.15 0.417 HADS-A 1.03 0.99 to 1.07 0.149 HADS-D 0.95 0.91 to 1.00 0.047 SWEMWBS 0.98 0.93 to 1.03 0.358 Work characteristics Demands 0.76 0.63 to 0.92 0.126 Control 1.16 0.96 to 1.41 0.511 Management support 1.07 0.87 to 1.31 0.222 Peer support 1.16 0.91 to 1.48 0.576 Relationships 1.06 0.86 to 1.30 0.654 Role 0.95 0.76 to 1.18 0.318 Change 0.91 0.76 to 1.09 0.001 n=1398

Multiple logistic regression [a] including 14-item WEMWBS as the independent variable, [b] including Rasch-derived 7-item interval scale (SWEMWBS) as the independent variable. AUDIT-C scale cut-off scores of ≥ 4 for women and ≥ 5 for men are used as an indicator of at-risk drinking (Gual et al. 2002). † Estimated odds for women relative to men. Guide to interpretation: [b] A one unit (year) increase in age is associated with a 2% reduction in the estimated odds of at-risk drinking (OR: 0.98, 95% CI: 0.97 to 0.99, p<0.001).

252

Table A7.2: Associations between scales and the estimated odds of reporting having experienced suicidal thoughts in the previous 12 months

[a] OR 95% CI p-value

Age 0.99 0.98 to 1.01 0.439 Gender† 0.77 0.56 to 1.04 0.092 AUDIT-C: low-risk§ 0.93 0.48 to 1.80 0.836 AUDIT-C: at-risk§ 0.91 0.48 to 1.72 0.769 HADS-A 1.15 1.09 to 1.20 <0.001 HADS-D 1.02 0.96 to 1.08 0.614 WEMWBS 0.93 0.90 to 0.96 <0.001 Work characteristics Demands 0.90 0.72 to 1.13 0.359 Control 0.98 0.77 to 1.24 0.841 Managerial support 1.23 0.95 to 1.59 0.117 Peer support 0.99 0.74 to 1.31 0.917 Relationships 0.78 0.63 to 0.99 0.045 Role 1.09 0.84 to 1.40 0.536 Change 0.91 0.73 to 1.14 0.409 n=1441

[b] OR 95% CI p-value

Age 1.00 0.99 to 1.02 0.590 Gender† 0.88 0.66 to 1.18 0.400 AUDIT-C: low-risk§ 0.97 0.52 to 1.80 0.921 AUDIT-C: at-risk§ 0.89 0.49 to 1.63 0.716 HADS-A 1.19 1.13 to 1.24 <0.001 HADS-D 1.05 1.00 to 1.11 0.070 SWEMWBS 0.87 0.81 to 0.92 <0.001 Work characteristics Demands 0.90 0.72 to 1.12 0.329 Control 0.98 0.78 to 1.23 0.856 Managerial support 1.12 0.89 to 1.42 0.367 Peer support 0.88 0.67 to 1.16 0.376 Relationships 0.85 0.67 to 1.07 0.164 Role 1.16 0.90 to 1.49 0.245 Change 0.91 0.74 to 1.13 0.407 n=1391

Multiple logistic regression [a] including 14-item WEMWBS as the independent variable, [b] including Rasch-derived 7-item interval scale (SWEMWBS) as the independent variable. † Estimated odds for women relative to men. § Estimated odds relative to non-drinkers. Guide to interpretation: [b] A one unit increase in score on the Rasch-derived 7-item interval scale (SWEMWBS) is associated with a 13% reduction in the estimated odds of reporting having experienced suicidal thoughts in the previous 12 months (OR: 0.87, 95% CI: 0.81 to 0.92, p<0.001). 253

Table A7.3: Associations between scales and WEMWBS or the Rasch-derived 7-item interval scale (SWEMWBS) score

[a] Coef. 95% CI p-value

Age -0.04 -0.06 to -0.01 0.012 Gender† -0.66 -1.25 to -0.07 0.028 AUDIT-C: low-risk§ 0.44 -0.79 to 1.68 0.482 AUDIT-C: at-risk§ 0.17 -1.02 to -1.36 0.776 HADS-A -0.61 -0.70 to -0.52 <0.001 HADS-D -1.22 -1.33 to -1.11 <0.001 Work characteristics Demands -0.35 -0.81 to 0.12 0.140 Control 1.12 0.66 to 1.58 <0.001 Managerial support 0.72 0.23 to 1.21 0.004 Peer support 1.26 0.68 to 1.84 <0.001 Relationships -0.80 -1.29 to -0.31 0.002 Role 1.34 0.82 to 1.86 <0.001 Change 0.00 -0.44 to 0.44 0.997 WHI_N: high‡ 0.57 -0.24 to 1.38 0.164 Suicidal ideation -1.62 -2.32 to -0.92 <0.001 n=1390

[b] Coef. 95% CI p-value

Age -0.02 -0.03 to -0.01 0.002 Gender† -0.21 -0.50 to 0.07 0.135 AUDIT-C: low-risk§ 0.15 -0.44 to 0.74 0.624 AUDIT-C: at-risk§ 0.00 -0.57 to 0.57 0.993 HADS-A -0.29 -0.33 to -0.24 <0.001 HADS-D -0.43 -0.48 to -0.38 <0.001 Work characteristics Demands -0.14 -0.36 to 0.08 0.218 Control 0.49 0.27 to 0.71 <0.001 Managerial support 0.31 0.08 to 0.54 0.010 Peer support 0.42 0.14 to 0.70 0.003 Relationships -0.28 -0.51 to -0.04 0.021 Role 0.81 0.5 to 1.06 <0.001 Change 0.05 -0.16 to 0.26 0.649 WHI_N: high‡ 0.33 -0.14 to 0.78 0.164 Suicidal ideation -0.40 -0.77 to -0.01 0.042 n=1390

Multiple linear regression with [a] 14-item WEMWBS as the dependent variable, [b] Rasch- derived 7-item interval scale (SWEMWBS) as the dependent variable. † Estimated coefficient for women relative to men; § Estimated coefficient relative to non-drinkers; ‡ Estimated coefficient for respondents with high WHI_N (defined as a score of > 1 SD above the mean score on the WHI_N scale, i.e. > 1.76) relative to all other respondents. Guide to interpretation: [b] A one unit increase in HADS-D score is associated with an estimated 0.43 unit reduction in the score on the Rasch-derived 7-item interval scale (SWEMWBS) (Coef. -0.43, 95% CI: -0.48 to -0.38, p<0.001).

254

Table A7.4: Associations between scales and the estimated odds of HADS-A possible or probable caseness (score ≥ 8)

[a] OR 95% CI p-value

Age 0.99 0.98 to 1.00 0.146 Gender† 1.26 0.95 to 1.67 0.110 AUDIT-C: low-risk§ 0.99 0.52 to 1.88 0.971 AUDIT-C: at-risk§ 1.15 0.62 to 2.13 0.667 WEMWBS 0.85 0.83 to 0.87 <0.001 Work characteristics Demands 0.60 0.48 to 0.74 <0.001 Control 0.73 0.58 to 0.92 0.007 Managerial support 1.11 0.87 to 1.42 0.392 Peer support 1.49 1.10 to 2.01 0.010 Relationships 0.51 0.40 to 0.66 <0.001 Role 0.90 0.69 to 1.17 0.421 Change 1.17 0.95 to 1.45 0.147 n=1449

[b] OR 95% CI p-value

Age 0.99 0.97 to 1.00 0.086 Gender† 1.27 0.96 to 1.69 0.093 AUDIT-C: low-risk§ 0.95 0.51 to 1.79 0.886 AUDIT-C: at-risk§ 1.09 0.59 to 1.99 0.791 SWEMWBS 0.68 0.64 to 0.71 <0.001 Work characteristics Demands 0.58 0.47 to 0.72 <0.001 Control 0.71 0.57 to 0.90 0.004 Managerial support 1.11 0.87 to 1.41 0.410 Peer support 1.32 0.99 to 1.78 0.063 Relationships 0.53 0.41 to 0.68 <0.001 Role 1.00 0.77 to 1.30 0.991 Change 1.18 0.95 to 1.45 0.129 n=1449

Multiple logistic regression [a] including 14-item WEMWBS as the independent variable, [b] including Rasch-derived 7-item interval scale (SWEMWBS) as the independent variable. † Estimated odds for women relative to men. § Estimated odds relative to non-drinkers. Guide to interpretation: [b] A one unit increase in score on the Rasch-derived 7-item interval scale (SWEMWBS) is associated with a 32% reduction in the estimated odds of HADS-A possible or probable caseness (OR: 0.68, 95% CI: 0.64 to 0.71, p<0.001).

255

Table A7.5: Associations between scales and the estimated odds of HADS-A probable caseness (score ≥ 11)

[a] OR 95% CI p-value

Age 0.99 0.98 to 1.01 0.147 Gender† 1.43 1.04 to 1.96 0.027 AUDIT-C: low-risk§ 0.83 0.42 to 1.64 0.596 AUDIT-C: at-risk§ 0.98 0.51 to 1.89 0.956 WEMWBS 0.85 0.83 to 0.87 <0.001 Work characteristics Demands 0.55 0.44 to 0.70 <0.001 Control 0.89 0.70 to 1.13 0.333 Managerial support 1.07 0.83 to 1.40 0.592 Peer support 1.18 0.87 to 1.58 0.288 Relationships 0.59 0.47 to 0.76 <0.001 Role 1.03 0.79 to 1.34 0.816 Change 1.02 0.81 to 1.27 0.888 n=1449

[b] OR 95% CI p-value

Age 0.99 0.98 to 1.01 0.325 Gender† 1.44 1.05 to 1.97 0.023 AUDIT-C: low-risk§ 0.75 0.38 to 1.46 0.396 AUDIT-C: at-risk§ 0.87 0.46 to 1.66 0.680 SWEMWBS 0.66 0.62 to 0.70 <0.001 Work characteristics Demands 0.54 0.43 to 0.68 <0.001 Control 0.87 0.68 to 1.10 0.235 Managerial support 1.05 0.81 to 1.36 0.709 Peer support 1.06 0.99 to 1.78 0.063 Relationships 0.63 0.49 to 0.80 <0.001 Role 1.11 0.85 to 1.44 0.446 Change 1.02 0.82 to 1.28 0.842 n=1449

Multiple logistic regression [a] including 14-item WEMWBS as the independent variable, [b] including Rasch-derived 7-item interval scale (SWEMWBS) as the independent variable. † Estimated odds for women relative to men. § Estimated odds relative to non-drinkers. Guide to interpretation: [b] A one unit increase in score on the Rasch-derived 7-item interval scale (SWEMWBS) is associated with a 34% reduction in the estimated odds of HADS-A probable caseness (OR: 0.66, 95% CI: 0.62 to 0.70, p<0.001).

256

Table A7.6: Associations between scales and the estimated odds of HADS-D possible or probable caseness (score ≥ 8)

[a] OR 95% CI p-value

Age 1.02 1.00 to 1.04 0.092 Gender† 0.82 0.56 to 1.21 0.326 AUDIT-C: low-risk§ 2.46 1.04 to 5.82 0.040 AUDIT-C: at-risk§ 2.46 1.06 to 5.68 0.035 WEMWBS 0.78 0.76 to 0.81 <0.001 Work characteristics Demands 0.88 0.67 to 1.16 0.380 Control 0.97 0.73 to 1.28 0.811 Managerial support 0.88 0.65 to 1.19 0.399 Peer support 0.83 0.59 to 1.16 0.277 Relationships 0.81 0.61 to 1.07 0.136 Role 0.97 0.72 to 1.31 0.823 Change 1.19 0.91 to 1.55 0.205 n=1449

[b] OR 95% CI p-value

Age 1.01 1.00 to 1.03 0.138 Gender† 0.88 0.61 to 1.30 0.493 AUDIT-C: low-risk§ 1.88 0.82 to 4.33 0.136 AUDIT-C: at-risk§ 1.87 0.83 to 4.20 0.130 SWEMWBS 0.56 0.52 to 0.61 <0.001 Work characteristics Demands 0.84 0.65 to 1.09 0.194 Control 0.90 0.69 to 1.18 0.446 Managerial support 0.82 0.61 to 1.11 0.195 Peer support 0.74 0.54 to 1.02 0.069 Relationships 0.87 0.66 to 1.14 0.310 Role 1.03 0.78 to 1.38 0.818 Change 1.19 0.92 to 1.54 0.174 n=1449

Multiple logistic regression [a] including 14-item WEMWBS as the independent variable, [b] including Rasch-derived 7-item interval scale (SWEMWBS) as the independent variable. † Estimated odds for women relative to men. § Estimated odds relative to non-drinkers. Guide to interpretation: [b] A one unit increase in score on the Rasch-derived 7-item interval scale (SWEMWBS) is associated with a 44% reduction in the estimated odds of HADS-D possible or probable caseness (OR: 0.56, 95% CI: 0.52 to 0.61, p<0.001). 257

Table A7.7: Associations between scales and the estimated odds of HADS-D probable caseness (score ≥ 11)

[a] OR 95% CI p-value

Age 0.97 0.93 to 1.00 0.060 Gender† 0.53 0.28 to 1.00 0.051 AUDIT-C: low-risk§ 1.08 0.30 to 3.91 0.908 AUDIT-C: at-risk§ 1.11 0.32 to 3.78 0.870 WEMWBS 0.75 0.71 to 0.80 <0.001 Work characteristics Demands 0.40 0.26 to 0.63 <0.001 Control 1.14 0.73 to 1.76 0.567 Managerial support 0.82 0.50 to 1.33 0.419 Peer support 0.92 0.56 to 1.51 0.744 Relationships 1.22 0.78 to 1.91 0.388 Role 1.21 0.77 to 1.92 0.415 Change 0.98 0.65 to 1.47 0.906 n=1449

[b] OR 95% CI p-value

Age 0.97 0.94 to 1.00 0.088 Gender† 0.66 0.37 to 1.20 0.177 AUDIT-C: low-risk§ 0.98 0.27 to 3.52 0.970 AUDIT-C: at-risk§ 1.06 0.31 to 3.61 0.924 SWEMWBS 0.48 0.41 to 0.56 <0.001 Work characteristics Demands 0.37 0.24 to 0.57 <0.001 Control 1.05 0.70 to 1.58 0.811 Managerial support 0.78 0.49 to 1.25 0.297 Peer support 0.83 0.52 to 1.33 0.442 Relationships 1.31 0.68 to 2.28 0.581 Role 1.28 0.77 to 1.99 0.625 Change 1.02 0.68 to 1.50 0.978 n=1449

Multiple logistic regression [a] including 14-item WEMWBS as the independent variable, [b] including Rasch-derived 7-item interval scale (SWEMWBS) as the independent variable. † Estimated odds for women relative to men. § Estimated odds relative to non-drinkers. Guide to interpretation: [b] A one unit increase in score on the Rasch-derived 7-item interval scale (SWEMWBS) is associated with a 52% reduction in the estimated odds of HADS-D probable caseness (OR: 0.48, 95% CI: 0.41 to 0.56, p<0.001).

Table A7.8: Pearson correlation coefficients between HADS, WEMWBS and HSE MSIT psychosocial work characteristics sub-scales

Managerial Peer Age HADS-A HADS-D WEMWBS Demands Control Relationships Role Change support support r r r r r r r r r r r

Age 1.00

HADS-A -0.19 1.00 p-value <0.001 HADS-D -0.02 0.63 1.00 p-value 1.000 <0.001 WEMWBS 0.10 -0.69 -0.76 1.00 p-value 0.003 <0.001 <0.001 Demands 0.07 -0.38 -0.34 0.32 1.00 p-value 0.253 <0.001 <0.001 <0.001 Control 0.39 -0.41 -0.34 0.45 0.34 1.00 p-value <0.001 <0.001 <0.001 <0.001 <0.001 Managerial support 0.05 -0.36 -0.40 0.48 0.37 0.45 1.00 p-value 1.000 <0.001 <0.001 <0.001 <0.001 <0.001 Peer support 0.03 -0.38 -0.44 0.50 0.38 0.42 0.72 1.00 p-value 1.000 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 Relationships 0.15 -0.39 -0.34 0.37 0.35 0.39 0.48 0.57 1.00 p-value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 Role 0.20 -0.37 -0.33 0.45 0.22 0.43 0.47 0.46 0.40 1.00 p-value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 Change 0.25 -0.34 -0.31 0.41 0.26 0.55 0.67 0.53 0.48 0.56 1.00 p-value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001

For ease of identification, significant negative correlations are highlighted: ; significant positive correlations are highlighted:

Table A7.9: Pearson correlation coefficients between HADS, the Rasch-derived 7-item interval scale (SWEMWBS) and HSE MSIT psychosocial work characteristics sub-scales

Managerial Peer Age HADS-A HADS-D WEMWBS Demands Control Relationships Role Change support support r r r r r r r r r r r

Age 1.00

HADS-A -0.19 1.00 p-value <0.001 HADS-D -0.02 0.63 1.00 p-value 0.351 <0.001 SWEMWBS 0.09 -0.66 -0.69 1.00 p-value <0.001 <0.001 <0.001 Demands 0.07 -0.39 -0.34 0.30 1.00 p-value 0.005 <0.001 <0.001 <0.001 Control 0.39 -0.41 -0.33 0.43 0.34 1.00 p-value <0.001 <0.001 <0.001 <0.001 <0.001 Managerial support 0.05 -0.36 -0.40 0.46 0.37 0.45 1.00 p-value 0.049 <0.001 <0.001 <0.001 <0.001 <0.001 Peer support 0.03 -0.38 -0.44 0.48 0.38 0.42 0.72 1.00 p-value 0.285 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 Relationships 0.15 -0.39 -0.34 0.36 0.35 0.39 0.48 0.57 1.00 p-value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 Role 0.20 -0.37 -0.33 0.46 0.22 0.43 0.47 0.46 0.40 1.00 p-value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 Change 0.25 -0.34 -0.31 0.41 0.26 0.55 0.67 0.53 0.48 0.56 1.00 p-value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001

For ease of identification, significant negative correlations are highlighted: ; significant positive correlations are highlighted: 259

261

Appendix 8

Synopsis of qualitative studies

Understanding suicidal thoughts and help-seeking behaviour among UK veterinary surgeons: a semi-structured interview study

Qualitative analysis of Coroners’ data for veterinary surgeon suicides within a region of England

263

A8.1 Understanding suicidal thoughts and help-seeking behaviour among UK veterinary surgeons: a semi-structured interview study

This study seeks to explore by semi-structured telephone interview the causal attribution of suicidal thoughts and help-seeking behaviour for emotional problems in 109 respondents to Survey 1 (Bartram et al. 2009a, b, c) who reported a positive response to three questions concerning suicidal ideation (see Figure 4.4) and provided their names and contact details in order that they might be considered to participate in a possible subsequent interview.

The University of Southampton Faculty of Medicine Ethics Committee granted full approval to the protocol and related documentation for the study in January 2009 (SOMSEC026.09), and the University Research Governance Office confirmed that the University was prepared to act as Research Sponsor in February 2009 (RGO ref 6194). Telephone interviews commenced in February 2009 and were completed in February 2010. Each interview was of approximately 40 minutes in duration. A summary of the interview study sample and attrition is displayed in Figure A8.1.

Fig. A8.1: Interview study sample and attrition

Purposive sample n=109

No recollection of Not willing to previous suicidal participate thoughts n=2 n=18

No response/not Established contact contactable but no interview n=3 arranged n=2

Completed interview n=84

264

Hand-written field notes have been recorded for each interview. Anonymised transcripts will be prepared and imported as documents into NVivo8 software to facilitate analysis. 265

A8.2 Qualitative analysis of Coroners’ data for veterinary surgeon suicides within a region of England

The study seeks to explore themes in the transcripts of coroners’ reports for deaths of veterinary surgeons which received a verdict of suicide. Twelve individuals who died 1989 - 2002 within a defined geographical area were identified from the records of the Wessex Suicide Audit (King 2001). The eight coroners concerned provided written consent for the files to be accessed for the purpose of the study.

The School of Medicine Ethics Committee granted full approval to the protocol and related documentation for the study in January 2010 (SOMSEC051.10), and the University Research Governance Office confirmed that the University was prepared to act as Research Sponsor in March 2010 (RGO ref 7093).

This study commenced in September 2010 and data collection was completed in December 2010. Files were examined for 11 deaths in total; one file could not be located by the coroner’s office (presumed destroyed).

Comprehensive hand-written field notes have been recorded for each file. Anonymised transcripts will be prepared and imported as documents into NVivo8 software to facilitate analysis.

267

Appendix 9

Related publications

BARTRAM, D.J. & BALDWIN, D.S. (2010) Veterinary surgeons and suicide: a structured review of possible influences on increased risk. Veterinary Record 166, 388-397

BARTRAM, D.J., SINCLAIR, J.M.A. & BALDWIN, D.S. (2010) Interventions with potential to improve the mental health and well-being of UK veterinary surgeons. Veterinary Record 166, 518-523

BARTRAM, D.J., SINCLAIR, J.M. & BALDWIN, D.S. (2012) Further validation of the Warwick-Edinburgh Mental Well-being Scale (WEMWBS) in the UK veterinary profession: Rasch analysis. Quality of Life Research doi: 10.1007/s11136-012-0144- 4

BARTRAM, D.J., YADEGARFAR, G. & BALDWIN, D.S. (2008) Reported alcohol consumption, depressive and anxiety symptoms, and mental well-being among UK veterinary surgeons: cross-sectional questionnaire survey. Journal of Psychopharmacology 22 (5 Abstract Suppl.), A30

BARTRAM, D.J., YADEGARFAR, G. & BALDWIN, D.S. (2009) A cross-sectional study of mental heath and well-being and their associations in the UK veterinary profession. Social Psychiatry and Psychiatric Epidemiology 44, 1075-1085

BARTRAM, D.J., YADEGARFAR, G., SINCLAIR, J.M.A. & BALDWIN, D.S. (2011) Validation of the Warwick-Edinburgh Mental Well-being Scale (WEMWBS) as an overall indicator of population mental health and well-being in the UK veterinary profession. The Veterinary Journal 187, 397-398

Papers Papers Papers Veterinary surgeons and suicide: a structured review of possible influences on increased risk

D. J. Bartram, D. S. Baldwin

Veterinary surgeons are known to be at a higher risk of suicide compared with the general population. There has been much speculation regarding possible mechanisms underlying the increased suicide risk in the profession, but little empirical research. A computerised search of published literature on the suicide risk and influences on suicide among veterinarians, with comparison to the risk and influences in other occupational groups and in the general population, was used to develop a structured review. Veterinary surgeons have a proportional mortality ratio (PMR) for suicide approximately four times that of the general population and around twice that of other healthcare professions. A complex interaction of possible mechanisms may occur across the course of a veterinary career to increase the risk of suicide. Possible factors include the characteristics of individuals entering the profession, negative effects during undergraduate training, work-related stressors, ready access to and knowledge of means, stigma associated with mental illness, professional and social isolation, and alcohol or drug misuse (mainly prescription drugs to which the profession has ready access). Contextual effects such as attitudes to death and euthanasia, formed through the profession’s routine involvement with euthanasia of companion animals and slaughter of farm animals, and suicide ‘contagion’ due to direct or indirect exposure to suicide of peers within this small profession are other possible influences.

MEMBERS of some occupational groups are at a greatly increased risk tors are taken into account (Charlton 1995, Stack 2001, Agerbo and of suicide (Agerbo and others 2007). An elevated risk has been report- others 2007). ed in healthcare professionals, including doctors (Hawton and oth- The risk factors for suicide include depression, alcohol and drug ers 2001, Schernhammer and Colditz 2004, Torre and others 2005), abuse, certain personality traits, and environmental factors such as pharmacists (Kelly and Bunting 1998), dentists (Alexander 2001) and chronic major difficulties and undesirable life events (Goldney 2005). nurses (Hawton and Vislisel 1999). Farmers are also at increased risk An interplay between various potentially malign influences has been (Malmberg and others 1999). suggested for the veterinary profession. The interrelations between The absolute number of suicides by veterinary surgeons is low, work, personality and mental health are well documented (Stansfeld due to the small size of the profession (approximately 16,000 veteri- 2002), but reports specific to the veterinary profession (Halliwell and nary surgeons practising in the UK) (RCVS 2007), but the profession Hoskin 2005) have tended to simply present the observations and is at a higher risk of suicide when compared with other occupations opinions of concerned individuals. and the general population. For example, a survey of the causes of As such, it is uncertain whether the increased risk of suicide mortality among male veterinarians resident in Britain followed derives from the characteristics of individuals entering the profession, up from between 1949 and 1953 until 1975 reported a twofold the nature of the work environment, or other factors. The authors increase in deaths from suicide compared with all men in the same review published data on suicide in the veterinary profession, other occupational social class (Kinlen 1983). While most differences in healthcare professions and the general population, and summarise the suicide risk between occupations are accounted for by differences in reported possible influences on the increased risk. income and employment status, the most striking exceptions are for veterinarians, doctors, nurses and pharmacists, all of whom have a Materials and methods significantly higher risk of suicide, even when demographic fac- Literature search Papers relevant to mental health among veterinary surgeons were identified by searching PubMed (which includes MEDLINE) and Google Scholar for articles published after 1970. The search terms Veterinary Record (2010) 166, 388-397 doi: 10.1136/vr.b4794 included, but were not limited to, veterinary and depression, suicide, stress, burnout, distress, abuse, alcohol drinking, substance-related D. J. Bartram, BVetMed, DipM, E-mail for correspondence: disorders, psychosocial working conditions, coping or psychiatry. MCIM, CDipAF, FRCVS, [email protected] Additional articles were identified by scrutinising the reference D. S. Baldwin, MB, BS, DM, lists of relevant articles, hand searching of conference proceedings, FRCPsych, Provenance: not commissioned; examining regular automatic customised e-mail alerts for relevant Division of Clinical Neurosciences, externally peer reviewed new articles in journals (Ovid Autoalert), and by consulting experts. School of Medicine, University of The search strategy excluded non-English-language records. Articles Southampton, RSH Hospital, Brintons were included as appropriate to provide a comprehensive, structured Terrace, Southampton SO14 0YG review of the evidence of increased suicide risk among veterinary

Veterinary Record | March 27, 2010

Online P&A March 27.indd 388 24/3/10 15:41:57 Papers Papers

TABLE 1: Reported proportional mortality rates (PMRs) for suicide among male and female Risk of suicide in veterinary veterinary surgeons in the UK surgeons United Kingdom Reference Study period Region Age range covered PMR 95% CI On the basis of PMRs in England and Males Wales (Charlton and others 1993, Kelly Charlton and others (1993) 1979-1990 England and Wales 16-64 364 NR and others 1995, Kelly and Bunting 1998, Kelly and Bunting (1998) 1982-1987 England and Wales 20-64 349 203-559 Mellanby 2005, Meltzer and others 2008) Kelly and Bunting (1998) 1991-1996 England and Wales 20-64 324 148-615 Mellanby (2005) 1979-1990 England and Wales 20-74 361 252-503 and Scotland (Stark and others 2006), vet- Mellanby (2005) 1991-2000 England and Wales 20-74 374 244-548 erinarians appear to be at particularly high Stark and others (2006) 1981-1999 Scotland 16-45 293 80-749* risk of suicide. In this occupational group Stark and others (2006) 1981-1999 Scotland 46-64 301 36-1088* the chances of a death being due to sui- Females cide are approximately four times that of Kelly and Bunting (1998) 1991-1996 England and Wales 20-59 500 136-1279 Mellanby (2005) 1979-1990 England and Wales 20-74 414 166-853 the general population, and around twice Mellanby (2005) 1991-2000 England and Wales 20-74 1240 446-2710 that of other healthcare professionals. The Meltzer and others (2008) 2001-2005 England and Wales 20-64 609 198-1422 PMRs for veterinary surgeons are consistently among the highest of all occupations. * When the 95 per cent confidence interval (CI) includes 100, the difference between the PMR for an occupation and the general population is not statistically significant (P<0.05) Reported PMRs for male and female vet- NR Not reported erinarians are shown in Table 1. Suicides by veterinarians do not appear to be confined to a restricted age range (Mellanby 2005, TABLE 2: Relative risk (RR) of suicide (relative to deaths from other causes) in high-risk Stark and others 2006), as the distribution occupational groups for men and women in England and Wales in 1990 to 1992, compared of suicides by age group is not substantially with reference groups (Charlton 1995)† different from that of all occupations com- Men aged 16-44 Men aged 45-64 Women aged 16-64‡ bined (Kelly and others 1995). RR 95% CI RR 95% CI RR 95% CI There are many difficulties when com- Veterinarians 4.61 1.49-14.25* 5.62 1.60-19.74* 7.62 1.04-55.94* paring the risk of suicide across occupational Pharmacists 1.15 0.37-3.52 4.15 2.00-8.58** 1.21 0.27-5.35 groups. These include the effects of different Dentists 2.26 0.93-5.47 5.19 2.29-11.76** – – sociodemographic factors, both between Farmers 0.88 0.60-1.30 1.93 1.48-2.51** – – occupations and within occupational spe- Medical practitioners 1.50 0.90-2.50 2.22 1.35-3.65* 4.54 2.54-8.13** cialisms, and potential confounding by * P<0.01 these factors should be adequately control- ** P<0.001 led (Wilhelm and others 2004). Charlton † Reference groups are married; UK-born; in the corresponding age range; not in the 10 occupational groups with (1995) linked data from death certificates the highest proportional mortality ratio for suicide; in urban wards with owner occupancy of at least 85 per cent, unemployment below 5 per cent, less than 6.5 per cent of occupants changing address per year and less than 9 per cent of and the characteristics of the electoral ward adults below pensionable age living alone of the deceased individuals’ usual resi- ‡ There were no significant model differences for women aged under and over 45 dence to undertake a case-control analysis CI Confidence interval of suicide risk. The risk of suicide relative to death from natural causes, among male veterinarians aged 16 to 44 years and 45 surgeons and possible influences on the increased risk, including to 64 years, and female veterinarians aged 16 to 64 years, is elevated evidence from other occupational groups, rather than a systematic by 4.6, 5.6 and 7.6 times, respectively, in comparison with individu- review with meta-analysis. als in the general population with similar demographic characteristics (Table 2). Interpretation of data The standard method used to make comparisons between the death Rest of the world rates by suicide in different occupations and the general population A study of mortality patterns among veterinarians in the USA from is by calculation of the proportional mortality ratio (PMR). PMRs 1947 to 1977 showed significantly elevated mortality from suicide as compare the proportion of deaths in an occupation from a specific opposed to other causes for males (1.7 times), particularly for those in cause to the proportion of deaths from the same cause in the gen- small animal practice (3.6 times), and with a higher proportion of self- eral population. A PMR for suicide of 200 indicates that twice the poisonings compared with the general population (Blair and Hayes expected proportion of deaths by suicide was recorded in that occu- 1980, 1982). Barbiturates were the means most commonly used for pational group (Kelly and Bunting 1998). While PMR is a widely suicide. In a separate study in California, male and female veterinari- used measure, it is affected by the relative frequency of other causes ans had significantly elevated mortality from suicide – respectively, 2.5 of death; an increased PMR can indicate lower mortality from other times and 5.9 times higher than the general population of California. causes as well as higher mortality from the cause being examined. The mortality from suicide was significantly higher among veterinar- The high PMR for suicide identified for some occupational groups ians who had worked in the profession for less than 30 years (Miller may reflect the fact that the overall mortality is low, and therefore and Beaumont 1995). the proportion of deaths from suicide is high relative to other causes In a large study in Norway, which examined suicide deaths by (Charlton 1995, Kelly and others 1995, Kelly and Bunting 1998, occupation over a 40-year period (Hem and others 2005), the highest Hawton 2001). Moreover, lower mortality from other causes may rate was among male veterinarians, with a suicide rate (44 suicides also explain the apparent paradox between high PMRs for suicide per 100,000 person years, 95 per cent confidence interval [CI] 25 to identified for some high-social-class occupational groups and robust 75) approaching twice that of the general population. Using a similar evidence of an inverse relationship between occupational social class method, a suicide rate among veterinarians of 45 per 100,000 person and risk of suicide (the lower the class, the higher the risk) (Platt and years (95 per cent CI 25 to 82) was reported for two Australian states Hawton 2000). combined, approximately four times the rate in the general population for the states concerned (Fairnie 2005, Jones-Fairnie and others 2008). International comparisons In that study, the principal method of suicide by veterinarians was self- International comparisons of suicide risk by occupation are hindered poisoning by drugs, mainly injectable barbiturates, and the suicides by variation between countries in the classification and recording of were not confined to a restricted age range; the age distribution of suicides (Andriessen 2006) and the use of a range of different measures suicides for veterinarians was not substantially different from that of of suicide risk across the literature. the general population of Australia.

March 27, 2010 | Veterinary Record

Online P&A March 27.indd 389 24/3/10 15:41:57 Papers Papers

TABLE 3: Percentage* of suicides by method for veterinarians and combined suicides for all medicines for self-poisoning, possible con- occupations, for males aged 20 to 64 years and for females aged 20 to 59 years, in England tributory factors for their high suicide risk. and Wales during 1982 to 1996 (Kelly and Bunting 1998) In addition, veterinarians are supervised less in their use of medicines than doctors are Method of suicide Poisoning with solid or Poisoning with gases Hanging and Firearms and (Fishbain 1986). liquid substance or other vapours suffocation Drowning explosives Other Deliberate self-poisoning is the most Male veterinarians 76 3 5 0 16 0 common method of suicide in both male All men 20 27 27 6 5 16 and female veterinarians, accounting for Female veterinarians 89 11 0 0 – 0 76 and 89 per cent of suicides, respectively, All women 46 10 17 9 – 18 compared with 20 and 46 per cent, respec- * Percentages may not add up to 100 due to rounding tively, in the general population (Kelly and Bunting 1998) (Table 3). Veterinarians and pharmacists have the highest proportions of suicides using this method for all occupa- Conceptual models of suicidal behaviour tional groups; medical practitioners also have an increased risk of this Many explanatory and predictive models of suicidal ideation and specific method of suicide (Kelly and Bunting 1998, Hawton and oth- behaviour have been proposed; these have included sociological, psy- ers 2000, Agerbo and others 2007). There are many reports of the use chiatric, biological and psychological explanations. Three models of drugs for veterinary anaesthesia and euthanasia as suicide agents are described in outline here, in order to offer a theoretical context in (for example, Cordell and others 1986, Sterken and others 2004). which to understand the mechanisms that may underlie the possible Barbiturates are the drugs most commonly used for suicide by doctors influences on suicide risk that occur across the course of a veterinary (Hawton and others 2000) and were used by at least half of male veter- career. The stress-diathesis model (Mann and others 1999) posits that inarians who completed suicide by deliberate self-poisoning between suicidal behaviour results from the interaction between stressful life 1982 and 1996 in England and Wales (Kelly and others 1995). Agerbo events and an individual predisposition or vulnerability. This vulner- and others (2007) showed that suicide rates among doctors are slightly ability, itself the product of psychobiological factors, genetic predis- elevated even when deaths by medicines are excluded, suggesting that position and past life events, influences how the individual perceives, the ready availability of lethal means is not the only factor increas- interprets and reacts to adverse life events. It is a dynamic system ing their occupational suicide risk (Reichenberg and MacCabe 2007). in which stress and diathesis influence each other (van Heeringen Firearms are the second most common method of suicide by male vet- 2002). Consistent with the stress-diathesis model, the ‘Cry of Pain’ erinarians, and the frequency of this method is also raised relative to model of suicidal behaviour (Williams and Pollock 2000, Williams the general population, accounting for 16 per cent of suicides among 2001) conceptualises suicidal behaviour as the response (the ‘cry’) to male veterinarians and 5 per cent among the general population (Kelly a stressful situation in which environmental cues result in feelings and Bunting 1998). Veterinarians working in equine, farm animal and of defeat, loss or humiliation that give rise to an overwhelming feel- mixed practice may have ready access to firearms for euthanasia of ing of needing to escape, a sense of being unable to escape (entrap- large animals. ment) and a sense that this state of affairs will continue indefinitely (no prospect of rescue). Judgements regarding perceptions of defeat, Attitudes to death and euthanasia entrapment and rescue are determined, at least in part, by psycho- Veterinary surgeons are often asked to end the lives of animals, either logical variables such as problem solving and the ability to think directly in the case of euthanasia, or indirectly in the case of involve- positively about the future (Williams and Pollock 2000). The rela- ment in the slaughter of meat-producing livestock. Familiarity with tionship between occupation and suicide can be conceptualised as a death and dying may affect attitudes in regard to the expendability model comprising four components contributing to the differential of life: 93 per cent of veterinary healthcare workers interviewed in a occupational suicide risk: demographics (the demographic composi- small-scale study indicated a favourable inclination towards euthana- tion of people in the occupation); internal occupational stress (stress sia of human beings (Kirwan 2005). This is a higher proportion than associated with the nature of the work); pre-existing psychiatric in the general population (Clery and others 2007) and contrasts with morbidity (the psychological profile of individuals attracted to the prevailing medical opinion (Seale 2009), although comparison of these occupation); and opportunity factors (opportunities available for studies is confounded by dissimilar research methods. Positive associa- access to lethal means of suicide) (Stack 2001). tions have been demonstrated between tolerance of suicide (more per- Suicidal behaviour can be conceptualised as a continuum of missive attitudes towards euthanasia, physician-assisted suicide and gradually increasing seriousness: feelings that life is not worth liv- unassisted suicide) and suicidal thoughts and behaviour (Neeleman ing, thoughts of taking one’s own life, seriously considering suicide, and others 1997, Etzersdorfer and others 1998, Zemaitiene and suicidal planning, suicidal attempt, and suicide completion. Another Zaborskis 2005, Gibb and others 2006, Joe and others 2007). The the- view considers that each of these forms of suicidal behaviour does not ory of cognitive dissonance (Harmon-Jones and Harmon-Jones 2007) together represent a continuum but is instead a discrete category, each – that psychological discomfort arising from conflicting thoughts or with a specific risk factor. A longitudinal study of suicidal planning beliefs motivates the modification of existing, or the acquisition of among doctors suggests that a continuum of severity may operate in new thoughts and beliefs to reduce the inconsistency and discomfort this occupational group (Tyssen and others 2004). Consistent with – may offer an explanation for any effect of euthanasia attitudes on the continuum hypothesis, for the purpose of the present structured suicide risk. Veterinary surgeons may experience uncomfortable ten- review, risk factors for suicidal behaviours in general are considered as sions between their desire to preserve life and an inability to treat a possible influences on the increased risk of suicide among veterinary case effectively, which could be ameliorated by modifying their atti- surgeons. tudes to preserving life to perceive euthanasia as a positive outcome. This altered attitude to death may then facilitate self-justification and Possible influences on increased risk of suicide lower their inhibitions towards perceiving suicide as a solution to their Access to means of suicide own problems. Access to potentially lethal means has a strong influence on the suicide rate. Decreases in the rate have been associated with changes to non- Suicide ‘contagion’ toxic domestic gas from coal gas, the installation of catalytic convert- Direct or indirect exposure to the suicidal behaviour of others can influ- ers in cars, smaller over-the-counter pack sizes of paracetamol and the ence attitudes and increase vulnerability to suicide (Maris and others installation of barriers on high bridges (Bennewith and others 2007, 2000). Knowledge of individual suicides can travel readily through Hawton 2007). Availability and knowledge of medicines is likely to the social networks of a small profession, and awareness of high lev- contribute to the suicide risk in doctors (Hawton and others 2000). els of suicide among professional peers may be a contributory risk Veterinarians also have ready access to medicines and knowledge of factor for suicidal behaviour in veterinary surgeons, creating a suicide

Veterinary Record | March 27, 2010

Online P&A March 27.indd 390 24/3/10 15:41:58 Papers Papers

‘contagion’ or modelling effect among vulnerable individuals within low EI. Fostering EI may have a place in veterinary education, both to a high-risk occupational group. improve the relationship between veterinary surgeons and clients and to act as a buffer against stress in the profession (Gelberg and Gelberg Cognitive and personality factors 2005, Timmins 2006). Personality and cognitive factors are important in the prediction of Personality traits of hopelessness, neuroticism and extroversion suicide risk (Sheehy and O’Connor 2008). The relationship between may be major influences on suicide (Brezo and others 2006). A lon- personality traits and vocational interest is well documented (Mount gitudinal study of UK medical graduates showed that career stress, and others 2005). Individuals have a preference for certain occupations burnout and satisfaction were predicted by trait measures of personal- based on their personality or life experiences, which could render ity that had been assessed five years earlier when they were medical them either more vulnerable or resilient to the work environment students (McManus and others 2004). A six-year longitudinal and (Kohn and Schooler 1982, Stansfeld 2002, Wilhelm and others 2004). nationwide study of Norwegian medical students demonstrated that The personality profiles of medical students (Meit and others 2007) the traits of neuroticism and high conscientiousness are risk factors for and doctors (Clack and others 2004) differ from those of the general stress (Tyssen and others 2007). Tyssen and others (2001a, b, 2004) population, and there are associations between personality factors have shown that specific personality traits and depression are com- and the choice of medical speciality (Borges and Savickas 2002): for mon predictors of mental health problems and suicidal ideation and example, mental health professionals are more likely to have early behaviours among medical students and young doctors, and demon- experiences of childhood trauma and family dysfunction than people strated the utility of screening final-year medical students to identify a in other professions (Elliott and Guy 1993), and women psychiatrists subgroup that might benefit from intervention. are more likely than women in other medical specialities to report Little is known about the coping strategies used by veterinar- personal or family histories of psychiatric disorders (Frank and others ians. Australian veterinary students were shown not to consistently 2001). The choice of a veterinary career may be influenced by factors employ a range of effective coping strategies to deal with the stres- such as a preference for working with animals rather than people; sors they encountered during their course of study (Williams and oth- previous interactions with animals may play a critical role in guiding ers 2005). Veterinarians in New Zealand have been shown to make veterinary students into their chosen career (Martin and others 2003, good use of their social networks to seek information and assistance Martin and Taunton 2005, Serpell 2005). in times of work-related stress, especially from informal sources such The veterinary profession may be particularly vulnerable to sui- as friends, family and colleagues, rather than from resources such as cide because of selection based on the very high academic requirements health professionals, counsellors and telephone helplines (Gardner for entry into veterinary schools (Halliwell and Hoskin 2005). These and Hini 2006). The importance of deploying both problem-focused admission requirements lead to an undergraduate cohort in which stu- and emotion-focused strategies to cope with the stresses of veterinary dents with a history of outperforming their peers are grouped with work has been emphasised (Bartram and Gardner 2008). similarly high academic achievers, creating an environment in which some students question their own abilities and fear that they will be Work-related stressors exposed as intellectual frauds (Zenner and others 2005). Similar senti- Work factors associated with psychological ill health in staff are the ments were expressed frequently by veterinary students in the USA demands of work (long hours worked, work overload and pressure) during counselling (Kogan and others 2005). This situation also applies and the associated effects of these on individuals’ personal lives; lack to students of other health professions, and predisposes to psychologi- of control over work; lack of participation in decision making; poor cal problems among those who do not adjust and ensure their expecta- social support; and unclear management and work role (Michie and tions of their performance are realistic (Henning and others 1998). Williams 2003). The demand-control model of job strain predicts that First-year students at a veterinary school in the USA possessed the combination of low decision latitude (low level of control over psychological characteristics consistent with other high-achieving the work environment and limited variety of work or opportunity competitive performance populations, such as professional athletes: for use of skills) and heavy job demands is associated with mental they had elevated anxiety levels, placed significant value on positive strain (Karasek 1979). Similarly, a combination of high demands and comparisons with the competence of their peers, and harboured a fear low decision latitude, imbalance between effort and reward, and poor of failure (Zenner and others 2005). Socially prescribed perfectionism social support at work from coworkers and supervisors are risk factors (an individual’s belief that others hold unrealistic and exaggerated for common mental disorders (Stansfeld and Candy 2006). High levels expectations of him/her that must be met in order to gain acceptance of decision latitude have been found to be protective of mental health and approval), self-criticism, concern about mistakes and doubts about in cross-sectional studies of veterinarians (Hesketh and Shouksmith action have been associated with suicidality (O’Connor 2007). 1986). Individuals without any pre-employment history of psychiat- Voracek (2004) reported a positive association of intelligence with ric disorders, exposed to high psychological job demands, had a two- suicide, but Gunnell and others (2005) and Andersson and others fold risk of onset of new depression and anxiety compared with those (2008) have reported inverse associations. Cognitive performance in with low job demands (Melchior and others 2007). Lack of support childhood appears to be significantly and inversely related to mor- from colleagues and supervisors, and deterioration in work charac- bidity and mortality from all causes in adulthood, even at the higher teristics, are associated with onset of depression (Waldenström and end of the intelligence spectrum, and independent of childhood others 2008). The central tenet of a complementary job stress model, socioeconomic status (Martin and Kubzansky 2005). Halliwell and the effort-reward imbalance model (Siegrist 1996), is that high effort Hoskin (2005) conjectured that the highly demanding veterinary in combination with low reward is a major risk factor for psychologi- undergraduate course has the potential to stifle the development of cal distress and ill health, especially in individuals characterised by a communication skills and emotional maturity, possibly more so than motivational pattern of overcommitment to their work. the medical curriculum. The considerable volume and pace of learn- Veterinary graduates typically move abruptly from the university ing at veterinary school is a stressor for students (Gelberg and Gelberg environment to the relative professional and social isolation of general 2005). Rice (2008) speculated that a factor influencing the high suicide private practice. Being on call, the financial aspects of the role, and lack risk could be that undergraduate admission criteria are based on high of surgical competence have been identified as the main initial difficul- academic achievement, which might select a greater proportion of ties, and staff turnover is high, with one in three veterinary graduates students with low ‘emotional intelligence’ (EI). EI is described as the leaving their first job within two years (Routly and others 2002). Many ability to perceive emotions in oneself and others, integrate emotions work with little supervision, do not always have access to assistance into thought processes, understand emotions and moderate emotions from colleagues and can make professional mistakes, which may have in oneself and others (Mayer and Salovey 1997). However, there is a a considerable emotional impact and may be significant in the develop- strong positive correlation between several dimensions of EI and aca- ment of suicidal thoughts (Mellanby and Herrtage 2004). demic achievement (Parker and others 2004, Petrides and others 2004, Veterinary work is perceived as stressful by over 80 per cent of UK Austin and others 2005), so it seems unlikely that the current admis- veterinarians (Robinson and Hooker 2006). Using a short validated sions procedure inevitably selects a high proportion of students with stress evaluation tool to measure and compare a number of work-

March 27, 2010 | Veterinary Record

Online P&A March 27.indd 391 24/3/10 15:41:58 Papers Papers

related stressors and stress outcomes across 26 different occupations The risk of suicide is raised in occupations in which those in the UK, veterinarians reported poorer psychological wellbeing employed are directly dependent upon clients for their income (Stack than workers in most other occupations (Johnson and others 2005). 2001). An exploratory study of 36 occupations found that the suicide A large cross-sectional survey of mental health and wellbeing in the rate was over 1.5 times higher for people in client-dependent occupa- UK veterinary profession (Bartram and others 2009b, c) showed that tions (such as physicians, dentists and retail proprietors) in comparison veterinary surgeons have less favourable (that is, a greater risk of work- with those in non-client-dependent occupations, and it was assumed related stress) psychosocial working conditions in relation to demands that this increased risk is associated with client dependency being a and managerial support than the general working population. The major source of stress (Labovitz and Hagedorn 1971). Veterinary sur- number of hours worked, making professional mistakes, and the geons in private practice work in a client-dependent environment. possibility of client complaints or litigation were the main reported contributors to stress (Bartram and others 2009c), consistent with the Effects of gender findings of a survey of work-related stress in the veterinary profession The PMR for suicide among veterinarians is greater among women in New Zealand (Gardner and Hini 2006). than men (Mellanby 2005), but this may reflect the small absolute Financial debt is a risk factor for suicidal behaviour in the general number of women concerned. This parallels the situation in the population (Hatcher 1994, Hintikka and others 1998, Yip and others medical profession, in which female doctors have higher suicide rates, 2007). Twenty-nine per cent of working doctors who died by suicide compared with the general population, than their male colleagues do in England and Wales had significant financial problems in the year (Baldwin and Rudge 1995, Hawton and others 2001, Schernhammer before death (Hawton and others 2004). There may be an association and Colditz, 2004, Petersen and Burnett 2008). In marked contrast to between rising student debt and suicide among veterinary surgeons the general population, in which the suicide rate for men is approxi- (Williams 2006). In 2008, the average debt faced by a final-year first- mately three times the suicide rate for women (Brock and others degree veterinary student in the UK was over £20,000, and 35 per cent 2006), the absolute suicide rate among doctors is similar in both sexes of final-year students reported their financial problems as either diffi- (Lindeman and others 1996, Hawton and others 2001). This differen- cult or severe (British Veterinary Association [BVA] and Association of tial increase in risk between the sexes requires particular monitoring in Veterinary Students [AVS] 2008). view of the large increase in the number of women entering the veteri- Veterinary practice occupies a difficult and complex moral posi- nary profession. However, Hawton and others (2001) have suggested tion because it serves animal and human interests, which may conflict that the risk among female doctors might be expected to decline as (Tannenbaum 1993, de Graaf 2005). Ethical challenges are common- they become the majority in a traditionally male-dominated occupa- place as veterinarians seek to balance their obligations to ensure the tion, and this may also apply within the veterinary profession. welfare of their patients while accommodating the owners’ expect Women have a greater preference for self-poisoning as a method of ations or demands, often within strict economic constraints, the views suicide than men, a pattern that is more pronounced in female doctors of professional peers, the wider interests of society as a whole, and the and veterinarians than in the general population (Kelly and Bunting commercial interests of private practice (Rollin 2006). Animal shelter 1998), so it is possible that in an occupational context of ready access workers involved with the euthanasia of unwanted animals report and knowledge, as in the medical and veterinary professions, the risk high levels of work-related stress, work-to-family conflict, poor physi- of suicide may increase more for women than for men. However, cal health, substance misuse and low levels of job satisfaction (Rollin male doctors and veterinarians also use self-poisoning for suicide more 1987, Reeve and others 2005, Rogelberg and others 2007). often than men in general (Kelly and Bunting 1998), so this is unlikely Interviews with US graduating veterinary students revealed that to be the sole explanation for the difference between the sexes. The their most common upsetting experiences involved procedures they particular stresses affecting women in these professions are likely to felt were unnecessary, such as when an animal’s suffering was pro- contribute to the difference (Gross 1997, Hawton and others 2000). longed because its owner did not accept that its disease was incurable By far the largest and most frequent gender-related source of stress and that death was inevitable (Herzog and others 1989). Procedures identified for female junior doctors is the conflict they feel between causing pain to animals, such as the castration of calves without their career and their marital and family life, and this is strongly linked anaesthesia, were also considered stressful to the students, and 25 per to depression (Firth-Cozens 1990). cent found euthanasia of animals personally distressing. Female veterinary students report higher levels of emotional empa- A survey of veterinarians in Finland reported high levels of work- thy with animals (Paul and Podberscek 2000), greater concerns for the related stress, particularly among those working in urban areas and welfare or rights of animals (Serpell 2005) and attach greater impor- academia (Reijula and others 2003). In a cross-sectional study of the tance to the human-animal bond (Martin and others 2003, Martin occupational health of veterinarians in Australia, the proportion of and Taunton 2005) than their male counterparts. This is reflected in respondents with high levels of psychological distress was double differences between male and female veterinary surgeons’ reported that of the general population; significant predictors of high levels of emotional responses to treatment failure and carrying out euthanasia distress included being less than 35 years old and having taken non- (Fogle and Abrahamson 1990), and in attitudes to pain control in ani- prescription drugs in the past 12 months. Working long hours and mals (Capner and others 1999, Raekallio and others 2003, Hugonnard being on call after hours were major contributory factors for stress. and others 2004). This may affect the ability of women to cope with Veterinarians working in general practice were significantly more the emotional stresses and moral conflicts inherent in veterinary prac- stressed than those in other roles (Fairnie 2005). In a survey of veteri- tice (Paul and Podberscek 2000). nary surgeons in the UK, deaths of animals caused by illness or eutha- Phillips-Miller and others (2000, 2001) surveyed married or part- nasia were reported to provoke significant short- and long-term emo- nered veterinarians in the USA about work satisfaction, work-related tional reactions in a substantial proportion of respondents, which may stress, marital and family stress, and perceived level of spousal support be relevant to the instigation of depression (Fogle and Abrahamson for their career. The two sexes reported similar levels of work satisfac- 1990). A study of veterinarians in the USA demonstrated that com- tion and work-related stress, but women reported a significantly high- municating bad news to clients is stressful and, for some individuals, er effect of marital and family stress on their career and lower levels the feelings of stress are prolonged (Ptacek and others 2004). A large of spousal support than their male counterparts. The areas of greatest cross-sectional study of veterinarians in Germany also recorded high work dissatisfaction for both sexes were income and long working levels of work-related stress, mainly attributable to workload and out- hours. Emergency calls outside office hours and unwillingness of cli- of-hours on-call duties (Harling and others 2009). Small surveys of vet- ents to agree to diagnostic or treatment procedures were also cited as erinary surgeons in Ireland have recorded similar outcomes (Connolly being particularly stressful. 2003, Kinsella 2006). Finally, Trimpop and others (2000a, b) demonstrated that stress associated with long working hours was a predictor Disease epidemics of traffic accident rates in German veterinarians: those who worked Major incidents such as the outbreak of foot-and-mouth disease in over 48 hours per week reported significantly higher levels of work the UK in 2001 can increase the levels of psychological morbidity in stress and incidences of driving accidents. affected communities (Peck 2005), potentially including veterinarians

Veterinary Record | March 27, 2010

Online P&A March 27.indd 392 24/3/10 15:41:59 Papers Papers

TABLE 4: Summary of studies included in this review with findings relevant to mental health in veterinarians

Response Study Design Sample population Location Sample size rate (%) Key findings

Anon 2002 Cross-sectional Veterinary surgeons and Australia 313 34 Euthanasia, specific client behaviours or characteristics, and long support staff 183 20 working hours were important stressors Bartram and others 2009a, b, c Cross-sectional Veterinary surgeons UK 1796 56 Symptoms of mental ill health elevated relative to general population. High psychological demands and low levels of support Brown 1994 Cross-sectional Veterinary students USA 207 96 Students deploy a range of both problem-focused and emotion- focused coping styles with relatively equal frequency Connolly 2003 Cross-sectional Veterinary surgeons Ireland 46 26 Workload reported as main source of stress Fairnie 2005 Cross-sectional Veterinary surgeons Australia 419 43 Proportion of highly distressed respondents was double that of the general population. Distress highest for those <35 years old Fishbain 1986 Case series Veterinary surgeons USA 4 – Pethidine by intramuscular injection was principal drug abused; history of drug abuse in veterinary school; veterinary-sourced Fogle and Abrahamson 1990 Cross-sectional Veterinary surgeons UK 167 56 Animal deaths provoke severe short-term (76 per cent) and long- term (20 per cent) emotional reactions in veterinarians Fritschi and others 2009 Cross-sectional Veterinary surgeons Australia 2125 37 One-third reported poor psychological health but levels of distress, anxiety and depression similar to other professions Gardner and Hini 2006 Cross-sectional Veterinary surgeons NZ 927 49 Those working in small animal practice, women and younger veterinarians reported the highest levels of stress Hafen and others 2006, 2008 Cohort Veterinary students USA T1:93 T1: 90 One-third of first-year students report clinical levels of depressive T2:78 T2: 84 symptoms. Predictors include academic concerns Hansez and others 2008 Cross-sectional Veterinary surgeons Belgium 216 9 Mean job strain and job engagement levels not higher than other professions, but greater negative work-home interaction Harling and others 2009 Cross-sectional Veterinary surgeons Germany 1136 53 Work-related stress mainly attributable to workload, out-of-hours on-call duties and difficult clients. High alcohol consumption Heath 2007 Cross-sectional Veterinary surgeons Australia 134 98 Only 52 per cent would opt for the veterinary profession if they could start their career again Herzog and others 1989 Cross-sectional* Graduating veterinary USA 24 – Distress was associated with prolongation of animal suffering, students procedures causing pain to animals, and euthanasia Hesketh and Shouksmith 1986 Cross-sectional Veterinary surgeons New Zealand 411 59 Decision latitude protective of mental health; lack of control over speed of work and discretion are associated with anxiety Johnson and others 2005 Cross-sectional Veterinary surgeons UK and 262 36† Low psychological wellbeing, intermediate job satisfaction and Ireland good physical health compared with 25 other occupations Kinsella 2006 Cross-sectional Veterinary surgeons Ireland 14 44 Unsociable hours and workload ranked highest as stressors Kirwan 2005 Cross-sectional* Veterinary surgeons, UK 15 – 93 per cent have favourable inclination towards euthanasia of nurses and receptionists human beings Kogan and others 2005 Cross-sectional Veterinary students USA 233 44 Non-academic stressors include additional employment, relationship concerns and poor self-care habits Martin and others 2003 Cross-sectional Veterinary students USA 146 NR Human-animal bond influences decisions to become veterinarians, especially for women Martin and Taunton 2005 Cross-sectional Veterinary surgeons USA 415 26 81 per cent indicated that the human-animal bond was important to their decisions to become veterinarians, especially for women Mellanby and Herrtage 2004 Cross-sectional Recent graduates UK 108 27 Poor supervision, minimal support, many mistakes result in distress to veterinarians Paul and Podberscek 2000 Cross-sectional Veterinary students UK 319 NR Females have higher emotional empathy with animals, which is sustained throughout their degree for females but declines for males Phillips-Miller and others 2000, 2001 Cross-sectional Veterinary surgeons USA 305 54 Women report greater effects of marital and family stress and lower spousal support Ptacek and others 2004 Cross-sectional* Veterinary surgeons USA 62 – Communicating bad news is stressful and, for some individuals, the feelings of stress are prolonged Reijula and others 2003 Cross-sectional Veterinary surgeons Finland 785 67 Those working in towns or involved in education and research reported the most stress Robinson and Hooker 2006 Cross-sectional Veterinary surgeons UK 9671 47 Work is perceived as stressful by 80 per cent of the profession. Only 53 per cent would choose veterinary career again Routly and others 2002 Cross-sectional Recent graduates and UK 58 54 Main difficulties for new graduates are being on call, financial senior partners 34 59 aspects of the role, and lack of surgical competence Serpell 2005 Cross-sectional Veterinary students USA 302 92 Females have greater concern for animal welfare/rights. Interactions with animals influences attitudes and career choice Trimpop and others 2000a, b Cross-sectional Veterinary surgeons and Germany 494 NR Stress associated with long working hours was a predictor of support staff 284 traffic accident rates Williams and others 2005 Cross-sectional Veterinary students USA 57 41 Narrow range of effective coping strategies Zenner and others 2005 Cross-sectional Veterinary students USA 61 80 Psychological characteristics similar to other high-achieving populations

* Interview based; all other included studies are questionnaire based † Response rate for veterinarians is not reported in Johnson and others (2005). Figure obtained from the original source Connolly (2002) NR Not reported

because of their involvement in large-scale slaughter and disposal of (Peck 2005). Nusbaum and others (2007) described a number of animals, provision of emotional support to farmers, and the economic negative psychological reactions in veterinarians involved with the effects on private practices. During this outbreak, veterinary surgeons outbreak, and advocated training in ‘psychologic first aid’ for public were turned to for emotional support by 40 per cent of farmers sur- health professionals involved in such incidents, to limit emotional dis- veyed, and were the second most cited source of support, after family, tress in the rural community and themselves. friends and other farmers (Peck and others 2002). Calls from veterinarians to Vet Helpline, a peer-support telephone helpline providing Complaints at work emotional support to members of the profession in the UK, increased Seventy-one per cent of working doctors who died by suicide in during the outbreak and decreased to pre-outbreak levels by mid-2002 England and Wales between January 1991 and December 1993

March 27, 2010 | Veterinary Record

Online P&A March 27.indd 393 24/3/10 15:41:59 Papers Papers

Alcohol or drug misuse Ready access to and knowledge of means

Self-selection: risk Negative effects in Work-related stressors Psychological morbidity factors present in people undergraduate training Suicide attracted to career Long working hours Feelings of entrapment Curriculum Client expectations Depression Previous life events Extramural studies Inadequate support Cognitive distortion Personality dimensions Personal finances Emotional exhaustion Self-referent negative thoughts Genetics Psychosocial factors Unexpected clinical outcomes Ruminative thinking Attitudes to euthanasia Poor coping strategies Disenchantment with career Suicidal thoughts established Hopelessness

Non-career-related chronic major Professional and social Perceived stigma of difficulties, undesirable life events isolation help-seeking behaviour or pre-existing psychiatric disorder

FIG 1: Schematic representation of a hypothetical model to explain the risk of suicide in veterinary surgeons (reproduced with permission from Bartram and Baldwin 2008)

had had significant problems at work in the year before their death; risk of suicide among doctors may be related to both an increased risk over one-third of those were facing complaints, which in most cases of psychiatric disorder and the increased risk of those with a psychi- appeared to have been a key factor leading to suicide, although most atric disorder to die by suicide. Although no cases of suicide among of them were also facing other problems at work and at home. Other veterinarians were identified in the cohort of suicides examined by common occupational problems included feeling overloaded by their Agerbo and others (2001), it is possible that the dual effect seen in doc- volume of work, working long hours and feeling unable to cope with tors may also apply in the veterinary profession. the responsibility of their job (Hawton and others 2004). The number The principal psychological problems experienced by doctors are of complaints received by the Royal College of Veterinary Surgeons depression and alcoholism, and these are more frequent than in the (RCVS) from the general public increases steadily each year, with the general population (Firth-Cozens 2001). Pre-existing psychiatric disor- main category being alleged inadequate care (RCVS 2007). ders, mainly depressive illness and alcohol or drug dependence, were present in over 86 per cent of doctors dying by suicide (Hawton and Perceived stigma others 2004), and in over 90 per cent of nurses (Hawton and others The stigma associated with mental illness is increasingly recognised as 2002). In a study of UK healthcare professionals referred to a special- an important obstacle to the provision of care to people with this dis- ist drug and alcohol treatment service, approximately one-third had order (Sartorius 2007). Mental illness may be particularly stigmatising previously received treatment for depression and about one-fifth had for those working in professions where vulnerabilities are not read- previously taken an overdose (Gossop and others 2001). No such data ily tolerated. Stigma is recognised as an important influence on doc- are available for veterinarians, but it is reasonable to speculate that tors’ decisions not to seek mental healthcare (White and others 2006, psychiatric disorders may be a similar factor in suicides by veterinary Worley 2008), and the risk of suicide may be greater in higher income surgeons. There is a ready opportunity in both professions for misuse earners who develop mental illness, as they may feel more stigma- of prescription medications. Among UK veterinary surgeons referred tised than others with lower incomes (Agerbo and others 2001). A to a health support programme, the order of preference of substance similar stigma may apply within the veterinary profession, with a misuse was alcohol, ketamine, benzodiazepines, opiates, street drugs consequent reduction in help-seeking behaviour. (cannabis, heroin, cocaine and ecstasy) and nitrous oxide, and approximately half of those treated admitted to having had suicidal thoughts Psychiatric illness (Veterinary Benevolent Fund [VBF] 2007). The risk of diagnosis with affective and stress-related disorders is high- Bartram and others (2009a, b, c) estimated that, compared with er in the human service professions (healthcare, education, social work the general population, UK veterinary surgeons report higher levels of and customer services) relative to all other occupations, after adjust- anxiety and depressive symptoms, a higher 12-month prevalence of ment for an extensive range of sociodemographic variables (Wieclaw suicidal thoughts, less favourable psychosocial working conditions in and others 2005, 2006). Specific professions contribute differentially relation to demands and managerial support, lower levels of positive to the magnitude of this increased risk: for example, teaching and mental wellbeing, and higher levels of negative work-home interac- social services display the highest risk of stress-related disorders, and tion. The level of alcohol consumption did not appear to be a negative health professionals have an elevated risk of depression. The elevated influence on mental health within the profession as a whole. This risk among the human service professions may be associated with last observation is consistent with the findings of Mellanby and oth- factors including job characteristics such as irregular working hours ers (2008) that the PMRs for alcohol-related deaths among male and and exposure to distressing events; high and conflicting job demands; female veterinary surgeons aged 20 to 64 years in England and Wales self-selection into certain occupations of individuals who are inclined during the period from 1993 to 2005 were lower than for the general to display a high degree of commitment; and men and women occu- population, although the differences were not statistically significant. pying occupations in which the opposite sex predominates. Harling and others (2009) explored the use of psychotropic substanc- Agerbo and others (2001) explored the impact of psychiatric faces among German veterinarians and found that the prevalence of tors on the relative risk of suicide across 55 occupations. Only modest smoking was lower than that in the general population, and that the associations between suicide and occupation were observed among prevalence of drug use was similar in both groups. However, alcohol individuals who had been hospitalised previously for psychiatric ill- consumption was higher among veterinarians, and the prevalence ness. This implies that mental disorder lies on the causal pathway of dangerous levels of alcohol consumption was markedly elevated between occupation and suicide, or that occupational differences among female veterinarians. Hafen and others (2006, 2008) reported are less important once a person suffers from a psychiatric illness. that one-third of first-year students at a veterinary school in the USA However, doctors were a notable exception, being at almost fourfold had symptoms of depression associated with both academic and non- higher risk than other occupations of dying by suicide if they had been academic stressors, but it is not clear to what extent these findings can hospitalised previously for a psychiatric disorder. Thus, the increased be generalised to veterinary undergraduates elsewhere. Allister (2009)

Veterinary Record | March 27, 2010

Online P&A March 27.indd 394 24/3/10 15:42:00 Papers Papers

reported elevated signs of mental ill health among fourth-year veteri- to veterinary school, whether there are negative influences on men- nary students in the UK. Finally, Fishbain (1986) compared the char- tal health during undergraduate training, and whether individuals’ acteristics of four US veterinary surgeons with psychiatric illness with maladaptive coping strategies might play a role in the development those of published descriptions of psychiatrically impaired doctors; of ill health. This in turn could lead to the development of enhanced similarities included the type of misused drug (mainly opiates), the assessments for entry to the veterinary degree course, and inform the number of drugs (multiple), and the source of supply (self-prescribed development and timing of educational interventions at a systemic or rather than street-sourced). The veterinarians in that study considered individual level to improve the wellbeing of students and their resil- overwork, fatigue, situational stress and problems with significant ience during their subsequent careers. others to be precipitating factors. The outcomes must be interpreted very cautiously, however, due to the small number of veterinarians References studied. ANON (2002) Survey details stress factors that influence Australian vets. Australian Studies with findings relevant to mental health in veterinarians Veterinary Journal 80, 522-524 are summarised in Table 4. Agerbo, E., Gunnell, D., Bonde, J. P., Mortensen, P. B. & Nordentoft, M. (2007) Suicide and occupation: the impact of socio-economic, demographic, and psychiatric differences. Psychological Medicine 37, 1131-1140 Towards a hypothetical model of suicide risk Agerbo, E., Mortensen, P. B., Eriksson, T., Qin, P. & Westergaard- in veterinary surgeons Nielsen, N. (2001) Risk of suicide in relation to income level in people admitted A hypothetical model to explain the risk of suicide in veterinary sur- to hospital with mental illness: nested case-control study. British Medical Journal 322, 334-335 geons has been proposed (Fig 1) (Bartram and Baldwin 2008). The Alexander, R. E. (2001) Stress-related suicide by dentists and other health care work- model attempts to clarify the complex interaction of possible influenc- ers: fact or folklore? Journal of the American Dental Association 132, 786-794 es, is based on specific testable constructs, and may facilitate a more Allister, R. J. (2009) Mental health and wellbeing in veterinary students. MSc Thesis. focused and systematic approach for suicide research and the devel- University of Edinburgh Andersson, L., Allebeck, p., Gustafsson, J-E. & Gunnell, D. (2008) opment of prevention strategies within the profession. The model Association of IQ scores and school achievement with suicide in a 40-year follow-up is congruent with the stress-diathesis (Mann and others 1999), Cry of a Swedish cohort. Acta Psychiatrica Scandinavica 118, 99-105 of Pain (Williams and Pollock 2000, Williams 2001) and differential ANDRIESSEN, K. (2006) Do we need to be cautious in evaluating suicide statistics? occupational risk (Stack 2001) models of suicide described above. European Journal of Public Health 16, 445-447 The principal hypotheses within this model are that suicide risk Austin, E. J., Evans, P., Goldwater, R. & Potter, V. (2005) A preliminary study of emotional intelligence, empathy and exam performance in first year medical in veterinarians is influenced by: individuals entering the profession students. Personality and Individual Differences 39, 1395-1405 having characteristics that confer a predisposition or vulnerabil- Baldwin, D. S. & Rudge, S. E. (1995) Depression and suicide in doctors and medi- ity; psychological morbidity induced by psychosocial factors during cal students. In Health Risks to the Health Care Professional. Ed P. Litchfield. Royal undergraduate training and work-related stress; attitudes to suicide College of Physicians. pp 77-90 Bartram, D. J. & Baldwin, D. S. (2008) Veterinary surgeons and suicide: induced by the practice of euthanasia of animals and awareness of influences, opportunities and research directions. Veterinary Record 162, 36-40 suicide within the profession; and access to and knowledge of means Bartram, D. J. & Gardner, D. (2008) Coping with stress. In Practice 30, 228-231 of suicide. The model attempts to clarify a complex interaction of pos- Bartram, D. J., Sinclair, J. M. A. & Baldwin, D. S. (2009a) Alcohol consump- sible mechanisms across the course of the veterinary career, and may tion among veterinary surgeons in the UK. Occupational Medicine 59, 323-326 Bartram, D. J., Yadegarfar, G. & Baldwin, D. S. (2009b) A cross-sectional serve as a useful heuristic to facilitate a more focused and systematic study of mental health and well-being and their associations in the UK veterinary approach to research. Research is required to validate or disprove the profession. Social Psychiatry and Psychiatric Epidemiology 44, 1075-1085 component hypotheses. Bartram, D. J., Yadegarfar, G. & Baldwin, D. S. (2009c) Psychosocial working conditions and work-related stressors among UK veterinary surgeons. Occupational Implications for research Medicine 59, 334-341 BENNEWITH, O., NOWERS, M. & GUNNELL, D. (2007) Effect of barriers on the Despite the low absolute number of suicides by veterinarians com- Clifton suspension bridge, England, on local patterns of suicide: implications for prepared with workers in other healthcare professions, the high PMR for vention. British Journal of Psychiatry 190, 266-267 suicide within the veterinary profession – one of the highest for any BLAIR, A. & HAYES, H. M. (1980) Cancer and other causes of death among US veteri- occupation – warrants dedicated research to expand the presently lim- narians, 1966-1977. International Journal of Cancer 25, 181-185 BLAIR, A. & HAYES, H. M. (1982) Mortality patterns among US veterinarians, ited evidence base, to inform the development and implementation 1947-1977: an expanded study. International Journal of Epidemiology 11, 391-397 of suitable interventions. Such research would be important, not only BORGES, N. J. & SAVICKAS, M. L. (2002) Personality and medical specialty choice: a for the wellbeing of individual members of the profession, but also in literature review and integration. Journal of Career Assessment 10, 362-380 view of the potentially deleterious impact of practitioners’ mental ill Brezo, J., Paris, J. & Turecki, G. (2006) Personality traits as correlates of suicidal ideation, suicide attempts, and suicide completions: a systematic review. Acta Psychiatrica health on the welfare of animals under their care, and the insight that Scandinavica 113, 180-206 research in this professional group might provide into influences on BROCK, A., BAKER, A., GRIFFITHS, C., JACKSON, G., FEGAN, G. & suicide in other occupations. MARSHALL, D. (2006) Suicide trends and geographical variations in the United Each stage of the veterinary career path – from the characteristics Kingdom, 1991-2004. Health Statistics Quarterly 31, 6-22 BROWN, S. L. (1994) Factor structure of a brief version of the Ways of Coping (WOC) of applicants to veterinary schools, undergraduate training, subsequent questionnaire: a study with veterinary medicine students. Measurement and Evaluation in employment, and through to retirement – can be examined to identify Counseling and Development 27, 308-315 early predisposing factors and later triggers for suicidal behaviour in BVA & AVS (2008) Survey results 2008. www.bva.co.uk/public/documents/2008_ AVS_ the profession. This ‘life course’ approach (Gunnell and Lewis 2005) Survey_Results.pdf. Accessed May 25, 2009 Capner, C. A., Lascelles, B. D. X. & Waterman-Pearson, A. E. (1999) would enable multiple points on the career continuum to be targeted Current British veterinary attitudes to perioperative analgesia for dogs. Veterinary Record with appropriate interventions. 145, 95-99 Interviews with veterinary surgeons who have experienced sui- Charlton, J. (1995) Trends and patterns in suicide in England and Wales. International cidal thoughts could yield informative insights into the circumstances Journal of Epidemiology 24 (Suppl 1), S45-S52 that they believe contributed to those thoughts, the nature and sources CHARLTON, J., KELLY, S., DUNNELL, K., EVANS, B. & JENKINS, R. (1993) Suicide deaths in England and Wales: trends in factors associated with suicide deaths. Population of help they sought, and their perceived barriers to seeking help. An Trends 71, 34-42 investigation based on examination of coroners’ records on deaths of Clack, G. B., Allen, J., Cooper, D. & Head, J. O. (2004) Personality differences veterinary surgeons that received a verdict of suicide or undetermined between doctors and their patients: implications for the teaching of communication cause could provide insight into the circumstances of suicides in the skills. Medical Education 38, 177-186 Clery, E., McLean, S. & Phillips, M. (2007) Quickening death: the euthanasia profession and help to identify proximal risk factors. debate. In British Social Attitudes: the 23rd Report – Perspectives on a Changing Society. The veterinary profession’s role in providing animal euthanasia Eds A. Park, J. Curtice, K. Thomson, M. Phillips, M. Johnson. Sage. pp 35-53 and so facilitating a ‘good death’, may normalise suicide, with death CONNOLLY, D. J. (2003) Stress in the veterinary profession: the nature, implications and perceived as a rational solution to intractable problems. There has been consequences for veterinarians in the West of Ireland. MA Thesis. National University of Ireland, Galway no rigorous examination of this hypothesis to date. CONNOLLY, D. J. (2002) An exploratory study of stress in the veterinary profession Research with veterinary students could help identify whether in the UK and Ireland. MSc thesis. University of Manchester Institute of Science and there is a predilection towards mental health problems in applicants Technology

March 27, 2010 | Veterinary Record

Online P&A March 27.indd 395 24/3/10 15:42:00 Papers Papers

CORDELL, W. H., CURRY, S. C., FURBEE, R. B. & MITCHELL-FLYNN, D. L. (1986) and psychological adjustment in medical, dental, nursing and pharmacy students. Veterinary drugs as suicide agents. Annals of Emergency Medicine 15, 939-943 Medical Education 32, 456-464 DE GRAAF, G. (2005) Veterinarians’ discourses on animals and clients. Journal of Agriculture HERZOG, H. A., Jr, VORE, T. L. & NEW, J. C., Jr (1989) Conversations with veterinary and Environmental Ethics 18, 557-578 students: attitudes, ethics, and animals. Anthrozoös 2, 181-188 Elliott, D. M. & Guy, J. D. (1993) Mental health professionals versus non-mental- HESKETH, B. & SHOUKSMITH, G. (1986) Job and non-job activities, job satisfaction health professionals: childhood trauma and adult functioning. Professional Psychology: and mental health among veterinarians. Journal of Occupational Behaviour 7, 325-339 Research and Practice 24, 83-90 Hintikka, J., Kontula, O., Saarinen, P., Tanskanen, A., Koskela, K. Etzersdorfer, E., Vijayakumar, L., Schöny, W., Grausgruber, A. & & Viinamäki, H. (1998) Debt and suicidal behaviour in the Finnish general popula- Sonneck, G. (1998) Attitudes towards suicide among medical students – com- tion. Acta Psychiatrica Scandinavica 98, 493-496 parison between Madras (India) and Vienna (Austria). Social Psychiatry and Psychiatric Hugonnard, M., Leblond, A., Keroack, S., Cadoré, J. & Troncy, E. Epidemiology 33, 104-110 (2004) Attitudes and concerns of French veterinarians towards pain and analgesia in Fairnie, H. M. (2005) Occupational injury, disease and stress in the veterinary profes- dogs and cats. Veterinary Anaesthesia and Analgesia 31, 154-163 sion. PhD Thesis. Curtin University of Technology, Australia. http://adt.curtin.edu.au/ Joe, S., Romer, D. & Jamieson, P. E. (2007) Suicide acceptability is related to theses/available/adt-WCU20070528.140327. Accessed March 7, 2008 suicide planning in US adolescents and young adults. Suicide and Life-Threatening Behavior Firth-Cozens, J. (1990) Sources of stress in women junior house officers. British 37, 165-178 Medical Journal 301, 89-91 Johnson, S., Cooper, C., Cartwright, S., Donald, I., Taylor, P. & Firth-Cozens, J. (2001) Interventions to improve physicians’ well-being and patient Millet, C. (2005) The experience of work-related stress across occupations. Journal of care. Social Science and Medicine 52, 215-222 Managerial Psychology 20, 178-187 FISHBAIN, D. A. (1986) Veterinarians with psychiatric impairment – a comparison with Jones-Fairnie, H., Ferroni, P., Silburn, S. & Lawrence, D. (2008) Suicide impaired physicians. Journal of the National Medical Association 78, 133-137 in Australian veterinarians. Australian Veterinary Journal 86, 114-116 FOGLE, B. & ABRAHAMSON, D. (1990) Pet loss: a survey of the attitudes and feelings KARASEK, R. A., Jr (1979) Job demands, job decision latitude, and mental strain: impli- of practicing veterinarians. Anthrozoös 3, 143-150 cations for job redesign. Administrative Science Quarterly 24, 285-308 FRANK, E., BOSWELL, L., DICKSTEIN, L. J. & CHAPMAN, D. P. (2001) Kelly, S. & Bunting, J. (1998) Trends in suicide in England and Wales, 1982-96. Characteristics of female psychiatrists. American Journal of Psychiatry 158, 205-212 Population Trends 92, 29-41 FRITSCHI, L., MORRISON, D., SHIRANGI, A. & DAY, L. (2009) Psychological well- KELLY, S., CHARLTON, J. & JENKINS, R. (1995) Suicide deaths in England and Wales, being of Australian veterinarians. Australian Veterinary Journal 87, 76-81 1982-1992: the contribution of occupation and geography. Population Trends 80, 16-25 Gardner, D. H. & Hini, D. (2006) Work-related stress in the veterinary profession KINLEN, L. J. (1983) Mortality among British veterinary surgeons. British Medical Journal in New Zealand. New Zealand Veterinary Journal 54, 119-124 287, 1017-1019 Gelberg, S. & Gelberg, H. (2005) Stress management interventions for veterinary KINSELLA, M. (2006) Suicide in the veterinary profession: the hidden reality – part two. students. Journal of Veterinary Medical Education 32, 173-181 Irish Veterinary Journal 59, 704-706 Gibb, B. E., Andover, M. S. & Beach, S. R. H. (2006) Suicidal ideation and atti- Kirwan, A. P. (2005) Mankind and other animals. MA Thesis. Open University tudes toward suicide. Suicide and Life-Threatening Behavior 36, 12-18 KOGAN, L. R., MCCONNELL, S. L. & SCHOENFELD-TACHER, R. (2005) Veterinary Goldney, R. D. (2005) Risk factors for suicidal behaviour. In Prevention and Treatment students and non-academic stressors. Journal of Veterinary Medical Education 32, 193-200 of Suicidal Behaviour: From Science to Practice. Ed K. Hawton. Oxford University Kohn, M. L. & Schooler, C. (1982) Job conditions and personality: a longitudinal Press. pp 161-182 assessment of their reciprocal effects. American Journal of Sociology 87, 1257-1286 GOSSOP, M., STEPHENS, S., STEWART, D., MARSHALL, J., BEARN, J. & Labovitz, S. & Hagedorn, R. (1971) An analysis of suicide rates among occupa- STRANG, J. (2001) Health care professionals referred for treatment of alcohol and tional categories. Sociological Inquiry 41, 67-72 drug problems. Alcohol and Alcoholism 36, 160-164 Lindeman, S., Läärä, E., Hakko, H. & Lönnqvist, J. (1996) A systematic Gross, E. B. (1997) Gender differences in physician stress: why the discrepant findings? review on gender-specific suicide mortality in medical doctors. British Journal of Psychiatry Women and Health 26, 1-14 168, 274-279 Gunnell, D. & Lewis, G. (2005) Studying suicide from the life course perspective: McManus, I. C., Keeling, A. & Paice, E. (2004) Stress, burnout and doctors’ implications for prevention. British Journal of Psychiatry 187, 206-208 attitudes to work are determined by personality and learning style: a twelve year lon- Gunnell, D., Magnusson, P. K. E. & Rasmussen, F. (2005) Low intelligence gitudinal study of UK medical graduates. BMC Medicine 2, 29 test scores in 18-year-old men and risk of suicide: cohort study. British Medical Journal Malmberg, A., Simkin, S. & Hawton, K. (1999) Suicide in farmers. British Journal 330, 167 doi: 10.1136/bmj.38310.473565.8F of Psychiatry 175, 103-105 Hafen, M., Jr, Reisbig, A. M. J., White, M. B. & Rush, B. R. (2006) Predictors of MANN, J. J., WATERNAUX, C., HAAS, G. L. & MALONE, K. M. (1999) Toward a depression and anxiety in first-year veterinary students: a preliminary report. Journal of clinical model of suicidal behaviour in psychiatric patients. American Journal of Psychiatry Veterinary Medical Education 33, 432-440 156, 181-189 Hafen, M., Jr, Reisbig, A. M. J., White, M. B. & Rush, B. R. (2008) The first-year Maris, R. W., Berman, A. L. & Silverman, M. M. (2000) The social relations veterinary student and mental health: the role of common stressors. Journal of Veterinary of suicides. In Comprehensive Textbook of Suicidology. Guilford Publications. pp Medical Education 35, 102-109 240-265 Halliwell, R. E. W. & Hoskin, B. D. (2005) Reducing the suicide rate Martin, F., Ruby, K. & Farnum, J. (2003) Importance of the human-animal bond among veterinary surgeons: how the profession can help. Veterinary Record 157, for pre-veterinary, first-year, and fourth-year veterinary students in relation to their 397-398 career choice. Journal of Veterinary Medical Education 30, 67-72 Hansez, I., Schins, F. & Rollin, F. (2008) Occupational stress, work-home inter- Martin, F. & Taunton, A. (2005) Perceptions of the human-animal bond in vet- ference and burnout among Belgian veterinary practitioners. Irish Veterinary Journal 61, erinary education of veterinarians in Washington State: structured versus experiential 233-241 learning. Journal of Veterinary Medical Education 32, 523-530 HARLING, M., STREHMEL, P., SCHABLON, A. & NIENHAUS, A. (2009) Martin, L. T. & Kubzansky, L. D. (2005) Childhood cognitive performance and Psychosocial stress, demoralization and the consumption of tobacco, alcohol and risk of mortality: a prospective cohort study of gifted individuals. American Journal of medical drugs by veterinarians. Journal of Occupational Medicine and Toxicology 4, 4 doi: Epidemiology 162, 887-890 10.11186/1745-6673-4-4 Mayer, J. D. & Salovey, P. (1997) What is emotional intelligence? In Emotional Harmon-Jones, E. & Harmon-Jones, C. (2007) Cognitive dissonance theory Development and Emotional Intelligence: Educational Implications. Eds P. Salovey, D. after 50 years of development. Zeitschrift für Sozialpsychologie 38, 7-16 Sluyter. Basic Books. pp 3-31 HATCHER, S. (1994) Debt and deliberate self-poisoning. British Journal of Psychiatry 164, MEIT, S. S., BORGES, N. J. & EARLY, L. A. (2007) Personality profiles of incoming 111-114 male and female medical students: results of a multi-site 9-year study. Medical Education Hawton, K. (2007) Restricting access to methods of suicide: rationale and evaluation Online 12, 7 of this approach to suicide prevention. Crisis 28 (Suppl 1), 4-9 Melchior, M., Caspi, A., Milne, B. J., Danese, A., Poulton, R. & Moffitt, Hawton, K., Clements, A., Sakarovitch, C., Simkin, S. & Deeks, J. J. T. E. (2007) Work stress precipitates depression and anxiety in young, working women (2001) Suicide in doctors: a study of risk according to gender, seniority and specialty and men. Psychological Medicine 37, 1119-1129 in medical practitioners in England and Wales, 1979-1995. Journal of Epidemiology and Mellanby, R. J. (2005) Incidence of suicide in the veterinary profession in England Community Health 55, 296-300 and Wales. Veterinary Record 157, 415-417 Hawton, K., Clements, A., Simkin, S. & Malmberg, A. (2000) Doctors who Mellanby, R. J. & Herrtage, M. E. (2004) Survey of mistakes made by recent kill themselves: a study of the methods used for suicide. Quarterly Journal of Medicine veterinary graduates. Veterinary Record 155, 761-765 93, 351-357 Mellanby, R. J., Platt, B., Simkin, S. & Hawton, K. (2008) Incidence of Hawton, K., Malmberg, A. & Simkin, S. (2004) Suicide in doctors: a psychologi- alcohol-related deaths in the veterinary profession in England and Wales, 1993-2005. cal autopsy study. Journal of Psychosomatic Research 57, 1-4 Veterinary Journal 181, 332-335 Hawton, K., Simkin, S., Rue, J., Haw, C., Barbour, F., Clements, A., Meltzer, H., Griffiths, C., Brock, A., Rooney, C. & Jenkins, R. (2008) Sakarovitch, C. & Deeks, J. (2002) Suicide in female nurses in England and Patterns of suicide by occupation in England and Wales: 2001-2005. British Journal of Wales. Psychological Medicine 32, 239-250 Psychiatry 193, 73-76 Hawton, K. & Vislisel, L. (1999) Suicide in nurses. Suicide and Life-Threatening MICHIE, S. & WILLIAMS, S. (2003) Reducing work related psychological ill health and Behavior 29, 86-95 sickness absence: a systematic literature review. Occupational and Environmental Medicine HEATH, T. J. (2007) Longitudinal study of veterinary students and veterinarians: the first 60, 3-9 20 years. Australian Veterinary Journal 85, 281-289 Miller, J. M. & Beaumont, J. J. (1995) Suicide, cancer and other causes of death HEM, E., HALDORSEN, T., AASLAND, O .G., TYSSEN, R., VAGLUM, P. & among California veterinarians, 1960-1992. American Journal of Industrial Medicine 27, EKEBERG, Ø. (2005) Suicide rates according to education with a particular focus on 37-49 physicians in Norway 1960-2000. Psychological Medicine 35, 873-880 MOUNT, M. K., BARRICK, M. R., SCULLEN, S. M. & ROUNDS, J. (2005) Higher- Henning, K., Ey, S. & Shaw, D. (1998) Perfectionism, the impostor phenomenon order dimensions of the big five personality traits and the big six vocational interest

Veterinary Record | March 27, 2010

Online P&A March 27.indd 396 24/3/10 15:42:01 Papers Papers

types. Personnel Psychology 58, 447-478 Occupational Health Psychology 1, 27-44 Neeleman, J., Haplern, D., Leon, D. & Lewis, G. (1997) Tolerance of suicide, STACK, S. (2001) Occupation and suicide. Social Science Quarterly 82, 384-396 religion and suicide rates: an ecological and individual study in 19 Western countries. Stansfeld, S. (2002) Work, personality and mental health. British Journal of Psychiatry Psychological Medicine 27, 1165-1171 181, 96-98 NUSBAUM, K. E., WENZEL, J. G. W. & EVERLY, G. S., Jr (2007) Psychologic first aid Stansfeld, S. & Candy, B. (2006) Psychosocial work environment and mental and veterinarians in rural communities undergoing livestock depopulation. Journal of the health – a meta-analytic review. Scandinavian Journal of Work, Environment and Health 32, American Veterinary Medical Association 231, 692-694 443-462 O’Connor, R. C. (2007) The relations between perfectionism and suicidality: a sys- Stark, C., Belbin, A., Hopkins, P., Gibbs, D., Hay, A. & Gunnell, D. (2006) tematic review. Suicide and Life-Threatening Behavior 37, 698-714 Male suicide and occupation in Scotland. Health Statistics Quarterly 29, 26-29 Parker, J. D. A., Creque, R. E., Sr, Barnhart, D. L., Harris, J. I., Majeski, Sterken, J., Troubleyn, J., Gasthuys, F., Maes, V., Diltoer, M. & S. A., Wood, L. M., Bond, B. J. & Hogan, M. J. (2004) Academic achievement Verborgh, C. (2004) Intentional overdose of Large Animal Immobilon. European in high school: does emotional intelligence matter? Personality and Individual Differences Journal of Emergency Medicine 11, 298-301 37, 1321-1330 TANNENBAUM, J. (1993) Veterinary medical ethics: a focus of conflicting interests. PAUL, E. S. & Podberscek, A. L. (2000) Veterinary education and students’ attitudes Journal of Social Issues 49, 143-156 towards animal welfare. Veterinary Record 146, 269-272 Timmins, R. P. (2006) How does emotional intelligence fit into the paradigm of veteri- Peck, D. F. (2005) Foot and mouth outbreak: lessons for mental health services. Advances nary medical education? Journal of Veterinary Medical Education 33, 71-75 in Psychiatric Treatment 11, 270-276 TORRE, D. M., WANG, N. Y., MEONI, L. A., YOUNG, J. H., KLAG, M. J. & FORD, PECK, D. F., GRANT, S., MCARTHUR, W. & GODDEN, D. (2002) Psychological D. E. (2005) Suicide compared to other causes of mortality in physicians. Suicide and impact of foot-and-mouth disease on farmers. Journal of Mental Health 11, 523-531 Life-Threatening Behavior 35, 146-153 Petersen, M. R. & Burnett, C. A. (2008) The suicide mortality of working physi- Trimpop, R., Austin, E. J. & Kirkcaldy, B. D. (2000a) Occupational and traffic cians and dentists. Occupational Medicine 58, 25-29 accidents among veterinary surgeons. Stress Medicine 16, 243-257 Petrides, K. V., Frederickson, N. & Furnham, A. (2004) The role of trait TRIMPOP, R., KIRKCALDY, B., ATHANASOU, J. & COOPER, C. (2000b) Individual emotional intelligence in academic performance and deviant behavior at school. differences in working hours, work perceptions and accident rates in veterinary surger- Personality and Individual Differences 36, 277-293 ies. Work and Stress 14, 181-188 Phillips-Miller, D. L., Campbell, N. J. & Morrison, C. R. (2000) Work tyssen, r., dolatowski, f. c., røvik, j. o., thorkildsen, r. f., and family: satisfaction, stress, and spousal support. Journal of Employment Counseling ekeberg, Ø., hem, e., gude, t., Grønvold, N. T. & Vaglum, P. (2007) 37, 16-30 Personality traits and types predict medical school stress: a six-year longitudinal and Phillips-Miller, D., Morrison, C. & CAMPBELL, N. J. (2001) same profession, nationwide study. Medical Education 41, 781-787 different career: a study of men and women in veterinary medicine. In Advances in Tyssen, R., Hem, E., Vaglum, P., Grønvold, N. T. & Ekeberg, Ø. (2004) Psychology Research. Vol 5. Ed F. Columbus. Nova Science. pp 1-53 The process of suicidal planning among medical doctors: predictors in a longitudinal Platt, S. & Hawton, K. (2000) Suicidal behaviour and the labour market. In The Norwegian sample. Journal of Affective Disorders 80, 191-198 International Handbook of Suicide and Attempted Suicide. Eds K. Hawton, K. van Tyssen, R., Vaglum, P., Grønvold, N. T. & Ekeberg, Ø. (2001a) Factors in Heeringen. Wiley. pp 309-384 medical school that predict postgraduate mental health problems in need of treatment: PTACEK, J. T., LEONARD, K. & MCKEE, T. L. (2004) ‘I’ve got some bad news…’: a nationwide and longitudinal study. Medical Education 35, 110-120 veterinarians’ recollections of communicating bad news to clients. Journal of Applied Tyssen, R., Vaglum, P., Grønvold, N. T. & Ekeberg, Ø. (2001b) Suicidal Social Psychology 34, 366-390 ideation among medical students and young physicians: a nationwide and prospective Raekallio, M., Heinonen, K. M., Kuussaari, J. & Vainio, O. (2003) Pain study of prevalence and predictors. Journal of Affective Disorders 64, 69-79 alleviation in animals: attitudes and practices of Finnish veterinarians. Veterinary Journal VAN HEERINGEN, K. (2002) Towards a psychobiological model of the suicidal proc- 165, 131-135 ess. In Understanding Suicidal Behaviour: the Suicidal Process Approach to Research, RCVs (2007) Annual Report 2007. Royal College of Veterinary Surgeons Treatment and Prevention. Ed K. van Heeringen. Wiley. pp 136-159 Reeve, C. L., Rogelberg, S. G., Spitzmüller, C. & DiGiacomo, N. (2005) VBF (2007) Annual Report and Accounts 2006. Veterinary Benevolent Fund The caring-killing paradox: euthanasia-related strain among animal-shelter workers. Voracek, M. (2004) National intelligence and suicide rate: an ecological study of 85 Journal of Applied Social Psychology 35, 119-143 countries. Personality and Individual Differences 37, 543-553 REICHENBERG, A. & MACCABE, J. H. (2007) Feeling the pressure: work stress and Waldenström, K., AHLBERG, G., BERGMAN, P., FORSELL, Y., STOETZER, mental health. Psychological Medicine 37, 1073-1074 U., Waldenström, M. & LUNDBERG, I. (2008) Externally assessed psycho- REIJULA, K., Räsänen, K., HäMäLäINEN, M., JUNTUNEN, K., LINDBOHM, social work characteristics and diagnoses of anxiety and depression. Occupational and M., TASKINEN, H., BERGBOM, B. & RINTA-JOUPPI, M. (2003) Work environ- Environmental Medicine 65, 90-96 ment and occupational health of Finnish veterinarians. American Journal of Industrial White, A., Shiralkar, P., Hassan, T., Galbraith, N. & Callaghan, R. Medicine 44, 46-57 (2006) Barriers to mental healthcare for psychiatrists. Psychiatric Bulletin 30, 382-384 RICE, D. (2008) Suicide by veterinary surgeons. Veterinary Record 162, 355 WIECLAW, J., AGERBO, E., MORTENSEN, P. B. & BONDE, J. P. (2005) Occupational Robinson, D. & Hooker, H. (2006) The UK Veterinary Profession in 2006: the risk of affective and stress-related disorders in the Danish workforce. Scandinavian Journal Findings of a Survey of the Profession Conducted by the Royal College of Veterinary of Work, Environment and Health 31, 343-351 Surgeons. Royal College of Veterinary Surgeons WIECLAW, J., AGERBO, E., MORTENSEN, P. B. & BONDE, J. P. (2006) Risk of Rogelberg, S. G., DiGiacomo, N., Reeve, C. L., Spitzmuller, C., Clark, affective and stress-related disorders among employees in human service professions. O. L., Teeter, L., Walker, A. G., Carter, N. T. & Starling, P. G. (2007) Occupational and Environmental Medicine 63, 314-319 What shelters can do about euthanasia-related stress: an examination of recommenda- Wilhelm, K., Kovess, V., Rios-Seidel, C. & Finch, A. (2004) Work and mental tions from those on the front line. Journal of Applied Animal Welfare Science 10, 331-347 health. Social Psychiatry and Psychiatric Epidemiology 39, 866-873 ROLLIN, B. E. (1987) Euthanasia and moral stress. Loss, Grief and Care 1, 115-126 WILLIAMS, D. (2006) Pressures of student debt. Veterinary Record 158, 383 Rollin, B. E. (2006) An Introduction to Veterinary Medical Ethics: Theory and Cases. WILLIAMS, J. M. G. (2001) Suicide and attempted suicide. Penguin 2nd edn. Blackwell WILLIAMS, J. M. G. & POLLOCK, L. R. (2000) The psychology of suicidal behaviour. ROUTLY, J. E., TAYLOR, I. R., TURNER, R., MCKERNAN, E. J. & DOBSON, H. In The International Handbook of Suicide and Attempted Suicide. Eds K. Hawton, K. (2002) Support needs of veterinary surgeons during the first few years of practice: per- van Heeringen. Wiley. pp 79-93 ceptions of recent graduates and senior partners. Veterinary Record 150, 167-171 WILLIAMS, S. M., ARNOLD, P. K. & MILLS, J. N. (2005) Coping with stress: a survey SARTORIUS, N. (2007) Stigma and mental health. Lancet 370, 810-811 of Murdoch University veterinary students. Journal of Veterinary Medical Education 32, Schernhammer, E. S. & Colditz, G. A. (2004) Suicide rates among physicians: 201-212 a quantitative and gender assessment (meta-analysis). American Journal of Psychiatry 161, WORLEY, L. L. M. (2008) Our fallen peers: a mandate for change. Academic Psychiatry 2295-2302 32, 8-12 Seale, C. (2009) Legalisation of euthanasia or physician-assisted suicide: survey of doc- Yip, P. S. F., Yang, K. C. T., Ip, B. Y. T., Law, Y. W. & Watson, R. (2007) Financial tors’ attitudes. Palliative Medicine 23, 205-212 debt and suicide in Hong Kong SAR. Journal of Applied Social Psychology 37, 2788-2799 Serpell, J. A. (2005) Factors influencing veterinary students’ career choices and atti- Zemaitiene, N. & Zaborskis, A. (2005) Suicidal tendencies and attitude towards tudes to animals. Journal of Veterinary Medical Education 32, 491-496 freedom to choose suicide among Lithuanian schoolchildren: results from three cross- Sheehy, N. & O’Connor, R. C. (2008) Cognitive style and suicidal behaviour. sectional studies in 1994, 1998 and 2002. BMC Public Health 5, 83 In Suicide: Strategies and Interventions for Reduction and Prevention. Ed S. Palmer. ZENNER, D., BURNS, G. A., RUBY, K. L., DEBOWES, R. M. & STOLL, S. K. (2005) Routledge. pp 69-83 Veterinary students as elite performers: preliminary insights. Journal of Veterinary Medical SIEGRIST, J. (1996) Adverse health effects of high-effort/low-reward conditions. Journal of Education 32, 242-248

March 27, 2010 | Veterinary Record

Online P&A March 27.indd 397 24/3/10 15:42:01 Papers Papers Papers Interventions with potential to improve the mental health and wellbeing of UK veterinary surgeons

D. J. Bartram, J. M. A. Sinclair, D. S. Baldwin

The proportion of UK veterinary surgeons who die by suicide as opposed to other causes is approximately four times that of the general population, and around twice that of other healthcare professionals. Recent research suggests that veterinary surgeons report high levels of psychological distress. This paper proposes a portfolio of evidence-based interventions, for both organisations and individuals, which have the potential to improve mental health and wellbeing in the veterinary profession.

Veterinary surgeons in the UK are known to be at increased risk A postal questionnaire survey of a large stratified random sample of death by suicide, compared with the general population (Charlton of veterinary surgeons practising within the UK was conducted to and others 1993, Kelly and others 1995, Kelly and Bunting 1998, assess the contribution of mental health and wellbeing to the elevated Mellanby 2005, Stark and others 2006, Meltzer and others 2008), risk of suicide. Compared with the general population, the sample with a proportional mortality ratio for suicide around four times that reported high levels of anxiety and depressive symptoms; a higher of the general population and around twice that of other healthcare 12-month prevalence of suicidal thoughts; less favourable psychoso- professions. Elevated mortality from suicide has also been reported cial working characteristics, especially in regard to demands and man- in the USA (Miller and Beaumont 1995), Australia (Jones-Fairnie and agerial support; lower levels of positive mental wellbeing; and higher others 2008) and Norway (Hem and others 2005). levels of negative work-home interaction (Bartram and others 2009b). There has been much speculation regarding the possible mecha- The level of alcohol consumption does not appear to be a negative nisms underlying increased suicide risk in the profession, but little influence on mental health within the profession as a whole (Bartram empirical research. A complex interaction of possible mechanisms and others 2009a, b). The levels of psychological distress reported sug- may occur across the career life course of a veterinary surgeon, lead- gest that ready access to, and knowledge of, lethal means is probably ing to increased suicide risk. Possible factors include the character- not operating in isolation to increase suicide risk within the profession istics of individuals entering the profession, negative effects during (Bartram and others 2009b). This paper aims to provide a portfolio undergraduate training, work-related stressors (such as long working of suggested evidence-based interventions, for both organisations and hours, inadequate support, emotional exhaustion, client expectations individuals, which have the potential to improve mental health and and unexpected clinical outcomes), ready access to and knowledge of wellbeing in the veterinary profession. means of self-harm (medicines are typically stored in practice premises, and deliberate self-poisoning is the most common method of suicide Implications for the profession in both male and female veterinarians), stigma associated with mental Interventions to improve psychological health have been implemented illness, professional and social isolation, and alcohol or drug misuse effectively in a variety of different work settings (Michie and Williams (mainly prescription drugs to which the profession has ready access, 2003). An occupation-specific suicide prevention programme imple- such as ketamine, benzodiazepines and opiates). Contextual effects mented in the US Air Force – aimed at decreasing stigma, enhancing such as attitudes to death and euthanasia (formed through the profes- social networks, facilitating help-seeking, and enhancing understand- sion’s routine involvement with euthanasia of companion animals ing of mental health – reported success in lowering suicide rates (Knox and slaughter of farm animals) and suicide ‘contagion’ (due to direct or and others 2003). indirect exposure to the suicide of peers within this small profession) Studies suggest that poor mental health among doctors is associ- are other possible influences (Bartram and Baldwin 2008a, b). ated with, for example, being irritable with patients, taking short cuts and not following procedures, which may markedly reduce standards of patient care (Firth-Cozens 2001, Taylor and others 2007). Given the similarities between medical and veterinary practice, intervention Veterinary Record (2010) 166, 518-523 doi: 10.1136/vr.b4796 should be important, not only for the wellbeing of individual members of the veterinary profession, but also to safeguard veterinary public D. J. Bartram, BVetMed, DipM, MCIM, E-mail for correspondence: health and the health and welfare of animals under veterinary care. CDipAF, FRCVS, [email protected] The structure of the UK veterinary profession is highly J. M. A. Sinclair, BSc, MB, BS, MSc, fragmented. Most practices are small businesses without the benefit MRCPsych, DPhil, Provenance: not commissioned; of dedicated human resources staff or occupational health profes- D. S. Baldwin, MB, BS, DM, FRCPsych, externally peer reviewed sionals to advance improvements in working conditions and the Division of Clinical Neurosciences: management and health of employees. Employers have ethical and Mental Health Group, School of legal duties to assess the risk of stress-related ill health arising from Medicine, University of Southampton, work activities and take measures to control that risk. However, Royal South Hants Hospital, Brintons a balance must be achieved between personal and organisational Terrace, Southampton SO14 0YG responsibility for mental health and wellbeing; a comprehensive

Veterinary Record | April 24, 2010

Online P&A Apriil 24.indd 518 21/4/10 15:51:13 Papers Papers

TABLE 1: Proposed individual-orientated interventions with to justify prompt concerted action, and is in accordance with the asser- the potential to help alleviate psychological distress among tions of Rose (1992) on preventive medicine: veterinary surgeons ‘Certainty is not a prerequisite for action. … that action may then pro- Intervention ceed alongside continuing research and evaluation, recognising that Primary new evidence may mean a change of policy.’ Enhanced assessments at undergraduate level to inform the development, timing Employee participation in the process of introducing interventions and targeting of educational interventions to improve resilience during the subsequent career can help to refine interventions to optimise their fit within the con- Consider screening undergraduates to identify at-risk individuals and provide text of a specific workplace, and may in itself help to diminish some appropriate levels of support dimensions of work-related stress (LaMontagne and others 2007). Mental health promotion: techniques to improve mental wellbeing, enhance The following are possible areas for intervention to improve the adaptive coping skills, increase awareness of the dangers of alcohol misuse, and mental health and wellbeing of veterinary surgeons, based on evidence reduce perceived stigma Education to facilitate early recognition of psychological problems in self and others of their need within the profession and efficacy of such interventions Consultation skills training to improve communication and thereby help to manage in other fields. client expectations and handle complaints effectively Introduction of emotional reflection into the Professional Development Phase Possible areas for intervention Increase awareness of existing support initiatives: Vet Helpline, Veterinary Surgeons’ Mental health promotion Health Support Working Party, Vetlife website Advice for veterinary surgeons considering alternative careers within or outside the Educational initiatives to improve mental health and wellbeing profession could be integrated into the early undergraduate curriculum and be Secondary made accessible to those already working in the profession. Topics Improving help-seeking behaviour, including encouragement to use existing support could include evidence-based techniques to: enhance intra- and initiatives interpersonal adaptive coping skills; enable early recognition of Support during complaints and disciplinary proceedings Provision of a service to help resolve employment-related disputes mental ill health in oneself and others; recognise and engage sui- Litigation under employment law cidal colleagues and help them to stay safe and seek further assist- Tertiary ance; increase awareness of the dangers of alcohol misuse; challenge Counselling and psychotherapy stigma and discrimination; and enhance help-seeking behaviours. Peer-led tuition could be provided by a national network of suitably trained veterinary surgeons. Primary interventions at undergraduate programme of interventions including both individual-orientated level are likely to have the greatest long-term benefit and be among and work environment-orientated elements has been shown to the easiest to implement (Gerrity 2001). Timely intervention with deliver the most favourable outcomes (Bond 2004, LaMontagne and young people who show early signs of mental ill health is impor- others 2007). tant, as younger age of onset of depression is associated with greater Rose (1992) observed that the prevalence of many diseases in a suicidal intent, irrespective of the age of onset of suicidal thoughts population is directly related to the population mean of the under- (Thompson 2008). lying risk factors: when there is a small shift in the mean of a nor- Consideration could be given to the production of a short educa- mal distribution there is a large shift in the tails of the distribution. tional video for use in veterinary schools and conferences as part of an Consequently, shifting the mean of the risk factors in the population outreach campaign to raise awareness of suggested measures to reduce by population-level interventions leads to a decline in the prevalence the levels of psychological distress in the profession. of the disease. The prevalence of a psychiatric disorder in a population is linearly related to the mean number of psychiatric symptoms in that Monitoring of trends population (Anderson and others 1993, Whittington and Huppert A recent cross-sectional study of mental health and wellbeing among 1996). Anderson and others (1993) concluded: veterinary surgeons in the UK (Bartram and others 2009b) provides a baseline of measures from which to monitor future change and trends ‘Populations thus carry a collective responsibility for their own mental in mental health and wellbeing. The RCVS Survey of the Profession health and well-being. This implies that explanations for the differing could provide a suitable medium for such monitoring as it is mailed to prevalence rates of psychiatric morbidity must be sought in the charac- the entire membership and achieves a response rate of approximately teristics of their parent populations; and control measures are unlikely 50 per cent (Robinson and Hooker 2006). The Warwick-Edinburgh to succeed if they do not involve population-wide changes.’ Mental Wellbeing Scale (WEMWBS) (Tennant and others 2007) It is important, therefore, that interventions within the veterinary would be an appropriate instrument for inclusion in the questionnaire profession are directed at the level of the entire occupational popula- as it is designed to measure population-level changes in mental well- tion, rather than merely targeting those in psychological distress. being and the score is correlated with anxiety and depressive symp- The cross-sectional nature of studies of mental health in the pro- toms, suicidal thoughts and psychosocial working conditions among fession (for example, Gardner and Hini 2006, Bartram and others veterinary surgeons (Bartram and others 2010). A shorter seven-item 2009b, Fritschi and others 2009) precludes inference of causality, but version of WEMWBS is also available, and benefits from brevity and the results suggest a number of primary (preventive), secondary (amel- more robust scaling properties than the full scale (Stewart-Brown and iorative) and tertiary (reactive) interventions, for both organisations others 2009). Monitoring would help in the evaluation of effective- and individuals, which have the potential to reduce psychological dis- ness of any interventions introduced. tress in the profession. In this paper, particular consideration is given to the concept of ‘countervailing interventions’ – those focused on Accessible and appropriate support services increasing the positive experience of work rather than decreasing the There is a need to evaluate the effectiveness of existing initiatives negative aspects – by incorporating some of the sources of satisfaction to support mental health in the UK veterinary profession, such as in veterinary work identified by Bartram and others (2009c), as these Vet Helpline and Veterinary Surgeons’ Health Support Working may help counteract the effects of workplace stressors (Kelloway and Party, to determine whether modification or additional provision others 2008). Proposed interventions are summarised in Tables 1 and is required. Consideration could be given to the introduction of a 2. Research is required to assess the effectiveness of these proposed telephone counselling service, possibly as part of an employee assist- interventions on the dual outcomes of the mental health of veteri- ance programme provided by individual veterinary employers or nary surgeons and their work performance in terms of animal and cli- associated with membership of a veterinary professional organisa- ent care (for example, reduction in complaints and litigation). This tion, such as the British Veterinary Association (BVA). Similarly, a approach applies the precautionary principle in recognising the need service to facilitate employment dispute resolution and encourage for further research on interventions while simultaneously advocating better employment relations within the veterinary profession has that there is adequate evidence from outside the veterinary profession potential to improve working life.

April 24, 2010 | Veterinary Record

Online P&A Apriil 24.indd 519 21/4/10 15:51:13 Papers Papers

TABLE 2: Proposed organisation-orientated interventions with ers 2005). Interpersonal relationships are a central theme of mental the potential to help alleviate psychological distress among health promotion programmes (Stewart-Brown 2005). Durkheim veterinary surgeons (1897) observed over a century ago that the frequency of suicide varied inversely with social cohesion: Intervention

Primary ‘… we might observe that since the strength of the collective is one Regular population-level monitoring of mental health and wellbeing of the of the obstacles that best limits suicide, this cannot weaken without veterinary profession using valid and reliable instruments suicide rising.’ Regular monitoring of psychosocial work characteristics of individual workplaces: ‘… in a coherent and vital community, there is a continual iterative process of measurement and modification exchange of ideas and feelings from all to each and from each to all Providing a safe means of recognising and discussing concerns about colleagues which is like mutual moral support, so that the individual, instead Improving management skills for senior veterinary surgeons: better supervision skills, recognition of risk, team leadership, regular staff appraisals, improved of being reduced to his resources only, participates in the collective communication of change energy and draws on it when his own is exhausted.’ Increase management support, including mentoring programmes to assist new graduates to gain confidence While it may be facile to suggest that measures to improve social Increase participation in decision-making support within the veterinary profession would make suicide less Changing work practices: reduction in working hours, use of dedicated providers of likely, the possibility should not be discounted. out-of-hours veterinary care, regular breaks, adequate clerical support The structure of the veterinary profession is presently undergoing Evaluation of the effectiveness of existing initiatives to support mental health Provide opportunities to enhance interpersonal relationships with colleagues considerable changes, including increasing numbers of female gradu- and clients ates, increasing numbers of practices in corporate ownership, a decline Opportunities for improved work-home interaction, for example, flexible in traditional mixed practice in favour of specialisation, and a trend work options towards consolidation of farm animal practices into fewer units cover- Establish wellbeing as an essential value of the organisation to create the ing larger geographical regions. Higher perceived levels of workplace appropriate culture Restoration of confidence in RCVS complaints and disciplinary processes change have negative effects on psychological health (for example, Secondary Loretto and others 2005). Perceived coworker support and accurate Provision of accessible and appropriate support services, for example, counselling, information about impending change can reduce the impact on health cognitive behavioural therapy, telephone helplines (Platt and others 1999). Organisations could provide employees with Reduce discriminatory work practices timely information to enable them to understand the reasons for pro- Adjustments to the work environment to enable an individual to continue working Tertiary posed changes and ensure adequate employee consultation on changes Policies to facilitate rehabilitation and reintegration back into the workplace and opportunities for employees to influence proposals. Consider controlling access to specific medicines All employers have a legal responsibility under existing UK Signage positioned in close proximity to medicine stores health and safety at work legislation to ensure the health, safety and Intervention after a suicide to support those affected and reduce the risk of contagion welfare of their employees, and are required to undertake risk assessments and take reasonably practicable steps to address the risks identified; however, failings in active health risk management systems in Working conditions UK small animal veterinary practices have been reported (D’Souza Veterinary surgeons report less favourable psychosocial work charac- and others 2009). The legal responsibility includes minimising the teristics than the general working population in relation to demands risk of stress-related illness. There is a need to ensure that attention and managerial support (Bartram and others 2009b, c). There is con- to mental health, and not merely physical health, is recognised by sistent evidence from a number of cross-sectional and longitudinal veterinary employers as an integral requirement of the legislation. studies that high psychological demands, low decision latitude (con- The incorporation of formalised procedures for surveillance of work- trol) and low social support at work are predictive of mental ill health place-specific psychosocial risks, and the development and imple- (Stansfeld and Candy 2006, Bonde 2008, Netterstrøm and others mentation of suitable plans for their mitigation, into the require- 2008). Working conditions are potentially modifiable and improvements for accreditation under the RCVS Practice Standards Scheme ments can enhance mental health and wellbeing. Successful interven- is worthy of consideration, especially in view of the potential effects tions that improve psychological health use training and organisation- of the mental health and wellbeing of veterinary surgeons on animal al approaches to increase participation in decision-making, increase care. By the omission of any reference to work-related stress in the support and feedback, and improve communication (Michie and RCVS Practice Standards Scheme requirements relating to health and Williams 2003). safety of employees (RCVS 2009), the RCVS may be inadvertently Veterinary surgeons in the UK report relationships with col- reinforcing any perception among employers that health and safety leagues and clients as important sources of work-related satisfaction legislation relates only to physical risks. Bartram and Turley (2009) (Bartram and others 2009c). Consequently, encouraging both formal provided guidance on risk assessment for work-related stress in the and informal supportive relationships between work colleagues and context of veterinary practices using an approach advocated by the with clients has the potential to improve mental health and well- Health and Safety Executive. Further materials could be developed being. The development of a workplace culture in which there is that present a compelling business case for psychological health and regular constructive feedback and where problems are addressed safety at work, beyond ethical and legal obligations, and inform vet- sensitively could generate a more supportive work environment. erinary businesses on how to create and sustain working environ- Opportunities for clinical supervision, mentorship and review ments that optimise the mental health of employees. Workplaces could be expanded and collaborative team-working encouraged. implementing specific measures to improve the mental health and The importance of workplace relationships has been recently high- wellbeing of veterinary surgeons should be identified and publicised lighted by research associating a poor team climate at work with as models of best practice. depressive disorders and subsequent antidepressant use (Sinokki and The RCVS Professional Development Phase aims to support new others 2009). Deficiencies in people-management skills, including graduates during their attainment of clinical skills in their first year in supervision, team leadership and competence in strategies for the practice. The incorporation of elements of emotional reflection into prevention and reduction of work-related stress among employees, this phase would give recognition to the importance of psychologi- could be identified and addressed in veterinary practices. Meetings to cal health, and encourage new graduates to consider their emotional discuss stressful work experiences appear to protect against suicidal needs and development opportunities early in their careers. thoughts (ideation) among female physicians (Fridner and others 2009), which suggests that regular implementation of such meetings Other personal and work-related stressors may have similar value among veterinary surgeons. The total number of hours worked and out-of-hours on-call duties are High levels of job satisfaction have been shown to protect doctors’ self-reported by veterinary surgeons to be major contributors to stress mental health against the harmful effects of job stress (Taylor and oth- (Bartram and others 2009c), although, in keeping with studies among

Veterinary Record | April 24, 2010

Online P&A Apriil 24.indd 520 21/4/10 15:51:14 Papers Papers

doctors (Tyssen and Vaglum 2002), a recent study (Bartram and oth- should be given to mechanisms to moderate the opportunity for veteri- ers 2009c) failed to show an association between working long hours nary surgeons to use certain medicines for suicide at times of high risk, and mental health and wellbeing. Nevertheless, the requirements of the notwithstanding the immediate implications for dispensing and admin- Working Time Regulations (Anon 1998) – which prescribe maximum istration of medicines. In other professions where access to lethal means working hours and minimum rest periods – should be observed. The is inevitable, such as the use of firearms by the military, at times of high related issue of whether inactive periods during on-call duties contribute risk, restrictions are placed on the opportunity to use lethal means for towards the total number of hours worked requires prompt resolution. suicide (for example, solitary armed duty), as distinct from imposing Making professional mistakes, the possibility of client com- blanket restrictions on access (Mahon and others 2005). This model plaints and/or litigation, and client expectations are reported to be may have application in the veterinary profession; for example, restric- major contributors to stress among veterinary surgeons in the UK tion of unsupervised access to injectable barbiturates at times of high (Bartram and others 2009b, c). Consultation skills training focused risk to minimise opportunity to remove them from the workplace for on improving competencies in communication to manage client self-administration. expectations and deal with complaints may help to mitigate these Placement of signage at geographical suicide ‘hot spots’, display- stressors. The curricula of some UK veterinary schools contain ele- ing details of accessible sources of emotional support, can be an effec- ments of professional communication skills training (Gray and oth- tive suicide prevention measure (for example, King and Frost 2005). ers 2006), but there is scope for this to be greatly expanded. The Signage is a low-cost intervention that capitalises on existing services, RCVS and professional indemnity insurance providers could give can be promptly implemented and is based on robust evidence of effec- consideration to how to improve the access of veterinary surgeons tiveness. Signs displaying helpline telephone numbers, positioned in to appropriate advice and support when cases are referred to them. close proximity to practice medicine stores, might have a similar effect There is a general consensus on a need to update the RCVS com- on individuals considering suicide, and could also serve to increase plaints and disciplinary procedures in the interests of both the public awareness among all practice employees of sources of support. and the profession (House of Commons Environment, Food and Consideration could be given by veterinary employers and profes- Rural Affairs Committee [EFRACom] 2008). Efforts to improve vet- sional bodies to the development and implementation of policies to erinary surgeons’ respect for and confidence in the procedures could facilitate the reintegration of veterinary surgeons back into the work- relieve some of their associated concerns. place with respect and sensitivity following a mental health crisis. Managing personal finances is self-reported as an important stressor Consideration could also be given to the development of a pro- for recent veterinary graduates (Bartram and others 2009c). It is likely tocol for effective intervention after a suicide to provide guidance that concerns will rise among future graduates given the introduction to veterinary employers on the communication of information to of tuition fees in 2006 and the prospect that fees may be increased. The employees relating to a colleague’s suicide and provision of support provision of debt management guidance during undergraduate training to all those affected. has the potential to help reduce the stress associated with the repayment of student loans after graduation. Consideration could be given to lob- Future research bying government to allow students’ costs associated with compulsory A qualitative interview-based study of veterinary surgeons who have extramural studies to be incorporated into the tuition fee loan structure. experienced suicidal ideation may identify trends in factors considered to have influenced those thoughts, and could explore help-seek- Work-home interaction ing behaviours and perceived barriers to seeking help. Such mixed- Some veterinary surgeons experience high levels of negative work- methods research, integrating qualitative and quantitative data, holds home interaction (Bartram and others 2009b). Renewed attention potential for a more comprehensive understanding of mental health to existing policies and practices is recommended to address the and wellbeing in the profession. extent to which work and home life are in conflict for both men and Prospective longitudinal studies would help to identify factors women in the profession. This may be achieved through flexible with a causal association with mental health. Studies reported to working hours, reduced working and on-call hours, or job sharing, date focus on practising veterinary surgeons. However, each stage of as negative work-home interaction is associated with the number of the veterinary career path – from the characteristics of applicants to hours worked and on call. The formation of out-of-hours coopera- veterinary schools, undergraduate training, subsequent employment, tives with local practices or the use of dedicated providers of out-of- and through to retirement – requires examination, to identify early hours veterinary care has potential to reduce negative work-home predisposing factors and later triggers for suicidal behaviour in the pro- interaction. fession. This ‘life course’ approach to studying suicide (Gunnell and Lewis 2005) will enable multiple points on the career continuum to be Crisis intervention and return to work targeted with appropriate interventions. Attempts could be made to facilitate the early identification of signs Research with veterinary students could help to identify whether of psychological distress at work and enable appropriate and support- there is a predilection towards mental health problems in applicants ive intervention, which may involve making reasonable adjustments to veterinary school, whether there are negative influences on mental to the work environment to enable an individual to continue work- health during undergraduate training, and whether individual-specific ing. Screening workers for early signs of depression, and encourag- maladaptive coping strategies might play a role in the development of ing effective treatment, can improve clinical outcomes and work mental ill health. performance (Wang and others 2007). Associated initiatives to help An investigation examining the coroners’ records for veterinary break the stigma of mental health as a character weakness would be surgeons who completed suicide or died by an undetermined cause complementary. would provide insight into the circumstances of suicides in the profes- Employers could create a positive culture of acceptability of sick- sion and help to identify proximal risk factors. It has been suggested ness absence that enables veterinary surgeons to take time off work that attitudes towards suicide (the degree to which an individual views without excessive feelings of guilt, encourages them to recover suf- suicide as an acceptable option in some circumstances) may moderate ficiently before returning to work, and manages the consequences of the link between hopelessness and depressive symptoms and levels of individual absences for their colleagues in a fair and equitable way. suicidal ideation (Gibb and others 2006). Attitudes can also influence The elevated levels of psychological distress reported in the UK vet- help-seeking behaviour. The veterinary profession’s role in the provi- erinary profession suggest that the ready availability of lethal means is sion of animal euthanasia and the facilitation of a ‘good death’ may probably not operating in isolation to increase suicide risk within the normalise suicide, with death perceived as a rational solution to intracta- profession (Bartram and others 2009b). However, in view of the over- ble problems; however, there has been no rigorous examination of this whelming evidence of an association between suicide rates and avail- hypothesis to date. ability of lethal means (Hawton 2007), and the effectiveness of restric- Translational research is required to develop interventions, which tions on access to lethal means as an approach to suicide prevention will then need to be piloted and their effectiveness and utility evalu- (Daigle 2005, Mann and others 2005, Hawton 2007), consideration ated across the range of different veterinary work settings.

April 24, 2010 | Veterinary Record

Online P&A Apriil 24.indd 521 21/4/10 15:51:14 Papers Papers

Conclusions Firth-Cozens, J. (2001) Interventions to improve physicians’ well-being and patient care. Social Science & Medicine 52, 215-222 The observations from recent research and the proposals for interven- FRIDNER, A., BELKIC, K., MARINI, M., MINUCCI, D., PAVAN, L. & tion described in this paper have numerous implications for the vet- SCHENCK-GUSTAFSSON, K. (2009) Survey on recent suicidal ideation among female erinary profession, and ongoing advocacy will be required across a university hospital physicians in Sweden and Italy (the HOUPE study): cross-sectional range of stakeholders within the profession to demonstrate the need associations with work stressors. Gender Medicine 6, 314-328 FRITSCHI, L., MORRISON, D., SHIRANGI, A. & DAY, L. (2009) Psychological well- for and encourage the development and implementation of suitable being of Australian veterinarians. Australian Veterinary Journal 87, 76-81 interventions. The establishment of a single collaborative body with Gardner, D. H. & Hini, D. (2006) Work-related stress in the veterinary profession a mandate from all the key veterinary professional organisations to in New Zealand. New Zealand Veterinary Journal 54, 119-124 provide direction to, and coordination and integration of initiatives Gerrity, M. S. (2001) Interventions to improve physicians’ well-being and patient care: that attempt to improve psychological health within the veterinary a commentary. Social Science & Medicine 52, 223-225 Gibb, B. E., Andover, M. S. & Beach, S. R. H. (2006) Suicidal ideation and atti- profession could help to ensure a unified strategy and maximise tudes toward suicide. Suicide and Life-Threatening Behavior 36, 12-18 effectiveness. GRAY, C. A., BLAXTER, A. C., JOHNSTON, P. a., LATHAM, C. E., MAY, S., PHILLIPS, C. A., TURNBULL, N. & YAMAGISHI, B. (2006) Communication educa- Acknowledgements tion in veterinary education in the United Kingdom and Ireland: the NUVACS project coupled to progressive individual school endeavours. Journal of Veterinary Medical Education The authors thank the many individuals with whom they have 33, 85-92 shared ideas for potential interventions, and whose comments Gunnell, D. & Lewis, G. (2005) Studying suicide from the life course perspective: and suggestions have contributed to the development of these implications for prevention. British Journal of Psychiatry 187, 206-208 proposals. Hawton, K. (2007) Restricting access to methods of suicide: rationale and evaluation of this approach to suicide prevention. Crisis 28 (Suppl 1), 4-9 HEM, E., HALDORSEN, T., AASLAND, O. G., TYSSEN, R., VAGLUM, P. & Addendum EKEBERG, Ø. (2005) Suicide rates according to education with a particular focus on A number of the interventions proposed have been implemented physicians in Norway 1960-2000. Psychological Medicine 35, 873-880 or are in development since the original manuscript was submitted. Jones-Fairnie, H., Ferroni, P., Silburn, S. & Lawrence, D. (2008) Suicide This is testament to the highly commendable and ongoing efforts of in Australian veterinarians. Australian Veterinary Journal 86, 114-116 KELLOWAY, E. K., TEED, M. & KELLEY, E. (2008) The psychosocial environment: various veterinary professional bodies in the UK, including, but not towards an agenda for research. International Journal of Workplace Health Management 1, limited to, the RCVS, BVA, Veterinary Benevolent Fund and Vetlife 50-64 Steering Committee, to promptly address concerns regarding mental Kelly, S. & Bunting, J. (1998) Trends in suicide in England and Wales, 1982-96. health and wellbeing in the profession. Furthermore, there are pres- Population Trends 92, 29-41 ently related academic research programmes underway at the univer- KELLY, S., CHARLTON, J. & JENKINS, R. (1995) Suicide deaths in England and Wales, 1982-1992: the contribution of occupation and geography. Population Trends sities of Edinburgh, Oxford and Southampton and a number of the 80, 16-25 UK veterinary schools to inform the development of evidence-based KING, E. & FROST, N. (2005) The New Forest Suicide Prevention Initiative (NFSPI). interventions with potential to improve mental health and wellbeing Crisis 26, 25-33 in the veterinary profession. The intention is to measure the effective- Knox, K. L., Litts, D. A., Talcott, G. W., Feig, J. C. & Caine, E. D. (2003) Risk of suicide and related adverse outcomes after exposure to a suicide prevention ness of the portfolio of existing and future interventions to track the programme in the US Air Force: cohort study. British Medical Journal 327, 1376- progress of these collective efforts. 1378 LAMONTAGNE, A. D., KEEGEL, T., LOUIE, A. M., OSTRY, A. & LANDBERGIS, References P. A. (2007) A systematic review of the job-stress intervention evaluation litera- ANON (1998) Statutory Instrument 1998 Number 1833. Working Time Regulations ture, 1990-2005. International Journal of Occupational and Environmental Health 13, 268- 1998. www.opsi.gov.uk/si/si1998/19981833.htm. Accessed May 6, 2009 280 ANDERSON, J., HUPPERT, F. & ROSE, G. (1993) Normality, deviance and minor LORETTO, W., POPHAM, F., PLATT, S., PAVIS, S., HARDY, G., MACLEOD, L. & psychiatric morbidity in the community. A population-based approach to General GIBBS, J. (2005) Assessing psychological well-being: a holistic investigation of NHS Health Questionnaire data in the Health and Lifestyle Survey. Psychological Medicine employees. International Review of Psychiatry 17, 329-336 23, 475-485 MAHON, M. J., TOBIN, J. P., CUSAK, D. A., KELLEHER, C. & MALONE, K. M. Bartram, D. J. & Baldwin, D. S. (2008a) Veterinary surgeons and suicide: influ- (2005) Suicide among regular-duty military personnel: a retrospective case-control study ences, opportunities and research directions. Veterinary Record 162, 36-40 of occupation-specific risk factors for workplace suicide. American Journal of Psychiatry Bartram, D. J. & Baldwin, D. S. (2008b) Veterinary surgeons and suicide: a 162, 1688-1696 hypothetical model to explain risk. Proceedings of the 12th European Symposium on Mann, J. J., Apter, A., Bertolote, J., Beautrais, A., Currier, D., Haas, Suicide and Suicidal Behaviour. Glasgow, August 27 to 30, 2008. pp 57-58 A. & OTHERS (2005) Suicide prevention strategies – a systematic review. Journal of the BARTRAM, D. J., SINCLAIR, J. M. A. & BALDWIN, D. S. (2009a) Alcohol consump- American Medical Association 294, 2064-2074 tion among veterinary surgeons in the UK. Occupational Medicine 59, 323-326 Mellanby, R. J. (2005) Incidence of suicide in the veterinary profession in England BARTRAM, D. & TURLEY, G. (2009) Managing the causes of work-related stress. In and Wales. Veterinary Record 157, 415-417 Practice 31, 400-405 Meltzer, H., Griffiths, C., Brock, A., Rooney, C. & Jenkins, R. (2008) BARTRAM, D. J., YADEGARFAR, G. & BALDWIN, D. S. (2009b) A cross-sectional Patterns of suicide by occupation in England and Wales: 2001-2005. British Journal of study of mental heath and well-being and their associations in the UK veterinary pro- Psychiatry 193, 73-76 fession. Social Psychiatry and Psychiatric Epidemiology 44, 1075-1085 MICHIE, S. & WILLIAMS, S. (2003) Reducing work related psychological ill health and BARTRAM, D. J., YADEGARFAR, G. & BALDWIN, D. S. (2009c) Psychosocial work- sickness absence: a systematic literature review. Occupational and Environmental Medicine ing conditions and work-related stressors among UK veterinary surgeons. Occupational 60, 3-9 Medicine 59, 334-341 Miller, J. M. & Beaumont, J. J. (1995) Suicide, cancer and other causes of death BARTRAM, D. J., YADEGARFAR, G., SINCLAIR, J. M. A. & BALDWIN, D. S. (2010) among California veterinarians, 1960-1992. American Journal of Industrial Medicine 27, Validation of the Warwick-Edinburgh mental well-being scale (WEMWBS) as an over- 37-49 all indicator of population mental health and well-being in the UK veterinary profes- NETTERSTRøM, B., CONRAD, N., BECH, P., FINK, P., OLSEN, O., RUGULIES, R. & sion. Veterinary Journal doi: 10.1016/j.tvjl.2010.02.010 STANSFELD, S. (2008) The relation between work-related psychosocial factors and the BOND, F. W. (2004) Getting the balance right: the need for a comprehensive approach development of depression. Epidemiologic Reviews 30, 118-132 to occupational health. Work & Stress 18, 146-148 PLATT, S., PAVIS, S. & AKRAM, G. (1999) Changing labour market conditions and BONDE, J. P. E. (2008) Psychosocial factors at work and risk of depression: a system- health. A systematic literature review (1993-98). Report for the European Foundation atic review of the epidemiological evidence. Occupational and Environmental Medicine 65, for the Improvement of Living and Working Conditions. Office for Official Publications 438-445 of the European Communities. www.eurofound.europa.eu/pubdocs/1999/15/en/1/ CHARLTON, J., KELLY, S., DUNNELL, K., EVANS, B. & JENKINS, R. (1993) Suicide ef9915en.pdf. Accessed May 6, 2009 deaths in England and Wales: trends in factors associated with suicide deaths. Population RCVS (2009) RCVS Practice Standards Scheme Manual (March 2009). Section 9: Safety Trends 71, 34-42 Procedures. RCVS. pp 36-43 D’SOUZA, E., BARRACLOUGH, R., FISHWICK, D. & CURRAN, A. (2009) Robinson, D. & Hooker, H. (2006) The UK Veterinary Profession in 2006: The Management of occupational health risks in small-animal veterinary practices. Findings of a Survey of the Profession Conducted by the Royal College of Veterinary Occupational Medicine 59, 316-322 Surgeons. London, RCVS Daigle, M. S. (2005) Suicide prevention through means restriction: assessing the risk of ROSE, G. (1992) The Strategy of Preventive Medicine. Oxford University Press substitution. A critical review and synthesis. Accident Analysis & Prevention 37, 625-632 Sinokki, M., Hinkka, K., Ahola, K., Koskinen, S., Klaukka, T., Durkheim, É. (1897) On Suicide. Penguin Classics. English translation (2006). pp Kivimäki, M., Puukka, P., Lönnqvist, J. & Virtanen, M. (2009) The 225-226 association between team climate at work and mental health in the Finnish Health efracom (2008) Veterinary Surgeons Act 1966: Sixth Report of Session 2007-08. 2000 Study. Occupational and Environmental Medicine 66, 523-528 The Stationery Office. www.publications.parliament.uk/pa/cm200708/cmselect/ Stansfeld, S. & Candy, B. (2006) Psychosocial work environment and mental cmenvfru/348/348.pdf. Accessed May 6, 2009 health – a meta-analytic review. Scandinavian Journal of Work, Environment and Health 32,

Veterinary Record | April 24, 2010

Online P&A Apriil 24.indd 522 21/4/10 15:51:14 Papers Papers

443-462 PARKINSON, J., SECKER, J. & STEWART-BROWN, S. (2007) The Warwick- Stark, C., Belbin, A., Hopkins, P., Gibbs, D., Hay, A. & Gunnell, D. (2006) Edinburgh Mental Well-being Scale (WEMWBS): development and UK validation. Male suicide and occupation in Scotland. Health Statistics Quarterly 29, 26-29 Health and Quality of Life Outcomes 5, 63 STEWART-BROWN, S. (2005) Interpersonal relationships and the origins of mental THOMPSON, A. H. (2008) Younger onset of depression is associated with greater sui- health. Journal of Public Mental Health 4, 24-29 cidal intent. Social Psychiatry and Psychiatric Epidemiology 43, 538-544 STEWART-BROWN, S., TENNANT, A., TENNANT, R., PLATT, S., PARKINSON, TYSSEN, R. & VAGLUM, P. (2002) Mental health problems among young doc- J. & WEICH, S. (2009) Internal construct validity of the Warwick-Edinburgh Mental tors: an updated review of prospective studies. Harvard Review of Psychiatry 10, 154- Well-being Scale (WEMWBS): a Rasch analysis using data from the Scottish Health 165 Education Population Survey. Health and Quality of Life Outcomes 7, 15 Wang, P. S., Simon, G. E., Avorn, J., Azocar, F., Ludman, E. J., TAYLOR, C., GRAHAM, J., POTTS, H., CANDY, J., RICHARDS, M. & RAMIREZ, McCulloch, J., Petukhova, M. Z. & Kessler, R. C. (2007) Telephone A. (2007) Impact of hospital consultants’ poor mental health on patient care. British screening, outreach, and care management for depressed workers and impact on clini- Journal of Psychiatry 190, 268-269 cal and work productivity outcomes: a randomized controlled trial. Journal of the American TAYLOR, C., GRAHAM, J., POTTS, H. W. W., RICHARDS, M. A. & RAMIREZ, Medical Association 298, 1401-1411 A. J. (2005) Changes in mental health of UK hospital consultants since the mid-1990s. WHITTINGTON, J. E. & HUPPERT, F. A. (1996) Changes in the prevalence of psychi- Lancet 366, 742-744 atric disorder in a community are related to changes in the mean level of psychiatric TENNANT, R., HILLER, L., FISHWICK, R., PLATT, S., JOSEPH, S., WEICH, S., symptoms. Psychological Medicine 26, 1253-1260

April 24, 2010 | Veterinary Record

Online P&A Apriil 24.indd 523 21/4/10 15:51:15 Qual Life Res DOI 10.1007/s11136-012-0144-4

Further validation of the Warwick-Edinburgh Mental Well-being Scale (WEMWBS) in the UK veterinary profession: Rasch analysis

David J. Bartram • Julia M. Sinclair • David S. Baldwin

Accepted: 16 February 2012 Ó Springer Science+Business Media B.V. 2012

Abstract indicator of population mental health and well-being in this Purpose To examine the psychometric properties of the occupational group with elevated suicide risk. 14-item Warwick-Edinburgh Mental Well-being Scale (WEMWBS) in the UK veterinary profession by the appli- Keywords Psychometrics Á Mental health Á Veterinarians Á cation of Rasch analysis, and to assess the external construct Occupational health validity of the derived interval scale measurements. Methods Data sets were derived from two independent Abbreviations cross-sectional surveys of the veterinary profession CI Confidence interval (n = 8,829 and n = 1,796). Rasch analysis (n = 500) DIF Differential item functioning included response option thresholds ordering, tests of fit, HADS Hospital Anxiety and Depression Scale differential item functioning, targeting, response depen- HADS-A Anxiety subscale of Hospital Anxiety and dency, and person separation index (PSI). Unidimension- Depression Scale ality was evaluated by principal component analysis of HADS-D Depression subscale of Hospital Anxiety residuals. The findings were validated across further and Depression Scale subsamples from both data sets. The external construct HSE MSIT Health and Safety Executive Management validity of the Rasch-fitting item set was evaluated by Standards Indicator Tool associations with other measures of psychological health or ICC Item characteristic curve psychosocial work characteristics. PSI Person separation index Results Data for the original 14 items deviated signifi- RCVS Royal College of Veterinary Surgeons cantly from Rasch model expectations (chi-square = SD Standard deviation 558.2, df = 112, P = \0.001, PSI = 0.918). A unidi- SWEMWBS Short Warwick-Edinburgh Mental Well- mensional 7-item scale (Short WEMWBS, SWEMWBS) being Scale with acceptable fit to the model (chi-square = 58.8, df = WEMWBS Warwick-Edinburgh Mental Well-being 56, P = 0.104, PSI = 0.832) was derived by sequential Scale removal of the most misfitting items. The external con- WHI_N Negative work–home interaction struct validity of SWEMWBS was supported. Conclusions SWEMWBS has robust interval-level measurement properties which support its suitability as an Introduction

Rating scales for the measurement of psychological health in occupational groups are valuable tools to assess, for D. J. Bartram (&) Á J. M. Sinclair Á D. S. Baldwin example, the levels of work-related stress or the response Faculty of Medicine, University Department of Psychiatry, to interventions. It is a requirement that health rating University of Southampton, Academic Centre, College Keep, 4-12 Terminus Terrace, Southampton SO14 3DT, UK scales are a rigorous measure of the variables (aspects of e-mail: [email protected] health) they claim to quantify, if clinically meaningful 123 Qual Life Res interpretations are to be made from response data. The Rasch analysis uses a mathematical model to evaluate main aims of psychometric methods are to ascertain whe- the legitimacy of summing item scores to generate mea- ther it is legitimate to produce total scale scores from items, surements [8] and to compute a raw ordinal score to and the extent to which these scale scores measure the interval scale transformation [9]. Assessment of the psy- variable they purport to measure (validity), are free from chometric properties of WEMWBS by the application of random error (reliability), and are sensitive to changes in Rasch analysis to a UK adult general population sample the variable over time (responsiveness). demonstrated that a 7-item version of WEMWBS (Short In the UK, mortality due to suicide is higher in the vet- WEMWBS, SWEMWBS) conformed to the Rasch model erinary profession than in the general population, the pro- requirements for interval-level scaling. However, psycho- portional mortality ratio for suicide being around four times metric properties are not inherent to an instrument, but that of the general population and twice that of other estimates that may vary across different samples and set- healthcare professions [1, 2]. The relative risk of suicide tings [10, 11]. Quantitative group comparisons between, across occupational groups is often explained by differences for example, the general population and the veterinary in demographic factors, but veterinary surgeons have a profession, are not defensible if the instrument scores higher risk of suicide even when these are taken into account cannot be interpreted in the same way across the groups [3]. It is uncertain whether the increased risk derives from the concerned [12]. Variation in the psychometric properties of characteristics of individuals entering the profession, the instruments across different occupational groups has also work environment, or other factors known to influence sui- been reported [13, 14]. cide risk such as ready access to lethal means [1]. Consequently, it is appropriate to examine the psycho- Robust measurement of the psychological health of this metric properties of an instrument in populations which occupational group is an important first step towards sui- may differ from those in which the properties have been cide prevention. The Warwick-Edinburgh Mental Well- assessed previously. The veterinary profession is not rep- being Scale (WEMWBS) [4] is a 14-item instrument resentative of the general population in demographic developed for assessing positive mental well-being at a structure (e.g. ethnicity, employment status, educational general population level. The scale operationalises a wide attainment, income, and social grade) [15] or the levels of conception of mental well-being including positive affect self-reported mental health and well-being [16]. It is (i.e. hedonic aspects of well-being: feelings of optimism, therefore meaningful to evaluate the properties of the cheerfulness, and relaxation); psychological functioning WEMWBS in the context of this occupational group. (i.e. eudaimonic aspects of well-being: energy, clear The aims of this study were to (a) evaluate the psycho- thinking, self-acceptance, personal development, compe- metric properties of the WEMWBS using Rasch analysis, tence, and autonomy); and interpersonal relationships [4]. (b) compute a raw score to interval scale transformation Psychometric properties of the WEMWBS have been for the set of items which fit model expectations, and reported for UK adult and student general population (c) examine the external construct validity of the Rasch- samples [4, 5] and the UK veterinary profession [6]. derived interval scale. However, in common with many health rating scales, it uses the Likert method of summated ratings, in which sequentially ordered response options for each item are Methods scored with sequentially ordered integers, and item scores are summed to give a total score. This leads to the potential Participants and data collection for misinterpretation because, although the total scores have a rank order, the intervals between each score are not The study was based on the data sets derived from two necessarily equal. The ordinal scores are sufficient to independent cross-sectional surveys: (a) a postal ques- separate respondents into groups based upon the magnitude tionnaire survey (mailed in October/November 2007) of a of the health dimension being measured and to indicate large stratified random sample of veterinary surgeons whether the severity has improved or worsened. However, practising in the UK, yielding a response rate of 56.1% it is not possible to make meaningful score comparisons (1,796/3,200) [16] (Survey 1), and (b) The Royal College between subgroups of respondents and it is difficult to of Veterinary Surgeons (RCVS) Survey of the Profession interpret any change in the score over time because the 2010 which was administered during January–February distance between each score on the scale does not neces- 2010 by the Institute of Employment Studies on behalf of sarily reflect equidistant steps in the severity of the the RCVS and mailed to all veterinary surgeons registered underlying construct [7]. Furthermore, the application of to practise in the UK, yielding a response rate of 37.4% parametric statistical methods is limited by their inherent (8,829/23,594) [15] (Survey 2). The WEMWBS and sev- assumption that the intervals between each score are equal. eral other psychometric instruments relating to mental 123 Qual Life Res health and well-being were embedded in the questionnaire predictions of those responses from the Rasch model of Survey 1. The WEMWBS was the only psychometric (which defines how a set of items should perform for the instrument embedded in the principally employment-rela- response data to be transformed into interval scaling). ted questionnaire of Survey 2. The demographic and Thus, the difference between expected and observed scores occupational profile of respondents was generally in close indicates the degree to which rigorous measurement is alignment with that of the original sample for both surveys achieved [22]. Item fit was evaluated iteratively: The data [15, 16]. for misfitting items were examined to try to identify potential explanations, the most misfitting items were Measures removed sequentially, and fit of the remaining items was re-evaluated until the criteria for Rasch model fit were Each of the 14 items of the WEMWBS is answered on a satisfied. The possibility of constructing a second scale 5-point Likert scale (1–5), based on the respondent’s from the deleted items was investigated by testing the fit of feelings over the previous 2 weeks, giving a minimum total these items to the Rasch model. score of 14 and a maximum total score of 70. A higher Cases with no missing WEMWBS scores were selected score indicates a higher level of mental well-being [4]. In in the Survey 2 data set (n = 7,837). Four random samples teenage student and adult samples of the UK general of n = 500 cases were drawn, without replacement, for use population, the scale fulfilled traditional psychometric in the Rasch analysis. Rasch analysis was applied to sub- criteria including Cronbach’s alphas: 0.87–0.91; 1-week sets of the full data set because there is evidence that some test–retest reliabilities (intra-class correlation coefficient): Rasch fit statistics for scales with polytomous response 0.66–0.83; and moderate to high correlations with several categories are highly dependent on sample size, which comparator scales measuring well-being, positive affect, or translates into a greater likelihood of Type I errors (falsely mental health [4, 5]. rejecting an item as not fitting the Rasch model) with The external construct validity of the Rasch-derived increased sample size [23]. Stable calibrations from Rasch interval scale was assessed in the present study by testing analysis are considered achievable with a sample size of associations with scores on the Hospital Anxiety and 250 or greater, irrespective of the targeting of the item Depression Scale (HADS) [17]; the negative work–home difficulty to the person ability of the sample [24]. In order interaction (WHI_N) subscale of the SWING instrument to obtain robust estimates, the findings of Rasch analysis on [18]; the demands, control, and managerial support sub- the first subset examined were cross-validated on each of scales of the Health and Safety Executive Management the other three subsets from the Survey 2 data set and a Standards Indicator Tool (HSE MSIT) [19]; and with the random sample of 500 cases with no missing WEMWBS response to a question on suicidal ideation over the pre- values drawn from the Survey 1 data set. The results should vious 12 months, sourced from the second National Survey be consistent across each of the five subsets of data. The of Psychiatric Morbidity, ‘Have you thought of taking your transformation from raw score to interval measurement life, even if you would not really do it? [20]’. was computed using a data set comprising all respondents with no missing item response data from Survey 1 and Rasch analysis Survey 2 combined (n = 9,754). Use of the large combined data set enhanced the precision of the estimates irrespective The Rasch model assumes that the probability of a par- of the targeting of the scale [25]. ticular respondent affirming a particular response option Fit to the Rasch model was assessed through overall for an item is a logistic function of the relative distance summary tests of fit, individual item fit, and individual between the level of the latent variable (in this study, person fit, using statistical (fit residual, chi-square) and mental well-being) expressed by the response option for graphical (item characteristic curve, ICC) indicators [7]. that item (the item difficulty) and the level of mental well- We also examined (a) evidence that the response options being of the person (the person ability) on a linear scale. for each item work as intended to reflect sequential levels The unit of measurement of the Rasch model is the logit of mental well-being (item response option threshold which is defined, for polytomous scoring (i.e. scales in ordering), (b) the extent to which the level of mental well- which the items have more than two response options, like being expressed by the item response options are matched the WEMWBS), as the log odds ratio of the probability of to the level of mental well-being of people in the sample responding with category X to the probability of respond- (targeting), (c) evidence that responses to items are not ing with category X - 1[21]. The model and analytical associated with responses to other items other than through process are described in detail elsewhere [7, 9]. In essence, the latent variable being measured (response dependency), a Rasch analysis examines the extent to which the observed and (d) evidence that responses to items are not biased by data (responses to scale items) are consistent with (i.e. ‘fit’) gender or age (differential item functioning, DIF). For the 123 Qual Life Res purpose of DIF assessment, the age of respondents was of anxiety or depressive symptoms, and an increase in HSE categorised into 4 different groups based on age quartiles MIST scores signifies more favourable psychosocial work computed from the full Survey 2 data set of respondents characteristics, that is, lower risk of work-related stress); with no missing WEMWBS values (under 33 years, (b) in multiple linear regression (the Rasch-derived interval 33–42 years, 43–54 years, and 55 years and over). scale as the as the dependent variable), suicidal ideation, A basic assumption when summing rating scale items high WHI_N (a score of [1 SD above the mean score on into a total score is that the items represent a common the WHI_N scale) and increases in scores on the HADS-A underlying construct; that is, they should be unidimen- and HADS-D subscales and the HSE MIST demands sional. Unidimensionality requires that there should be subscale would each be associated with a decrease in the only random associations between the items once the factor score on the Rasch-derived interval scale; and (c) in mul- accounting for the principal associations between the items tiple logistic regression (the Rasch-derived interval scale as (the ‘Rasch factor’) has been taken into account. To test the independent variable), an increase in score would be this formally, the items were divided into two subsets based associated with decreased odds of reporting HADS-A on whether they loaded positively or negatively on the first caseness (HADS-A score C8), HADS-D caseness (HADS- principal component of the residuals. These two subsets of D score C8), and suicidal ideation. items were used to make separate person estimates for each To assess discriminant construct validity, the extent to person which were compared by the application of a series which scores are not associated with theoretically unrelated of t tests. Fewer than 5% of tests should be significantly constructs or variables [29], it was hypothesised that the different (or the lower bound of the 95% binomial confi- interval scale score would be unrelated to country/region of dence interval should overlap 5%) for the scale to be graduation (UK, overseas EU, or overseas non-EU), and considered unidimensional [9, 26, 27]. therefore, the difference in mean scores across this variable Reliability was assessed using the person separation would not be statistically significant. This hypothesis was index (PSI), which is analogous to Cronbach’s alpha, tested by applying one-way ANOVA tests. substituting the person estimate from the Rasch model To assess group difference validity, the extent to which (logit value) for the ordinal raw score in the same formula scores can discriminate between groups which are pre- [9]. Reliability for group-level or individual use is indi- dicted to have differences in the level of the underlying cated by values[0.7 or[0.85, respectively [9]. Bonferroni construct [7], it was hypothesised, based on the findings of adjustment was applied to the 0.05 alpha level for both fit a recent UK psychiatric morbidity study [30] and the and DIF statistics due to the multiple testing. Rasch anal- results of the WEMWBS in a UK general population ysis was performed with a partial credit model using sample [4], that the mean score for men would be higher RUMM2030 software [28]. than for women. The significance of the difference was evaluated using independent t tests after checking for External construct validity homogeneity of variance (Levene’s test). It was considered that the results would demonstrate To assess convergent construct validity, the extent to which evidence of construct validity if C75% of these hypotheses scores are associated with other measures of the same or were confirmed [11]. Statistical significance was defined as theoretically similar constructs [29], Pearson correlation P \ 0.001 to account for multiple testing and the large coefficients were computed to explore possible pairwise sample size. Statistical analysis was performed using correlations between scores on the Rasch-derived interval IBMÒ SPSSÒ Statistics version 17.0 (SPSS Inc.). scale and scores on HADS, WHI_N, and the demands, control and managerial support subscales of HSE MSIT, using the data set from Survey 1. In addition, using the Results same data set, multiple regression was applied to estimate any associations between scores on the Rasch-derived Participant characteristics interval scale, the above measures and suicidal ideation after adjustment for age and gender and each of the other These have been reported in full elsewhere [15, 16]. In scales. We hypothesised that (a) the Rasch-derived interval summary, the demographic and occupational characteris- scale would show moderate negative correlations (r = tics of participants, for Surveys 1 and 2, respectively, were -0.3 to -0.7) with HADS-A and HADS-D subscales, (a) proportion male, 50, 50%; (b) mean (SD) age, 40.8 which capture anxiety and depressive symptoms, respec- (11.0) years, 45.5 (15.9) years; (c) graduated in the UK, 83, tively, and moderate positive correlations with the HSE 79%; (d) working in practice, 84, 81%; and (e) full-time MIST subscales (the directions of correlation are different employment, 82, 64%. The sampling frame for Survey 1 because an increase in HADS scores signifies higher levels included almost exclusively veterinary surgeons who were 123 Qual Life Res

Fig. 1 Response option thresholds ordering for item 9. Category time’, 3 ‘often’, and 4 ‘all of the time’. It can be seen that the probability curves for item 9 ‘I’ve been feeling close to other people’ responses to this item fall in a logical progressive order—category showing ordered response category thresholds. Each curve represents responses represent an increase in mental well-being (logits) for each a response category. The x-axis symbolises the construct, with the category. For example, at the lowest level of mental well-being (-6 level of mental well-being increasing to the right. The y-axis shows logits), the most probable score is 0 ‘none of the time’, and at the the probability of endorsing the response categories at different levels highest level of mental well-being (?5 logits), the most probable of mental well-being: 0 ‘none of the time’, 1 ‘rarely’, 2 ‘some of the score is 4 ‘all of the time’

Fig. 2 Threshold map for the 14-item WEMWBS. Items are listed in order of increasing item number (item 1 to item 14) on the y-axis. Person location (logits) is on the x-axis

currently in or seeking employment. By contrast, ques- response is equally probable) for each item was assessed by tionnaires in Survey 2 were sent to all registered veterinary comparing the logit ability rating for each response option surgeons, including those in retirement. This accounts for using each item’s threshold category probability curve. All the higher mean age (mean difference = 4.7 years, 95% item response option thresholds were found to be ordered; CI = 3.9–5.4 years, P \ 0.001) of respondents to Survey 2 that is, within each item, the transition from one category and the lower proportion in full-time employment (differ- to the next represents an increase in the underlying trait of ence =-18%, 95% CI =-20 to -16%, P \ 0.001). mental well-being (Figs. 1, 2).

Rasch analysis Individual item ﬁt Response option thresholds ordering Three indicators were examined: (a) ﬁt residuals (item– The ordering of response option thresholds (i.e. the tran- person interaction), (b) chi-square probability values sition between adjacent response options where either (item–trait interaction), and (c) ICCs. 123 Qual Life Res

Seven of the 14 items had fit residuals outside the rec- Fig. 3 Item characteristic curves for: a items 8, b item 12, and c item 9. c ommended range of ±2.5 (if the data fit the Rasch model, The ICC is a graph for an individual item. It plots the expected response (predicted from the model) to an item at each level of the measurement the mean value is expected to be approximately zero). High continuum. The available responses for WEMWBS items are: 0 ‘none positive fit residual values indicate misfit to model expec- of the time’, 1 ‘rarely’, 2 ‘some of the time’, 3 ‘often’, and 4 ‘all of the tations, and high negative fit residuals indicate item time’. a The fit residual for item 8 is negative, indicating an over- redundancy. Six items with the highest fit residuals were discriminating item relative to the model. The dots represent the observed scores for groups of people with similar total scores (class removed from the scale: item 5, ‘I’ve had energy to spare’ intervals) at the different ability levels. The chi-square probability (?5.81); item 8, ‘I’ve been feeling good about myself’ (P \ 0.001, displayed in the line of text above the graph) indicates a (-10.59); item 10, ‘I’ve been feeling confident’ (-6.22); statistically significant difference between expected and observed item 12, ‘I’ve been feeling loved’ (?7.94); item 13 ‘I’ve scores for this item. The class interval dots tend to be under the line at the left-hand end and above the line at the right-hand end (i.e. the been interested in new things’ (?2.99); and item 14, ‘I’ve observed scores form a steeper curve than the expected scores). been feeling cheerful’ (-9.61). Therefore, among people with higher levels of mental well-being, more The ICCs for the worst-fitting (items 8 and 12) and best- than predicted report feeling good about themselves ‘often’ or ‘all of the fitting (item 9, ‘I’ve been feeling close to other people’ time’, and among people with lower levels of mental well-being, fewer than predicted report feeling good about themselves ‘none of the time’ [-0.68]) items are illustrated in Fig. 3. or ‘rarely’. b By contrast, the fit residual for item 12 is positive, indicating an under-discriminating item relative to the model. [c] The fit Individual person fit residual for item 9 is almost zero, indicating good fit to the model. The chi-square probability (P = 0.209) indicates that the difference between expected and observed scores for this item is not statistically Examination of individual person-fit statistics showed that significant 14.4% (284/1,971) of participants had fit residuals outside the acceptable range of ±2.5. The range of individual person-fit residuals was -5.88 to ?4.95. A high negative locally dependent, which can affect parameter estimates person-fit residual is indicative of fitting the model ‘too and inflate reliability [7]. Response dependency was well’. This is generally regarded as being less problematic identified between items 6 and 7 only (residual correlation to the analyses than a high positive person-fit residual [31]. ?0.312) for the WEMWBS. There was no dependency Examination of the response patterns of individuals with between any item pairs after the removal of items 4, 5, 8, high positive fit residuals identified several who had 10, 11, 12, and 13 (the residual correlation between items 6 affirmed response categories which indicated opposite and 7 was reduced to ?0.238). extremes of mental well-being. However, no respondents were removed from the analyses. Fit of item sets to the Rasch model

DIF Initial fit of the WEMWBS to model expectations was poor (Analysis 1, Table 1). Item 8, item 10, and item 12 all Item 8 and item 4 (‘I’ve been feeling interested in other showed significant individual item misfit and were deleted. people’) showed uniform DIF for gender, and items 1, 2, 3, This led to a marginal improvement in fit (Analysis 2). 10, 11, and 12 showed uniform DIF for gender and age Item 14 was also deleted due to significant individual item group. A degree of DIF cancellation can occur at the misfit. This led to a further marginal improvement in fit summed scale level when certain items in the scale have (Analysis 3). Item 4 ‘I’ve been feeling interested in other DIF of similar magnitude and opposite direction for a people’ was deleted due to significant DIF for gender, and specific variable, thereby reducing the overall impact [32]. Item 5 ‘I’ve had energy to spare’ and item 13 ‘I’ve been For example, when items 1 and 3 were grouped into a interested in new things’ were deleted due to individual subtest, the DIF for both gender and age group cancelled. item misfit. A series of t tests performed on the person Cancellation was retained when item 11 was added to the estimates derived from the two most different subsets of subtest. Consequently items 1, 3, and 11 were retained in items identified from the factor loadings on the first prin- the scale. Item 4 was deleted from the scale (females affirm cipal component of the residuals indicated that only 5.6% a higher response category than males at the same level of (95% CI = 4.5–6.5%) of cases had statistically significant mental well-being). P values, which supported the scale’s unidimensionality (Analysis 4). This Rasch-derived 7-item scale, comprising Response dependency items 1, 2, 3, 6, 7, 9, and 11 only, is identical to the Rasch model-fitting SWEMWBS reported by Stewart-Brown Response dependency was identified through inspection of et al. [33]. As might be expected with a shorter scale, the residual correlation matrices. Pairs of items with positive PSI, which provides an indication of the ability of the residual correlations exceeding ?0.3 are considered to be measure to discriminate among respondents with different 123 Qual Life Res

levels of the latent variable being measured, had fallen the robustness of the solution (Analysis 4, Table 1) was from 0.918 (Analysis 1) to 0.832 (Analysis 4) but remained tested on the four other data subsets for the purpose of acceptable. Rasch fit statistics for the seven items are cross-validation (Analyses 5–8, respectively). These data presented in Table 2. subsets also showed good fit to model expectations. Fit Given the 7-item scale’s satisfactory fit to the Rasch of the deleted items to the Rasch model was poor model and the evidence supporting its unidimensionality, (Analysis 9).

123 Qual Life Res

Table 1 Summary of model fit statistics for original scale, successive revisions, and cross-validation Sample/notes/action Analysis Item-fit residuals Person-fit residuals Overall model fita PSI Unidimensionality number t tests %b Mean SD Mean SD Chi-square df P value (95% CI)

S2_500a 1 -0.574 5.341 -0.540 1.644 558.2 112 <0.001 0.918 11.7 Full 14-item scale (10.7–12.6) S2_500a 2 -0.551 3.436 -0.532 1.504 241.6 88 <0.001 0.892 9.2 Delete items: 8, 10, 12 (8.3–10.3) S2_500a 3 -0.335 1.795 -0.507 1.433 161.9 80 <0.001 0.876 8.6 Delete item: 14 (7.7–9.6) S2_500a 4 -0.570 2.326 -0.520 1.295 58.8 56 0.104 0.832 5.6 Delete items: 4, 5, 13 (4.5–6.5) S2_500b 5 -0.065 1.923 -0.476 1.222 57.4 56 0.068 0.833 4.6 Analysis 4 repeated (3.4–5.5) S2_500c 6 0.112 1.473 -0.389 1.933 56.1 56 0.041 0.894 5.9 Analysis 4 repeated (4.9–7.0) S2_500d 7 0.511 2.852 -0.287 1.342 62.8 56 0.119 0.845 4.7 Analysis 4 repeated (3.7–5.6) S1_500 8 -0.054 2.777 -0.516 1.459 71.9 56 0.235 0.832 5.4 Analysis 4 repeated (4.3–6.4) S2_500a 9 -0.344 5.913 -0.497 1.375 269.3 56 <0.001 0.854 8.3 Deleted items only (7.3–9.2) Bold signifies misfit to the Rasch model. Bonferroni adjustment not applied, statistical significance defined as P \ 0.05 (it is more conservative to not apply Bonferroni adjustment for summary of model fit statistics, thereby reducing the chance of Type I error, that is, incorrectly declaring fit to the Rash model) SD standard deviation, PSI person separation index, df degrees of freedom a Overall item–trait interaction b % of statistically significant t tests

Table 2 Rasch ﬁt statistics for the 7 retained items Item Location SE Fit residual Chi-square P value

1: I’ve been feeling optimistic about the future 0.48 0.03 0.27 7.00 0.536444 2: I’ve been feeling useful 0.00 0.04 -0.79 10.86 0.209974 3: I’ve been feeling relaxed 1.06 0.03 1.74 14.54 0.068788 6: I’ve been dealing with problems well 0.08 0.04 2.34 19.31 0.009984 7: I’ve been thinking clearly -0.66 0.04 2.15 18.78 0.046289 9: I’ve been feeling close to other people 0.11 0.03 2.47 11.48 0.175927 11: I’ve been able to make up my own mind about things -1.07 0.04 -0.70 20.28 0.009337 Reliability (PSI) = 0.832; Bonferroni adjustment to the alpha level for the number of items (P = 0.05/7 = 0.007143) SE standard error

Targeting of the derived 7-item scale across the range of person estimates (-6.109 to ?6.376) for the sample, with adequate coverage across the mea- Since the level of mental well-being in respondents and surement continuum. There were no item thresholds at the item responses can be placed on the same logit scale, it is extremes of the highest and lowest levels of mental well- possible to test how well the ﬁnal set of items and their being (i.e. items do not discriminate at these levels). response options are targeted, that is, how well the level of However, ﬂoor or ceiling effects were minimal. There was mental well-being expressed by the items and their minimal stacking of item thresholds, which provided evi- response options is matched to the level of mental well- dence to support that all the items contributed to the being in respondents (Fig. 4). There was an acceptable explanatory power of the scale (low item redundancy). The breadth of spread of item thresholds (-4.614 to ?4.700) mean person score of ?1.15 logits (SD = 1.56) indicated

123 Qual Life Res

Fig. 4 Person–item threshold distribution for the derived 7-item scale. The ﬁgure shows the targeting between the distribution of the respondent level of mental well-being (top histogram) and the distribution of item threshold locations (inverted histogram)

Table 3 Raw score to interval Raw Logit measure Transformed logit Raw Logit measure Transformed logit measure conversion table for the score (interval) measure (interval) score (interval) measure (interval) Rasch-derived 7-item scale 7 -6.11 7.00 22 0.11 20.94 8 -5.12 9.22 23 0.44 21.68 9 -4.38 10.87 24 0.78 22.45 10 -3.83 12.10 25 1.14 23.26 11 -3.39 13.11 26 1.52 24.12 12 -3.00 13.96 27 1.93 25.04 13 -2.66 14.73 28 2.37 26.01 14 -2.35 15.44 29 2.82 27.02 15 -2.04 16.12 30 3.27 28.04 16 -1.74 16.79 31 3.73 29.07 17 -1.45 17.45 32 4.22 30.16 18 -1.15 18.12 33 4.77 31.39 19 -0.84 18.81 34 5.47 32.97 20 -0.53 19.50 35 6.38 35.00 21 -0.22 20.21 that, on average, the mean level of mental well-being of the The mean interval measure SWEMWBS score for the sample was slightly higher than the mean level of metal sample of veterinary surgeons in Survey 1 (n = 1,743) and well-being represented by the item set (0 logits, Survey 2 (n = 7,990) was 23.64 (SD = 3.78) and 23.27 SD = 0.70) (by convention, the mean item location is (SD = 3.44), respectively (mean difference = 0.37, 95% centred on zero logits in the Rasch model). CI = 0.18–0.57, P \ 0.001).

External construct validity Transformation from raw score to interval measurement Tests supported the external construct validity of the Ras- The summed raw scores were mapped against the corre- ch-derived 7-item interval scale (Table 4). sponding estimated Rasch-transformed logit scores for the set of ﬁtting items. To calibrate an interval-level scale, logit measures for the ﬁtting items were rescaled into the Discussion score range of the ordinal scale (7–35) (Table 3). The change in measurement implied by a one unit change in The aims of the study were to examine the psychometric raw score varied 3.4-fold across the range of the scale, properties of the 14-item WEMWBS in the UK veterinary from 0.29 logits (raw score, 16–17) to 0.99 logits (raw profession by the application of Rasch analysis, and to score, 1–2). assess the external construct validity of the derived interval

123 Qual Life Res

Table 4 Tests of external construct validity for the Rasch-derived 7-item interval scale Statistical method/variable

Convergent construct validity (n = 1,390–1,499) Pearson correlation coefficient r, (95% CI), P value HADS-A -0.66 (-0.69 to -0.63), P \ 0.001 HADS-D -0.69 (-0.71 to -0.67), P \ 0.001 HSE MSIT demands subscale 0.30 (0.26 to 0.34), P \ 0.001 HSE MSIT control subscale 0.43 (0.39 to 0.48), P \ 0.001 HSE MSIT managerial support subscale 0.46 (0.42 to 0.50), P \ 0.001 Linear regressiona Coef. (95% CI), P value HADS-A -0.29 (-0.33 to -0.24), P \ 0.001 HADS-D -0.43 (-0.48 to -0.38), P \ 0.001 HSE MSIT demands subscale 20.14 (20.36 to 0.08), P = 0.218 HSE MSIT control subscale 0.49 (0.27 to 0.71), P \ 0.001 HSE MSIT managerial support subscale 0.31 (0.08 to 0.54), P = 0.010 Suicidal thoughts in previous 12 months 20.40 (20.77 to 20.01), P = 0.042 WHI_N 0.33 (20.14 to 0.78), P = 0.164 Logistic regressionb OR, (95% CI), P value HADS-A possible or probable caseness (score C 8) 0.68 (0.64 to 0.71), P \ 0.001 HADS-A probable caseness (score C11) 0.66 (0.62 to 0.70), P \ 0.001 HADS-D possible or probable caseness (score C8) 0.56 (0.52 to 0.61), P \ 0.001 HADS-D probable caseness (score C11) 0.48 (0.41 to 0.56), P \ 0.001 Suicidal thoughts in previous 12 months 0.87 (0.81 to 0.92), P \ 0.001 Discriminant construct validity One-way ANOVA Country/region of graduation F(2, 1,786) = 0.3, P = 0.751 Group difference validity Independent samples t test Gender group differences in mean score Male (mean) (95% CI) n = 875, 23.10 (22.84 to 23.36) Female (mean) (95% CI) n = 884, 22.27 (22.02 to 22.52) Mean differencec (95% CI), P value -0.84 (-1.19 to -0.48), P \ 0.001 Bold signifies values that are inconsistent with the a priori hypotheses HADS-A Anxiety subscale of Hospital Anxiety and Depression Scale, HADS-D Depression subscale of Hospital Anxiety and Depression Scale, HSE MSIT Health and Safety Executive Management Standards Information Tool, WHI_N Negative work–home interaction a WEMWBS as the dependent variable; b WEMWBS as the independent variable; c Independent samples t tests, equal variances assumed (Levene’s test P [ 0.05) scale measurements. Our results provide evidence to sup- contrast, Rasch measurement theory (RMT) provides a port the interval scale measurement properties of the mathematical model that embodies the principle of SWEMWBS. The external construct validity of the scale invariant comparison—the relative location of (i.e. distance was also supported. between) two people on the continuum does not depend on The choice of an appropriate measurement model to the items to which they respond, and the relative location examine rating scale data is the subject of considerable of two items does not depend on the people from whom the debate [7, 34–38]. The different paradigms are outlined estimates are made—and guides the construction of stable here to justify our approach. Item response theory (IRT) linear measures [8]. Therefore, the aim of a Rasch analysis models are statistical models used to explain data, and as is to determine the extent to which observed rating scale such, the criterion for the choice of a model is determined data satisfy the measurement model. When the data do not by its fit to the observed rating scale data [34, 36, 37]. In fit the model, the data are examined carefully to try and general, when the data do not fit the chosen IRT model, explain the misfit, but ultimately data are chosen that sat- another model is sought to better explain the data. By isfy the model’s requirements [39]. This is what

123 Qual Life Res distinguishes the RMT diagnostic paradigm from the IRT response category than males at the same level of mental statistical modelling paradigm [35]. The goal of our study well-being. However, this item did not display DIF for gender was to examine the WEMWBS as an explicit hypothesis of in the UK general population sample [33]. A possible how mental well-being can be measured. Therefore, we explanation for the difference is that women in the present elected to exploit the RMT diagnostic approach to provide study are members of an occupational group in which the an extensive test of this hypothesis in its original format proportion of women is increasing and prospects for suitable (i.e. electing not to introduce additional post hoc statisti- future employment are perceived more favourably. cally led item parameterisation). The raw score to interval measure transformations are Initial estimation of model fit suggested that WEMWBS inevitably not identical because they are derived from data showed deviation from Rasch model expectations and, different data sets. It is suggested that the transformation therefore, cannot be assumed to be compatible with inter- derived in the present study is used for future analyses in val-level scaling when used in this population. On the basis the context of the veterinary profession. Although future of item-fit statistics and examination for DIF, the most analyses in the context of the general population are likely misfitting items were removed sequentially to derive a to use the transformation published by Stewart-Brown et al. 7-item scale which satisfied the criteria for Rasch model fit. [33], the differences are minimal and unlikely to affect the Removing items from a scale should be carefully con- interpretation of comparisons between the two population sidered because content validity may be compromised [40], samples. However, this assumption should be tested. and evidence for the external validity of an existing pub- The SWEMWBS scale offers superior measurement lished scale is likely to have included all items, so the properties compared to the 14-item WEMWBS from which revised scale would need to be re-validated [25]. Moreover, it is derived. These include (a) interval-level measurement, protecting the integrity of the item set would facilitate which means that mental well-being can be quantified with comparisons across studies in which the full scale was greater precision and assessments of within-group changes administered. At the outset of the Rasch analysis, there was and between-group differences are strengthened by the no explicit intention to derive a new measure for the sake compatibility of the interval measurements with parametric of shortening the instrument to minimise respondent bur- statistics applied to test cross-sectional or longitudinal den. However, brevity of assessments is often valued by differences in scores; (b) unidimensionality, which means respondents and provided adequate reliability is preserved, that the items define a single underlying construct, that is, removing items to improve measurement properties is mental well-being; and (c) measurement invariance, which considered to be a desirable goal [41]. means that the instrument functions in the same way across The Rasch-derived 7-item scale had minimal item bias subgroups of the sample population of different gender, age (DIF), minimal response dependency, and its polytomous and level of mental well-being. response structure worked as intended, with higher scores Evidence to support the external construct validity of the within an item reflecting greater overall mental well-being. SWEMWBS has not previously been reported for any pop- The person–item threshold distribution indicated that tar- ulation sample. The associations between the interval mea- geting of the Rasch-derived 7-item scale was acceptable. surements of the SWEMWBS and comparator measures of The PSI was acceptable across each of the samples tested psychological health and perceptions of psychosocial work (range, 0.832–0.894), exceeding the minimum value of 0.7 characteristics were similar to those reported for the 14-item for group-level measurement [42]. This supports the ability WEMWBS [6]. The associations of SWEMWBS interval of the measure to discriminate among groups of respon- measurements with HADS-A and HADS-D caseness and dents with different levels of mental well-being. with suicidal ideation are particularly relevant in the context Four of the seven deleted items showed DIF for gender. of a profession with an elevated risk of suicide and support This can be a cause of multidimensionality, which may their use as overall indicators of mental health and well- explain why it was not possible to construct a second being in this occupational group. meaningful scale from these items. A strength of the study is the use of two large, inde- It is interesting that the 7-item scale derived in the present pendently derived data sets, which meant that the results of study is identical to the SWEMWBS, derived from the the Rasch analysis could be validated both within and application of Rasch analysis to WEMWBS data from a UK across samples. It is possible, however, that some partici- general population sample [33]. This will enable valid com- pants may have responded to both surveys, albeit over parison of results between the two populations. Although the 2 years apart. The consistency of the results suggests that same items were deleted in both studies, the reasons for they should be generalisable across the wider population of deletion were sometimes different. For example, item 1 (‘I’ve veterinary surgeons in the UK. A limitation of the evalu- been feeling optimistic about the future’) displayed DIF for ation of the convergent construct validity of the SWE- gender in the present study, with females affirming a higher MWBS interval measures is that evidence to support a 123 Qual Life Res range of psychometric properties for the comparator scales 5. Clarke, A., Friede, T., Putz, R., Ashdown, J., Martin, S., Blake, A., is restricted to samples of other populations. Moreover, the et al. (2011). Warwick-Edinburgh Mental Well-being Scale (WE- MWBS): validated for teenage school students in England and comparator scales were ordinal and were not transformed Scotland.A mixed methodsassessment. BMC Public Health, 11,487. to interval scales before their associations with SWE- 6. Bartram, D. J., Yadegarfar, G., Sinclair, J. M. A., & Baldwin, D. MWBS were tested using parametric methods. The find- S. (2011). Validation of the Warwick-Edinburgh Mental Well- ings should be interpreted with this in mind. being Scale (WEMWBS) as an overall indicator of population mental health and well-being in the UK veterinary profession. The Veterinary Journal, 187, 397–398. 7. Hobart, J., & Cano, S. (2009). Improving the evaluation of Conclusions therapeutic interventions in multiple sclerosis: The role of new psychometric methods. Health Technology Assessment, 13(12), iii, ix–x, 1–177. This study extends evidence that the summed total score 8. Rasch, G. (1960). Probabilistic models for some intelligence and of the WEMWBS does not meet the strict criteria for attainment tests. Copenhagen: Danish Institute for Educational measurement demanded by the Rasch model, with some Research. items demonstrating DIF and over- or under-discrimina- 9. Tennant, A., & Conaghan, P. G. (2007). The Rasch measurement model in rheumatology: What is it and why use it? When should tion. The Rasch-derived 7-item scale (SWEMWBS) satis- it be applied, and what should one look for in a Rasch paper. fied all criteria, including unidimensionality. A linear Arthritis and Rheumatism, 57, 1358–1362. transformation of the raw score from SWEMWBS can be 10. Meredith, W., & Teresi, J. A. (2006). An essay on measurement used with confidence in parametric analyses for the UK and factorial invariance. Medical Care, 44, S69–S77. 11. Terwee, C. B., Bot, S. D. M., de Boer, M. R., van der Windt, D. veterinary profession, given appropriate distribution. The A. W. M., Knol, D. L., Dekker, J., et al. (2007). Quality criteria convergent construct validity of the interval scale is sup- were proposed for measurement properties of health status ported by evidence of its expected associations with other questionnaires. Journal of Clinical Epidemiology, 60, 34–42. measures of psychological health or psychosocial work 12. Gregorich, S. E. (2006). Do self-report instruments allow meaningful comparisons across diverse population groups? Testing characteristics. To our knowledge, this is the first report of measurement invariance using the confirmatory factor analysis the convergent construct validity of the SWEMWBS framework. Medical Care, 44, S78–S94. interval scale. The results suggest its suitability as an 13. Leiter, M. P., & Schaufeli, W. B. (1996). Consistency of the indicator of population mental health and well-being in the burnout structure across occupations. Anxiety, Stress and Coping, 9, 229–243. UK veterinary profession. However, the properties of test– 14. Vandenberg, R. J., & Lance, C. E. (2000). A review and synthesis retest reliability and responsiveness should be evaluated of the measurement invariance literature: Suggestions, practices, further within the context of the veterinary profession and recommendations for organizational research. Organizational before the scale can be advocated as being appropriate for Research Methods, 3, 4–70. 15. Robertson-Smith, G., Robinson, D., Hicks, B., Khambhaita, P., & the purpose of monitoring the population-level changes in Hayday, S. (2010). The 2010 RCVS survey of the UK veterinary and mental health and well-being or as an outcome measure for veterinary nursing professions. Brighton: Institute for Employment intervention studies. Studies. http://www.rcvs.org.uk/publications/rcvs-survey-of-the- professions-2010/surveyprofessions2010.pdf. Accessed February Acknowledgments This research would not have been possible 13, 2012. without the support of the veterinarians who invested their time into 16. Bartram, D. J., Yadegarfar, G., & Baldwin, D. S. (2009). A cross- completing and returning the questionnaires. The RCVS included the sectional study of mental heath and well-being and their associ- WEMWBS in the RCVS Survey of the Profession 2010 and supplied ations in the UK veterinary profession. Social Psychiatry and the response data in a suitable format for analysis. Psychiatric Epidemiology, 44, 1075–1085. 17. Zigmond, A. S., & Snaith, R. P. (1983). The Hospital Anxiety and Depression Scale. Acta Psychiatrica Scandinavica, 67, 361–370. 18. Geurts, S. A. E., Taris, T. W., Kompier, M. A. J., Dikkers, J. S. E., van Hooff, M. L. M., & Kinnunen, U. M. (2005). Work- References home interaction from a work psychological perspective: Development and validation of a new questionnaire, the SWING. 1. Bartram, D. J., & Baldwin, D. S. (2010). Veterinary surgeons and Work & Stress, 19, 319–339. suicide: A structured review of possible influences on increased 19. Cousins, R., Mackay, C. J., Clarke, S. D., Kelly, C., Kelly, P. J., risk. Veterinary Record, 166, 388–397. & McCaig, R. H. (2004). ‘Management standards’ and work 2. Platt, B., Hawton, K., Simkin, S., & Mellanby, R. J. (2010). related stress in the UK: Practical development. Work & Stress, Systematic review of the prevalence of suicide in veterinary 18, 113–136. surgeons. Occupational Medicine, 60, 436–446. 20. Singleton, N., Bumpstead, R., O’brien, M., Lee, A., & Meltzer, 3. Charlton, J. (1995). Trends and patterns in suicide in England and H. (2001). Psychiatric morbidity among adults living in private Wales. International Journal of Epidemiology, 24(Suppl. 1), households, 2000. London: The Stationery Office. S45–S52. 21. Lamoureux, E. L., Pesudovs, K., Thumboo, J., Saw, S.-M., & 4. Tennant, R., Hiller, L., Fishwick, R., Platt, S., Joseph, S., Weich, Wong, T. Y. (2009). An evaluation of the reliability and validity S., et al. (2007). The Warwick-Edinburgh Mental Well-being of the visual functioning questionnaire (VF-11) using Rasch Scale (WEMWBS): development and UK validation. Health and analysis in an Asian population. Investigative Ophthalmology & Quality of Life Outcomes, 5, 63. Visual Science, 50, 2607–2613.

123 Qual Life Res

22. Cano, S. J., Barrett, L. E., Zajicek, J. P., & Hobart, J. C. (2011). 33. Stewart-Brown, S., Tennant, A., Tennant, R., Platt, S., & Par- Beyond the reach of traditional analyses: Using Rasch to evaluate kinson, J. (2009). Internal construct validity of the Warwick- the DASH in people with multiple sclerosis. Multiple Sclerosis Edinburgh Mental Well-being Scale (WEMWBS): A Rasch Journal, 17, 214–222. analysis using data from the Scottish Health Education Popula- 23. Smith, A. B., Rush, R., Fallowfield, L. J., Velikova, G., & Sharpe, tion Survey. Health and Quality of Life Outcomes, 7, 15. M. (2008). Rasch fit statistics and sample size considerations for 34. Edelen, M. O., & Reeve, B. B. (2007). Applying item response polytomous data. BMC Medical Research Methodology, 8, 33. theory (IRT) modeling to questionnaire development, evaluation, 24. Linacre, J. M. (1994). Sample size and item calibration stability. and refinement. Quality of Life Research, 16, 5–18. Rasch Measurement Transactions, 7, 328. 35. Andrich, D. (2011). Rating scales and measurement. Expert 25. Lundgren-Nilsson, A˚ ., & Tennant, A. (2011). Past and present Review of Pharmacoeconomics and Outcomes Research, 11, issues in Rasch analysis: The Functional Independence Measure 571–585. TM (FIM ) revisited. Journal of Rehabilitation Medicine, 43, 36. Massof, R. W. (2011). Understanding Rasch and item response 884–891. theory models: Applications to the estimation and validation of 26. Pallant, J. F., Miller, R. L., & Tennant, A. (2006). Evaluation of interval latent trait measures from responses to rating scale the Edinburgh Post Natal Depression Scale using Rasch analysis. questionnaires. Ophthalmic Epidemiology, 18, 1–19. BMC Psychiatry, 6, 28. 37. Hambleton, R. K., & Swaminathan, H. (1985). Item response 27. Tennant, A., & Pallant, J. F. (2006). Unidimensionality matters! theory: Principles and applications. Boston, Massachussets: (A tale of two Smiths?). Rasch Measurement Transactions, 20, Kluwer-Nijhoff. 1048–1051. 38. Wright, B. D. (1997). A history of social science and measure- 28. Andrich, D., Sheridan, B. S., & Luo, G. (2010). RUMM2030: A ment. Educational Measurement: Issues and Practice, 16, 33–45. Windows program for the analysis of data according to Rasch 39. Cano, S. J., & Hobart, J. C. (2011). The problem with health unidimensional models for measurement. Perth, Australia: measurement. Patient Preference and Adherence, 5, 279–290. RUMM Laboratory. 40. Bohlig, M., Fisher, W. P., Masters, G. N., & Bond, T. (1998). 29. Campbell, D. T., & Fiske, D. W. (1959). Convergent and dis- Content validity and misfitting items. Rasch Measurement criminant validation by the multitrait-multimethod matrix. Psy- Transactions, 12, 607. chological Bulletin, 56, 81–105. 41. Heinemann, A. W., & Deutsch, A. (2011). Commentary on ‘Past 30. Mcmanus, S., Meltzer, H., Brugha, T. S., Bebbington, P. E., & and present issues in Rasch analysis: The FIM revisited’. Journal Jenkins, R. (2009). Adult psychiatric morbidity in England, 2007: of Rehabilitation Medicine, 43, 958–960. Results of a household survey. London: National Centre for 42. Reeve, B. B., Hays, R. D., Bjorner, J. B., Cook, K. F., Crane, P. Social Research. http://www.ic.nhs.uk/cmsincludes/_process_ K., Teresi, J. A., et al. (2007). Psychometric evaluation and document.asp?sPublicationID=1231750469828&sDocID=5446. calibration of health-related quality of life item banks: Plans for Accessed February 13, 2012. the Patient-Reported Outcomes Measurement Information Sys- 31. Muller, S., & Roddy, E. (2009). A Rasch analysis of the Man- tem (PROMIS). Medical Care, 45(Suppl 1), S22–S31. chester foot pain and disability index. Journal of Foot and Ankle Research, 2, 29. 32. Teresi, J. A., & Fleishman, J. A. (2007). Differential item functioning and health assessment. Quality of Life Research, 16, 33–42.

123 Abstract Supplement to Volume 22 No 5 July 2008

Abstract Supplement to Volume 22 No 5 July 2008 Abstract Supplement to Volume 22 No 5 July 2008 Abstracts A1

SUPPLEMENT TO JOURNAL OF PSYCHOPHARMACOLOGY VOLUME 22 NUMBER 5 2008

These papers were presented at the Summer Meeting of the

BRITISH ASSOCIATION FOR PSYCHOPHARMACOLOGY

20 – 23 July, Harrogate, UK

Indemnity:

The scientific material presented at this meeting reflects the opinions of the contributing authors and speakers. The British Association for Psychopharmacology accepts no responsibility for the contents of the verbal or any published proceedings of this meeting.

BAP OFFICE 36 Cambridge Place, Hills Road Cambridge CB2 1NS

Website: bap.org.uk A30 Abstracts

MD08

REPORTED ALCOHOL CONSUMPTION, DEPRESSIVE AND ANXIETY SYMPTOMS, AND MENTAL WELL-BEING AMONG UK VETERINARY SURGEONS: CROSS-SECTIONAL QUESTIONNAIRE SURVEY Bartram DJ, Yadegarfar G, Baldwin DS. Clinical Neuroscience Division, School of Medicine, University of Southampton, RSH Hospital, Southampton, SO14 0YG, [email protected]

Veterinary surgeons are at high risk of suicide, with a proportional mortality ratio around four times that of the general population and approximately twice that of other healthcare professions. Although there has been much speculation regarding mechanisms of increased suicide risk in the profession, there is scant empirical research, and the contribution of alcohol misuse, anxiety and depression is uncertain. We wished to examine relationships between alcohol consumption, depressive and anxiety symptoms, and overall mental well-being through a questionnaire survey of a large stratified sample of veterinary surgeons practising within the UK. The questionnaire was mailed to 3,200 veterinary surgeons (approximately 20% of the membership of the Royal College of Veterinary Surgeons). Reported alcohol consumption was graded through completion of the Alcohol Use Disorders Identification Test alcohol consumption Questions (AUDIT-C), individuals being classified into non-drinkers, low-risk drinkers (score 1-3 for women and 1-4 for men) and at-risk drinkers (score 4 for women and 5 for men). Depressive and anxiety symptoms were assessed through completion of the Hospital Anxiety and Depression Scale (HADS) and mental well-being through the Warwick Edinburgh Mental Well-being Scale (WEMWBS). 1796 participants returned the completed questionnaire, a response rate of 56.1%: after imputation of minor omissions, valid data was available for 1757 individuals (881 men, 876 women). There were 95 non-drinkers (5.4%), 563 low-risk (32.0%) and 1099 (62.5%) at-risk drinkers. Mean scores on the HADS anxiety (HADS-A) and depression (HADS-D) sub-scales in the overall sample were 7.92 and 4.65, respectively. Mean HADS sub-scale scores and WEMWBS did not differ significantly (One-way ANOVA and Kruskal-Wallis tests) between non-drinkers, low-risk and at-risk drinkers: HADS-A, 7.96 vs 7.69 vs 8.04; HADS-D, 5.02 vs 4.67 vs 4.61; WEMWBS, 47.27 vs 49.42 vs 48.79. In this population, self-reported depressive and anxiety symptom severity did not differ significantly across three levels of reported alcohol consumption. The high prevalence of at-risk alcohol consumption and the HADS-A score indicating anxiety symptoms of possible clinical significance require further exploration and, if substantiated, would cause some concern if this sample is representative of the UK veterinary profession. Sources of funding: Veterinary Business Development printed and mailed the questionnaires; BUPA Giving provided financial support for the project.

MD09

ETHANOL AND PERFORMANCE IN THE LABORATORY AND EVERYDAY LIFE Tiplady B, Oshinowo B, Thomson J, Drummond G. Anaesthetics, University of Edinburgh, Edinburgh, EH16 4SA, [email protected]

Most research on the effects of ethanol on performance is carried out in the laboratory, with epidemiological studies confirming the significance of performance impairment in domains such as driving. There is increasing interest in assessment in an everyday life setting, using methods such as the internet, handheld PCs and mobile phones. Using the same volunteers, we compared assessments of the effects of ethanol on performance, in a controlled laboratory study and in normal life using mobile phones. 38 healthy volunteers (20 male) aged 18-54 years (mean 22.8) took part. They were asked not to alter their drinking habits. Text (SMS) messages were sent twice a day to the mobile phones at different times over 14 days. The application collected information on ethanol consumed, visual analogue ratings, and administered tests of memory, attention, and reaction time. 26 of the volunteers took part in the lab study. They received ethanol and placebo on separate days in random order and completed the same assessments at intervals up to 2h after the drink. Thirty volunteers reported consuming at least five units of ethanol in the previous 6h at least once during the two week period. Performance was compared to similar times with no ethanol in the past 24h in the same volunteers. Mean blood alcohol concentrations in the lab study were 124 mg/100 ml. The expected drunkenness and impairments to speed and accuracy of performance were seen in both settings. Performance was slower in the everyday setting, and the ethanol impairment greater, particularly for errors, although the inferred ethanol levels in the everyday setting were somewhat lower. The mean number of incorrect responses for number pairs (attention) was 6.52 for placebo, 9.28 for ethanol in the lab (p<0.01, ANOVA); and 6.45 (no ethanol), 12.2 (ethanol) in the everyday setting (p<0.05 Paired t-test). Laboratory and everyday assessments differ in many ways, including: the rate of drinking, distraction, time of day, and social context. It is therefore not surprising that results are not identical. The poorer performance in the everyday setting could be due to greater distraction, which should be further investigated. Two overall conclusions may be drawn (1) Performance impairments are found in both settings, and are at least as great in real life as in the lab, and (2) Lab results suggesting that errors are an important aspect of alcohol impairment are supported by this study of volunteers in their everyday circumstances. No external funding.

MD10

EFFECTS OF ACUTE ALCOHOL CONSUMPTION ON THE PROCESSING OF PERCEPTUAL CUES OF EMOTIONAL EXPRESSION Ataya AA, Attwood. A, Benton C, Penton-Voak I, Munafò M. Department of Experimental Psychology, University of Bristol, 12a Priory Road, Bristol, BS8 ITU, [email protected]

Introduction: The mechanisms underlying the relationship between alcohol and aggression are not particularly well understood. Alcohol may facilitate aggression via alterations in the processing of the emotional content of facial cues. Studies have reported impairments in the processing of emotional facial cues in alcohol dependent participants (Townshend & Duka 2003). Recently, studies have shown modified processing of emotional facial cues after acute doses of alcohol in non-dependent social drinkers (Kano et al. 2003). The studies reported here further explore these effects using adapted psychophysical tasks in order to measure threshold sensitivity and categorization of emotional expressions, and examining the effects of alcohol dose and expectancy. The effect of alcohol dose on the processing of facial cues in male and female social drinkers was also examined. Both studies were funded by the Alcohol Education and Research Council (AERC) Method: Study 1 (n = 100) was a between-subjects balanced placebo design, in which participants attended one session (0.0 or 0.4 g/kg alcohol) and were randomly allocated to one of four groups; received alcohol/told alcohol, received alcohol/told placebo, received placebo/told alcohol, received placebo/told placebo. A psychophysical task in which two faces were presented for each trial (neutral vs emotional) was employed, and participants were required to identify the emotional face. This task enables identification of perceptual sensitivity to small changes in facial emotional expressions. Sad, happy and angry emotional expressions were tested in male and female target faces. Study 2 (n = 96) employed a similar design to Study 1 however a miscategorisation task was used. A target face was presented consisting of a morph between two emotional exemplars (e.g., happy and angry face) and participants were asked to identify the emotion of the face (i.e., happy or angry). This task was run separately for angry-happy and angry-disgusted facial morphs. Results: : Data were analyzed within 2x2x2x2 mixed model ANOVAs with drink (alcohol, placebo), expectancy (told alcohol, told placebo) and participant sex (male, female) as between subject factors and target sex (male, female) as a within-subjects factor. Emotion was also included as a within-subjects factor compromising there levels for Study 1 (happy, angry, sad) and two levels for Study 2 (angry-happy, angry- disgusted).Study one revealed a near significant emotion by drink interaction (F [2, 178] = 2.94, p = 0.055), with higher thresholds after alcohol for sad, but not happy or angry, emotional expressions .Study two indicated a significant emotion x target sex x alcohol interaction (F [1, 72] = 5.52, p = 0.02), with participants showing a bias towards categorisation of disgusted faces as angry after alcohol but not after placebo consumption. There were no effects of alcohol on the angry-happy categorisation condition (ps > 0.05) Conclusions: These data suggest that alcohol may differentially affect processing of different emotional expressions. In study one, after alcohol consumption, participants showed reduced sensitivity to recognising sad emotion in faces compared to placebo, but no effects were found for angry or happy emotions. Study two showed that alcohol may lead to individuals miscategorising negative (disgusted), but not positive (happy), faces as angry, which has implications for real world situations in which a negative facial expression may be erroneously perceived as provocative. However these miscategorisation effects were obtained in male, but not female, targets, possibly due to greater expectancy of alcohol-related aggression in men. Soc Psychiat Epidemiol (2009) 44:1075–1085 DOI 10.1007/s00127-009-0030-8

ORIGINAL PAPER

A cross-sectional study of mental health and well-being and their associations in the UK veterinary profession

David J. Bartram Æ Ghasem Yadegarfar Æ David S. Baldwin

Received: 5 September 2008 / Accepted: 2 March 2009 / Published online: 18 March 2009 Springer-Verlag 2009

Abstract depression, and co-morbid anxiety and depression was Background Veterinary surgeons are at elevated risk of 26.3, 5.8 and 4.5%. 5.4% of respondents were non-drink- suicide, with a proportional mortality ratio around four ers, 32.0% low-risk drinkers, and 62.6% ‘at-risk’ drinkers times that of the general population and approximately (i.e. AUDIT-C score C4 for women, C5 for men). The 12- twice that of other healthcare professions. There has been month prevalence of suicidal thoughts was 21.3%. much speculation regarding possible mechanisms under- Conclusions Compared to the general population, the lying increased suicide risk in the profession but little sample reported high levels of anxiety and depressive empirical research. We aimed to assess the contribution of symptoms; higher 12-month prevalence of suicidal mental health and well-being to the elevated risk, through a thoughts; less favourable psychosocial work characteris- postal questionnaire survey of a large stratified random tics, especially in regard to demands and managerial sample of veterinary surgeons practising within the UK. support; lower levels of positive mental well-being; and Methods A questionnaire was mailed twice to 3,200 higher levels of negative work–home interaction. The veterinary surgeons. Anxiety and depressive symptoms, levels of psychological distress reported suggest ready alcohol consumption, suicidal ideation, positive mental access to and knowledge of lethal means is probably not well-being, perceptions of psychosocial work characteris- operating in isolation to increase suicide risk within the tics, and work–home interaction were assessed using valid profession. and reliable existing instruments and a series of bespoke questions previously developed through informal focus Keywords Mental health Á Veterinary surgeon Á groups. Depression Á Anxiety Á Well-being Results Evaluable questionnaires were returned by 1,796 participants, a response rate of 56.1%. The demographic and occupational profile of respondents was representative Introduction of the UK veterinary profession. The prevalence of ‘caseness’ (i.e. HADS subscale score C8) for anxiety, Members of some occupational groups are at greatly increased risk of suicide [1], this elevated risk being seen among healthcare professionals including doctors [22, 40, 48] pharmacists [29], dentists [2], and nurses [23]. Farmers D. J. Bartram (&) Á D. S. Baldwin are also at increased risk [30]. Division of Clinical Neurosciences, School of Medicine, The absolute number of suicides by veterinary sur- RSH Hospital, University of Southampton, Brintons Terrace, Southampton SO14 0YG, UK geons is low due to the small size of the profession e-mail: [email protected] (approximately 16,000 veterinary surgeons practising in the UK), but the profession is at increased risk of suicide G. Yadegarfar when compared to other occupations and the general Department of Epidemiology and Biostatistics, School of Public Health, population. Proportional mortality ratios (PMRs) in Eng- Isfahan University of Medical Sciences, Isfahan, Iran land and Wales [29, 32, 34] and Scotland [45], and 123 1076 Soc Psychiat Epidemiol (2009) 44:1075–1085 similar estimates in the USA [36], Australia [26] and Methods Norway [25], indicate that the veterinary profession has around four times the proportion of all deaths certified as The study was reviewed and approved by Southampton and suicide than would be expected from the proportion for South West Hampshire Research Ethics Committee (B) the general population, and around twice that for other (REC reference number: 07/H0504/122). healthcare professionals. While most differences in suicide risk between occupations are accounted for by Sample differences in income and employment status, the most striking exceptions are for vets, doctors, nurses and A random stratified sample of 3,200 veterinary surgeons pharmacists, all having significantly higher rates of sui- practising in the UK was identified. This number represents cide, even when demographic factors are taken into approximately 20% of the membership of the Royal Col- account [1, 9, 43]. Lower mortality from other causes lege of Veterinary Surgeons (RCVS), excluding those may account for the high PMRs for suicide identified for practising overseas or retired. Veterinary surgeons listed in some high social class occupational groups, including the sampling frame (Vetfile, Veterinary Business Devel- healthcare professionals [29]. opment Ltd) were stratified according to type of work There has been much speculation regarding possible within the profession and selected at random within each mechanisms underlying increased suicide risk in the stratum in proportion to the number of veterinary surgeons profession but little empirical research. A hypothetical in each type of work practising in the UK. The gender and model postulates an interplay between various potentially decade of qualification profile of the entire study cohort malign influences [3]. Such influences include the char- was then compared with RCVS membership data for vet- acteristics of individuals entering the profession, negative erinary surgeons practising in the UK to check the effects during undergraduate training, work-related representativeness of the sample. Assuming a response rate stressors, ready access to and knowledge of lethal means, of 50% (1,600 completed questionnaires), point estimates stigma associated with mental illness, professional and of prevalence ranging from 10 to 50% would have confi- social isolation, alcohol or drug misuse, and veterinary dence intervals of ±1.39 to ±2.32, respectively at 95% surgeons’ attitudes to death and euthanasia. These occu- confidence level. pation-specific factors are assumed to act in association with other variables known to be more widely associ- Data collection ated with completed suicide, including male gender and single status [29], the presence of anxiety or depressive Questionnaires were mailed with a covering letter and symptoms and the presence of recurrent suicidal thoughts stamped addressed envelope for return to each member of [7]. the sample (in October and November 2007). The follow- We aimed to assess the contribution of mental ill health up mailing was sent to all veterinary surgeons in the and well-being to the elevated risk, through a postal sample, as it was not possible to identify those who had questionnaire survey of a large stratified random sample of already responded anonymously. Data entry was automated veterinary surgeons practising within the UK. Our primary by using an electronic optical reading system with related hypothesis was that the veterinary profession has higher software TeleForm (Verity Inc., Sunnyvale, CA) to scan levels of mental ill health, lower levels of mental well- the returned questionnaires. being and less favourable psychosocial working conditions, when compared to the general population. The secondary Measures hypothesis was that self-reported measures of mental ill health, mental well-being and psychosocial working con- The 4-page questionnaire comprised a total of 120 items ditions differ significantly with demographic factors and which included: occupational factors within the profession. Research into suicide among veterinary surgeons is • Background information consisting of brief demo- important, not only with a view towards enhancing the graphic and occupational details, including age, sex, well-being of individuals within the profession, but also to main type of work in the profession, position in the help mitigate the potentially deleterious impact of any practice (if applicable), hours of work and hours on call mental ill health among practitioners on the health and in a typical week. welfare of animals under their care, and the additional • The hospital anxiety and depression scale (HADS): this insight that research in this professional group might pro- is a 14-item self-report measure of the prevalence and vide into influences on mental health and well-being in severity of both anxiety (HADS-A; 7 items) and other occupations. depressive (HADS-D; 7 items) symptoms separately, 123 Soc Psychiat Epidemiol (2009) 44:1075–1085 1077

developed for use in general medical outpatient clinics most favourable working conditions (low risk of stress and now extensively validated and widely used in at work), respectively. The overall score for each of the clinical practice and research [42, 52]. Items in both seven stressor domain scales is calculated for each subscales are scored on a 4-point Likert scale (0–3), respondent by adding the item scores for each question resulting in a range of 0–21. A cutoff score of C8is answered in that scale and dividing by the total number used as an indicator of ‘caseness’, with score 8–10 of questions answered in that scale. Cronbach’s alpha indicating possible case and score C11 indicating reliability coefficient demonstrated satisfactory internal probable case [4, 42]. The two subscales demonstrated consistency in our sample for each stressor domain high internal consistency in our sample (HADS-A, (demands a = 0.84; control a = 0.81; managerial Cronbach’s a = 0.85; HADS-D, Cronbach’s support a = 0.84; peer support a = 0.79; relationships a = 0.79). a = 0.79; role a = 0.78; change a = 0.62). • The alcohol use disorders identification test alcohol • Negative and positive work–home interaction (WHI_N consumption questions (AUDIT-C) [8]: this grades and WHI_P) subscales of the SWING instrument [16]. reported alcohol consumption, measuring frequency of This comprises a total of 13 items (WHI_N: 8 items; drinking, typical quantity consumed and frequency of WHI_P: 5 items), each scored on a 4-point Likert scale heavy drinking. Each question is scored on a 5-point from 0–3. The overall score for each subscale is Likert scale (0–4), resulting in a range of 0–12. Cutoff calculated for each respondent by summating the item scores of C4 for women and C5 for men are used as an scores for each of the questions answered and dividing indicator of ‘at-risk’ drinking [18]. The scale demon- by the total number of questions. The two subscales strated moderate internal consistency in our sample demonstrated satisfactory internal consistency in our (Cronbach’s a = 0.70). sample (WHI_N, Cronbach’s a = 0.89; WHI_P, Cron- • Three questions on suicidal ideation derived from the bach’s a = 0.75). second National Survey of Psychiatric Morbidity [41]. • A series of 27 original items specifically focusing on These were originally sourced from the 5-item ques- potential sources of stress in the veterinary profession tionnaire developed by Paykel and others [38] and (a domain of 9 items referred to clinical work and was referred to the previous 12 months (yes/no): Have you only completed by respondents to whom this domain felt that life was not worth living?; Have you wished was relevant). Examples include: client expectations, that you were dead?; Have you thought of taking your euthanasia of animals, dealing with client grief, life, even if you would not really do it? The scale administrative and clerical tasks. These were developed demonstrated high internal consistency in our sample in collaboration with informal focus groups and revised (Cronbach’s a = 0.85). following pre- and pilot testing. Respondents scored on • The Warwick-Edinburgh mental well-being scale (WE- a 5-point Likert scale (0–4; from ‘not at all’ to ‘very MWBS) [47]. This is a 14-item measure for assessing much’) how much each item contributed to the stress population positive mental health which captures a they experienced. wide conception of mental well-being including affec- • An open question inviting respondents to identify in tive–emotional aspects, cognitive–evaluative free text up to three main sources of pleasure and/or dimensions and psychological functioning. Each item satisfaction in practice. Responses were grouped is answered on a 5-point Likert scale (1–5), giving a according to themes and a coding frame was developed minimum score of 14 and a maximum score of 70; a to describe the thematic content. higher score indicates a higher level of mental well- being. The scale demonstrated high internal consistency in our sample (Cronbach’s a = 0.94). Statistical analysis • The Health and Safety Executive management standards indicator tool (HSE MSIT) [11]. This measures Mean imputation was used to replace missing scale or perceptions of psychosocial work characteristics and subscale scores provided that no more than one (HADS comprises 35 questions grouped into seven key stressor subscale; HSE MSIT subscales, except demands and con- domains: demands (8 items), control (6 items), mana- trol), two (demands and control subscales of HSE MSIT) gerial support (5 items), peer support (4 items), or three (WEMWBS) values were missing. If greater than relationships (4 items), role (5 items), and change (3 the specified number of values was missing, then the scale items), which have the potential to have a negative or subscale was judged as invalid for that respondent and impact on employee mental health and well-being. not included in the analysis. Each question scores 1–5 from the least favourable Descriptive statistics including mean, standard deviation, working conditions (high risk of stress at work) to the median and proportions were used to describe the data. To 123 1078 Soc Psychiat Epidemiol (2009) 44:1075–1085 explore the data initially, means were compared using t test and one-way analysis of variance (one-way ANOVA) or the non-parametric equivalents, Mann–Whitney U and Krus- kal–Wallis tests. Chi-square test was employed to compare proportions. To investigate the effect of predictor variables on response variables, multiple linear regression and 0.218 logistic regression were applied to continuous and binary 0.288 = = P P outcome variables, respectively. Odds ratios (ORs) and Co-morbid anxiety and depression Probable D and Probable A regression coefficients were accompanied by 95% confidence intervals (CIs). Statistical significance was defined as (%) ?

P \ 0.05. Statistical analysis was performed using Statis- Probable case tical Package for the Social Sciences (SPSS) version 15.0 (SPSS Inc.) and STATA version 9.0 (StataCorp LP). ] 12 Possible case 0.367 0.598 = Results = P P HADS-D Depression subscale Non- case

Evaluable questionnaires were returned by 1,796 partici- (%) 0–7 (%) 8–10 (%) 11 pants, a response rate of 56.1%. ? Probable case

Demographic characteristics of respondents

The mean age of respondents was 40.9 years (SD = 11.0), Possible case 0.001 and 50.0% were male. 83.9% of respondents worked in 0.001 \ general practice, of which 69.2% reported small animal \ P Non- case P HADS-A Anxiety subscale practice as their main type of work. The median duration of hours worked and on call in a typical week for study respondents working full-time (n = 1458) is 48 h (mean = 47.8, SD = 8.8) and 15 h (mean = 20.2,

SD = 24.9), respectively. a 0.001 The demographic and occupational proﬁle of study 0.001 \ respondents was generally in close alignment with that of \ P P RCVS membership which suggests that study respondents are representative of the wider population of veterinary surgeons practising in the UK.

HADS 0.019 0.268 = Table 1 shows that, in contrast to HADS-D scores, HADS-A = P scores differ significantly between genders and across age P groups. Female veterinary surgeons have significantly higher HADS-A mean score (F = 45.8, P \ 0.001) and proportion of HADS-A cases (v2 = 32.0, P \ 0.001). HAD-A mean score (F = 19.6, P \ 0.001) and the proportion of HADS-A cases (v2 = 37.1, P \ 0.001) decline 0.001 in higher age groups. The HADS-A, HADS-D and HADS-T 0.001 \ \ P P Mean SD Median Mean SD Median Mean SD Median 0–7 (%) 8–10 (%) 11 mean scores for veterinary surgeons are significantly higher HADS-A HADS-D HADS-T than for the general population [12] (HADS-A, 7.9 vs. 6.1

[t(3547) = 13.57, P \ 0.001]; HADS-D, 4.6 vs. 3.9 n [t(3547) = 6.41, P \ 0.001]; HADS-T, 12.6 vs. 9.8 HADS-subscale and -total scores, and prevalence of anxiety and depression caseness for the total sample, by gender and by age group [t(3547) = 13.01, P \ 0.001]). Possible or probable clinically signiﬁcant depression and anxiety are deﬁned by a value value HADS-T is the summated total of the item scores for each of the subscales. It may be used as a global measure of psychological distress [ 75 5 2.6 1.1 3 2.4 1.1 2 5.0 2.0 6 100 – – 100 – – – 35 613 8.6 4.0 8 4.5 3.3 4 13.1 6.6 12 42.2 27.6 30.2 82.2 12.2 5.5 4.9 Age 35–5455–74C 913P 7.9 217 6.3 4.1 8 3.9 6 4.8 4.2 3.5 4 3.4 3 12.7 10.5 6.8 12 6.7 10 48.9 63.1 24.5 22.1 26.6 14.7 78.6 83.4 15.2 12.9 6.1 3.7 4.6 2.3 TotalGender Male 1,757Female 7.9P 881 876 4.1 7.3 8 8.6 4.0 4.6 4.1 7 8 3.4 4.6 4 4.7 3.4 12.6 3.5 4 4 6.8 12 11.8 13.3 48.4 6.6 6.9 11 13 25.3 54.5 42.3 26.3 24.3 26.3 80.6 21.1 31.4 13.6 81.6 79.7 5.8 13.0 14.3 4.5 5.5 6.1 4.0 5.0 Table 1 Characteristic a cutoff score of C8 on the HADS-A and HADS-D scales \ 123 Soc Psychiat Epidemiol (2009) 44:1075–1085 1079

is 2% decline for each additional year (OR = 0.98, 95% CI: 0.97–0.99, P \ 0.001). There is no association between age and HADS-D caseness after adjustment for gender (OR = 1.00, 95% CI: 0.99–1.01, P=0.767). Veterinary surgeons working in university-based non-clinical roles have signiﬁcantly lower estimated risk of HADS-A caseness than those working in small animal practice (OR = 0.27, 95% CI: 0.12–0.63, P = 0.003).

AUDIT-C

Fig. 1 Prevalence of anxiety, co-morbid depression and anxiety, and The proportion of veterinary surgeons in each AUDIT-C depression caseness. COM co-morbid depression and anxiety. Prev- alence of HADS-A or HADS-D score C11 (probable case). Figures in category is displayed in Table 2. There is a significant parentheses represent prevalence of HADS-A or HADS-D score 8–10 difference between males and females across AUDIT-C (possible case) categories (v2 = 6.85, P = 0.033). Around one in 20 veterinary surgeons (5.4%, 95% CI: 4.4–6.6%) report not [43]. The prevalence of anxiety, depression, and co-morbid drinking alcohol. Women are more likely than men to be anxiety and depression caseness among veterinary surgeons non-drinkers (6.7 vs. 4.1%, v2 = 6.03, P = 0.014). Almost is illustrated in Fig. 1. The prevalence of HADS-A probable two-thirds of veterinary surgeons (total 65.1%; men 70.7%, cases is 2.1 times higher and the prevalence of HADS-D women 59.5%) drink more than twice a week, and 38.1% probable cases is 1.6 times higher than for the UK adult of men and 24.3% of women drink four or more times a general population (HADS-A, 26.3% (95% CI: 24.3– week. The difference between males and females across 28.4%) vs. 12.6%; HADS-D, 5.8% (95% CI: 4.8–7.0%) drinking frequencies is significant (v2 = 45.65, vs. 3.6%) and the difference between the distribution of P \ 0.001). One in four men (24.5%) and one in eight veterinary surgeons and the general population across women (12.5%) who drink consume five or more units of HADS-A and HADS-D case categories is significant alcohol on a typical day when drinking. The difference (v2 = 113.8, P \ 0.001). between the typical quantities consumed by males and As HADS outcomes differed significantly between females is significant (v2 = 60.54, P \ 0.001). For 2.2% of genders and across age groups, a logistic regression anal- veterinary surgeons (3.4% of men and 0.9% of women) ysis was performed to determine the estimated relative risk drinking six or more drinks on one occasion is a daily or of HADS-A or HADS-D caseness (score C 8) after almost daily occurrence and for 15.9% (21.3% of men and adjusting for the variables of gender and age. Females are 10.6% of women) it is a weekly occurrence. The difference at 38% greater risk of HADS-A caseness after adjustment between males and females in the frequency of drinking six for age (OR = 1.38, 95% CI: 1.12–1.69, P = 0.002). or more units on one occasion is significant (v2 = 68.33, There is no significant difference between genders for P \ 0.001). The highest prevalence of at-risk drinkers HADS-D caseness after adjustment for age (OR = 1.14, (73.7%) and lowest prevalence of non-drinkers (3.7%) is in 95% CI: 0.89–1.47, P = 0.300). Gender adjusted ORs for the 20–29 years age group (n = 297). The prevalence of age groups indicate a linear decline in the risk of HADS-A at-risk drinkers falls to 40.0% for the oldest age group caseness with increasing age in comparison with the (70? years). At-risk drinking is significantly associated youngest age group (20–29 years). As a linear trend is with age after adjustment for gender: a 1-year increase in seen, age was entered into the logistic regression model as age is associated with a 2% reduction in the risk of at-risk a continuous variable to assess the dose–response rela- drinking (OR = 0.98, 95% CI: 0.97–0.99, P = 0.001). tionship between age and HADS-A. The gender adjusted In comparison with alcohol consumption figures for the predicted overall age response relationship with HADS-A general population [10], veterinary surgeons in our sample

Table 2 Proportion of n Non-drinkers Low-risk drinkers (%) At-risk drinkers (%) veterinary surgeons in each (%) Score 1 to 4 for men Score C5 for men AUDIT-C drinking category Score 0 score 1 to 3 for women score C4 for women

Total 1,757 5.4 32.0 62.6 Male 881 4.1 33.5 62.4 Female 876 6.7 30.6 62.7

123 1080 Soc Psychiat Epidemiol (2009) 44:1075–1085

Table 3 Twelve-month prevalence of suicidal ideation Characteristic Life not worth livinga Wished you were deadb Suicidal thoughtsc Any suicidal ideationd n % OR 95% CI n % OR 95% CI n % OR 95% CI n % OR 95% CI

Total 1,751 23.0 – – 1,751 15.0 – – 1,749 21.3 – – 1,750 29.4 – – Gender Male 878 20.6 1 – 879 13.5 1 – 878 20.2 1 – 878 27.4 1 – Female 873 25.4 1.31* 1.05–1.64 872 16.5 1.26 0.97–1.64 871 22.5 1.15 0.92–1.45 872 31.3 1.20 0.98–1.48 Age \35 611 23.7 1 – 611 15.5 1 – 610 22.8 1 – 611 30.4 1 – 35–54 910 22.3 0.92 0.72–1.18 911 14.9 0.95 0.72–1.27 909 20.9 0.90 0.70–1.15 909 28.3 0.90 0.72–1.13 55–74 216 23.1 0.97 0.67–1.40 215 12.6 0.78 0.49–1.23 216 19.4 0.82 0.56–1.20 216 31.0 1.03 0.73–1.44 C75 50 – – 50 – – 50 – – 50 – – OR odds ratio, CI confidence interval * P = 0.017 a A positive response to the question: ‘Have you felt that life was not worth living?’ b A positive response to the question: ‘Have you wished that you were dead?’ c A positive response to the question: ‘Have you thought of taking your life, even if you would not really do it?’ d A positive response to any of the three questions above; an extended definition of suicidal thoughts used by Gunnell and Harbord [19] are less likely to be non-drinkers (5 vs. 12%), drink more thoughts by 1.9% (OR = 1.02, 95% CI: 1.01–1.03, frequently than the general population (65 vs. 48% drink P = 0.007). By contrast, risk is unaffected by an increase more than twice per week), but consume less on a typical in number of hours on call in a typical week (OR = 1.00, drinking day and a have a prevalence of daily and weekly 95% CI: 1.00–1.01, F = 10.9, P = 0.527). binge-drinking that is similar to the general population. A 1-unit increase in HADS-A score is associated with a 15% higher estimated relative risk of reporting having Suicidal ideation experienced suicidal thoughts in the previous 12 months (OR = 1.15, 95% CI: 1.09–1.20, P \ 0.001); a 1-unit The proportion of veterinary surgeons giving a positive increase in the relationships domain score for HSE MSIT response to each of the questions regarding negative is associated with a 22% lower risk (OR = 0.78, 95% CI: thoughts about their life in the previous 12 months is dis- 0.62–0.99, P = 0.045); and a 1-unit increase in WEMWBS played in Table 3. The 12-month prevalence of any score is associated with a 7% lower risk (OR = 0.93, 95% suicidal ideation is 29.4% (life was not worth living, CI: 0.90–0.96, P \ 0.001). 23.0%; death wishes, 15.0%; suicidal thoughts, 21.3%). In the second National Survey of Psychiatric Morbidity Women are more likely than men to think that life is not [41] of adults in Great Britain, which used identical sui- worth living (OR = 1.31, 95% CI: 1.05–1.64, P = 0.017) cidal ideation questions, the 12-month prevalence of but there is otherwise no statistically significant difference suicidal thoughts was 3.9% (4.1% of women, 3.6% of men) between genders or the age groups examined. Logistic [35]; the prevalence was three or more times higher among regression analysis with adjustment for age and gender was younger people and the middle-aged than among older used to identify demographic and occupational risk factors people [14]. The 12-month prevalence (21.3%, 95% CI: for suicidal thoughts. Veterinary surgeons working in 19.5–23.3) of suicidal thoughts in this sample is around 5.5 mixed practice (OR = 0.46, 95% CI: 0.29–0.71, times higher than among the general population. Although P = 0.001) and in university-based clinical roles the gender difference in 12-month prevalence of suicidal (OR = 0.29, 95% CI: 0.09–0.94, P = 0.039) are at thoughts among veterinary surgeons is not statistically reduced risk of suicidal thoughts than those working in significant, the higher prevalence among women is con- small animal practice (the reference category). Veterinary sistent with the gender difference in this parameter among surgeons employed as full-time assistants are at increased the general population in Great Britain [35]. The 12-month risk of suicidal thoughts (OR = 2.07, 95% CI: 1.15–3.73, prevalence of suicidal thoughts among veterinary surgeons P = 0.016) than those working as sole principals. Longer is similar across all the age groups examined, in contrast to working hours is also a risk factor: a 1-h increase in hours the decline in prevalence with increasing age observed in worked in a typical week increases the risk of suicidal the general population in Great Britain [14].

123 Soc Psychiat Epidemiol (2009) 44:1075–1085 1081

WEMWBS (high risk of stress at work). The scores derived for each scale cannot be compared across scales [51]. The mean WEMWBS score for the sample of veterinary Veterinary surgeons self-report less favourable psycho- surgeons is 48.85 (SD = 9.06) and the score is signifi- social work characteristics than the general population cantly higher for men than for women (49.86 vs. 47.83, across all of the seven stressor domains of HSE MSIT [51] P \ 0.001). There is a significant relationship between and the differences are statistically significant (except for WEMWBS and age adjusted for gender: the score increa- the control domain). The greatest differences between the ses by 0.05 for every 1-year increase in age (b = 0.05, mean scores for the veterinary profession and the general 95% CI: 0.01–0.09, P=0.012). There is also a significant population are for the demands [2.96 vs. 3.57; difference in mean WEMWBS between genders adjusted t(2267) = 16.50, P \ 0.001] and managerial support [3.14 for age: the mean score for females is 1.65 less than vs. 3.76; t(2252) = 13.74, P \ 0.001] scales. In keeping the mean score for males (b =-1.65, 95% CI: -2.55 to with other studies that consider gender-related differences -0.74, P \ 0.001). The mean scores for veterinary sur- in stress, e.g. [13], male veterinary surgeons report more geons working in university-based non-clinical university- favourable working conditions associated with control, based clinical roles are higher in comparison with those relationships, role and change than female veterinary working in small animal practice after adjusting for age and surgeons. Younger veterinary surgeons (\49 years) report gender (university–non-clinical: b = 3.50, 95% CI: 0.37– the least favourable working conditions across all stressor 6.62, P=0.029; university–clinical: b = 2.76, 95% CI: domains. Reported working conditions for demands, man- 0.48–5.03, P=0.018). agerial support and peer support domains are least The mean WEMWBS scores for the total sample of favourable for those working the longest hours. Reported veterinary surgeons, and for male and female veterinary working conditions for the seven stressor domains vary surgeons separately, are significantly lower than the cor- with other occupational factors. For example, veterinary responding means for a representative general population surgeons working for charities and in university-based sample from Scotland [6] (48.85 vs. 51.05 [t(2728) = clinical roles report the least favourable working conditions -6.20, P \ 0.001]; 49.86 vs. 51.21 [t(1408) =-2.80, related to demands and control; those working in govern- P = 0.003]; 47.83 vs. 50.92 [t(1561) =-6.84, ment roles report the least favourable working conditions P \ 0.001], respectively). related to relationships and role.

Psychosocial work characteristics Work–home interaction

Mean and median scores for each of the HSE MSIT scales The mean WHI_N score for the total sample (n = 1,749) is for the veterinary profession are displayed in Table 4.A 1.19 (SD = 0.57). Multiple linear regression indicated that low score indicates less favourable working conditions there is no signiﬁcant difference between mean score for

Table 4 Stressor domain scores for the veterinary profession Demands Control Managerial support Peer support n Mean SD Median n Mean SD Median n Mean SD Median n Mean SD Median

Total 1,762 2.96 0.70 3.00 1,749 3.47 0.78 3.50 1,472 3.14 0.89 3.2 1,736 3.75 0.73 3.75 Gender Male 876 2.97 0.71 3.00 878 3.70 0.72 3.67 681 3.17 0.90 3.2 868 3.74 0.73 3.75 Female 871 2.96 0.69 3.00 871 3.24 0.76 3.33 791 3.11 0.88 3.2 868 3.75 0.73 3.75 Relationships Role Change n Mean SD Median n Mean SD Median n Mean SD Median

Total 1,743 4.01 0.69 4.25 1,737 4.21 0.63 4.20 1,711 3.22 0.94 3.33 Gender Male 875 4.06§ 0.66 4.25 873 4.28 0.62 4.40 858 3.44 0.89 3.67 Female 868 3.96 0.71 4.00 864 4.15 0.65 4.20 853 2.99 0.93 3.00

§ P \ 0.01, one-way ANOVA; P \ 0.001, one-way ANOVA

123 1082 Soc Psychiat Epidemiol (2009) 44:1075–1085 men and women after adjusting for age, although the mean Non-response bias is significantly higher for women prior to adjustment for age (1.23 vs. 1.14, F = 11.3, P=0.001). There is a sig- The demographic and occupational profile of study nificant relationship between mean score and age after respondents was generally in close alignment with that of adjustment for gender; the score decreases by 0.010 for the original sample. A comparative analysis to identify every one year increase in age (95% CI: -0.012 to -0.007, any differences between outcome measures for late P \ 0.001). Veterinary surgeons working in government responders (n = 272) and earlier responders showed that and university-based non-clinical roles have lower mean the HADS scores and 12-month prevalence of suicidal WHI_N score than those working in small animal practice thoughts were higher and WEMWBS score lower among after adjusting for age and gender (Government: b =-0.13, late responders, but the difference was only significant for 95% CI: -0.25 to -0.02, P=0.021; University non- HADS-D (mean = 5.11 vs. 4.57, t(1760) =-2.40, clinical: b =-0.23, 95% CI: -0.43 to -0.04, P=0.019). P = 0.017). The mean WHI_P score for the total sample (n = 1,716) is 0.97 (SD = 0.56). Multiple linear regression indicated that there is no significant difference between mean score for Discussion men and women after adjusting for age, although the mean is significantly lower for women prior to adjustment for age Main findings (0.92 vs. 1.01, F = 10.9, P=0.001). There is a significant relationship between mean score and age after adjustment Compared to the general population, the sample of veter- for gender: the score increases by 0.005 for every 1-year inary surgeons reported high levels of anxiety and increase in age (95% CI: 0.002–0.007, P \ 0.001). depressive symptoms; higher 12-month prevalence of sui- The mean WHI_N score for the sample is higher than cidal thoughts; less favourable psychosocial working for a sample of the working population from the Nether- characteristics, especially in regard to demands and man- lands [16] and lower than for Belgian veterinary surgeons agerial support; lower levels of positive mental well-being; [20]. The mean WHI_P score for the sample is higher than and higher levels of negative work–home interaction. The for the sample of the working population from the Neth- levels of psychological distress reported suggest ready erlands [16] and lower than for Belgian veterinary surgeons access to and knowledge of lethal means is probably not [20]. The results suggest that veterinary surgeons report operating in isolation to increase suicide risk within the higher levels of both negative and positive work–home profession. interaction, but comparative data must be interpreted cau- The prevalence of possible co-morbid anxiety and tiously due to possible cultural differences in interpretation depression among this sample of veterinary surgeons of the scale items and the small size of the Belgian sample. [defined by coexisting probable and possible cases (scores C 8)] is 16.8%. The presence of co-morbid anxiety Contributors to stress disorders increases the risk of suicidal behaviour in people with depression [21, 39], and there is evidence to suggest Number of hours worked, making professional mistakes, that co-morbidity, and not depression itself, is a risk factor client expectations, and administrative and clerical tasks for suicide attempts [17] and completed suicide [46]. are reported as the greatest contributors to stress for the Alcohol has a well-established role in suicidal behaviour sample population of veterinary surgeons. The percentage through its induction of negative affect, promotion of of respondents reporting that these stressors contribute adverse life events, impairment of problem-solving skills, quite a lot or very much to their stress is 42.9, 40.4, 38.0 disinhibitory effects and the social disintegrative effects of and 27.9%, respectively. Respondents who treated clinical abuse [5]. However, the level of alcohol consumption does cases were asked a further set of questions regarding pos- not appear to be a negative influence on mental health sible stressors. The possibility of client complaints or within the profession as a whole. litigation, unexpected clinical outcomes and out-of-hours Suicidal thoughts are a key stage in the pathway on-call duties are reported as the greatest contributors to leading to suicide [19]. Positive associations have been stress for respondents treating clinical cases. demonstrated between favourable attitudes towards suicide (the degree to which an individual views suicide as Sources of satisfaction an acceptable option under some circumstances) and levels of suicidal thoughts [15, 17]. Favourable attitudes The greatest sources of satisfaction cited included: good towards suicide appear to increase the attractiveness of clinical outcomes (41.5%), relationships with colleagues suicide should situational cues arise, placing an individual (33.7%) and intellectual challenge/learning (32.4%). at increased risk of suicidal ideation. The high prevalence 123 Soc Psychiat Epidemiol (2009) 44:1075–1085 1083 of suicidal ideation among veterinary surgeons may be may be important in the aetiology of mood disorders associated with the profession’s acceptance of and among the working population [50]. familiarity with animal euthanasia which may change attitudes to suicide as a possible solution to their own Limitations problems [2]. Ready access to means of suicide is posited as a key factor that influences the translation of suicidal The data are limited by the cross-sectional design which thoughts into an actual suicide act [24]. Ready access to allows investigation of associations, but does not allow lethal means may also act more distally in the suicide conclusions to be drawn about the direction of causality. A process and account for the high prevalence of suicidal further limitation in the study methodology is bias related thoughts among veterinary surgeons, if the ease at which to the self-reporting of symptoms and working conditions. a suicide can be completed cues the consideration of The difference between outcome measures for late and suicide as a possible solution. earlier responders suggests that the reported prevalence of Psychosocial working conditions may be important risk mental ill-health may have been higher if the study factors contributing to suicidal behaviours [37] and studies response rate was higher. The content of the questionnaire have demonstrated a causal association between work stress may have led to some selection bias. Veterinary surgeons and rates of depression and anxiety [31, 49]. A recent meta- with mental health problems may have been disinclined to analysis provided robust evidence that a combination of high respond due to their symptoms or concerns that they might demands and low decision latitude, effort–reward imbal- be identified, or conversely those without mental health ance, and low social support at work from co-workers and problems may have considered that the questionnaire was supervisors are risk factors for common mental disorders not relevant to them. The survey did not collect informa- [44]. The less favourable psychosocial working character- tion on some potentially relevant mediating or moderating istics reported in comparison with the general population variables such as marital status, adverse work and life suggest that high demands and low managerial support may events, attitudes to euthanasia and suicide, history of any be important contributors to work-related stress among previous psychological distress, family history of affective veterinary surgeons, in keeping with the demand–control– disorder, use of prescribed or non-prescribed psychotropic support model of work-related stress [27, 28]. medication, duration of exposure to possible explanatory variables (e.g. length of time in current job), social network Strengths size and perceived social support outside work, and personality factors such as negative affectivity and Our study has several strengths. First, it is based on a large attributional style, which may have confounded associa- nationwide sample of veterinary surgeons in a range of tions without adjustment for them. Comparison of different types of employment and the demographic and measures of mental health and well-being for the study occupational profile of respondents is broadly representa- sample with those for the general population must be tive of the UK veterinary profession which gives some undertaken with great care due to the possible influence of confidence in the generalisability of the results. Second, the variables including socio-demographic factors. questionnaire uses standard instruments with known psychometric properties to help ensure the validity and Future research reliability of the results and availability of normative data. The short series of bespoke questions on stressors pertinent There is scope for further research to explore the results of to the veterinary profession was developed using focus the current study with, for example, qualitative interviews groups and subsequently pre- and pilot-tested to help with a purposive sample of respondents, and longitudinal ensure relevance, comprehensiveness and absence of bias. studies would help to determine whether the cross-sec- The response rate of 56% is less than optimal but compares tional associations identified in the current study are causal. favourably with other postal surveys of the veterinary Research is also required to develop interventions and profession (e.g. Mellanby and Herrtage [33]), which may evaluate their effectiveness and utility. reflect a high level of personal salience of the subject of Acknowledgments The study was funded by Veterinary Times and mental health to the sample population. The study adopts a BUPA Giving. We thank the veterinary surgeons throughout the UK comprehensive approach to the assessment of mental who participated in the survey. Catherine Carr provided administra- health in the veterinary profession, seeking to identify tive support. Julia Sinclair kindly reviewed and commented on a late sources of pleasure in veterinary work and complementing draft of the manuscript. indicators of psychological morbidity with measures of Conflict of interest statement The authors declare that they have mental well-being and work–home interaction. The latter no competing interests.

123 1084 Soc Psychiat Epidemiol (2009) 44:1075–1085

References 21. Hawgood J, De Leo D (2008) Anxiety disorders and suicidal behaviour: an update. Curr Opin Psychiatry 21:51–64 1. Agerbo E, Gunnell D, Bonde JP, Mortensen PB, Nordentoft M 22. Hawton K, Clements A, Sakarovitch C, Simkin S, Deeks JJ (2007) Suicide and occupation: the impact of socio-economic, (2001) Suicide in doctors: a study of risk according to gender, demographic and psychiatric differences. Psychol Med 37:1131– seniority and specialty in medical practitioners in England and 1140 Wales, 1979–1995. J Epidemiol Community Health 55:296–300 2. Alexander RE (2001) Stress-related suicide by dentists and other 23. Hawton K, Vislisel L (1999) Suicide in nurses. Suicide Life health care workers. Fact or folklore? J Am Dent Assoc 132:786– Threat Behav 29:86–95 794 24. Hawton K (2007) Restricting access to methods of suicide: 3. Bartram DJ, Baldwin DS (2008) Veterinary surgeons and suicide: rationale and evaluation of this approach to suicide prevention. influences, opportunities and research directions. Vet Rec Crisis 28(Suppl 1):4–9 162:36–40 25. Hem E, Haldorsen T, Aasland OG, Tyssen R, Vaglum P, Ekeberg 4. Bjelland I, Dahl AA, Haug TT, Neckelmann D (2002) The O (2005) Suicide rates according to education with a particular validity of the hospital anxiety and depression scale. An updated focus on physicians in Norway 1960–2000. Psychol Med 35:873– literature review. J Psychosom Res 52:69–77 880 5. Brady J (2006) The association between alcohol misuse and 26. Jones-Fairnie H, Ferroni P, Silburn S, Lawrence D (2008) Suicide suicidal behaviour. Alcohol Alcohol 41:473–478 in Australian veterinarians. Aust Vet J 86:114–116 6. Braunholtz, S, Davidson S, Myant K, O’Connor R. (2007) Well? 27. Karasek R, Theorell T (1990) Healthy work: stress, productivity, What do you think? (2006): the third national Scottish survey of and the reconstruction of working life. Basic Books, New York public attitudes to mental health, mental wellbeing and mental 28. Karasek RA Jr (1979) Job demands, job decision latitude, and health problems. The Scottish Executive, Edinburgh mental strain: implications for job redesign. Adm Sci Q 24:285– 7. Bronisch T, Wittchen HU (1994) Suicidal ideation and suicide 308 attempts: comorbidity with depression, anxiety disorders, and sub- 29. Kelly S, Bunting J (1998) Trends in suicide in England and stance abuse disorder. Eur Arch Psychiatry Clin Neurosci 244:93–98 Wales, 1982–96. Popul Trends 92:29–41 8. Bush K, Kivlahan DR, McDonell MB, Fihn SD, Bradley KA 30. Malmberg A, Simkin S, Hawton K (1999) Suicide in farmers. Br (1998) The AUDIT alcohol consumption questions (AUDIT-C): J Psychiatry 175:103–105 an effective brief screening test for problem drinking. Arch Intern 31. Melchior M, Caspi A, Milne BJ, Danese A, Poulton R, Moffitt TE Med 158:1789–1795 (2007) Work stress precipitates depression and anxiety in young, 9. Charlton J (1995) Trends and patterns in suicide in England and working women and men. Psychol Med 37:1119–1129 Wales. Int J Epidemiol 24(Suppl 1):S45–S52 32. Mellanby RJ (2005) Incidence of suicide in the veterinary pro- 10. Coulthard M, Farrell M, Singleton N, Meltzer H (2002) Tobacco, fession in England and Wales. Vet Rec 157:415–417 alcohol and drug use and mental health. The Stationery Office, 33. Mellanby RJ, Herrtage ME (2004) Survey of mistakes made by London recent veterinary graduates. Vet Rec 155:761–765 11. Cousins R, MacKay CJ, Clarke SD, Kelly C, Kelly PJ, McCaig 34. Meltzer H, Griffiths C, Brock A, Rooney C, Jenkins R (2008) RH (2004) ‘Management standards’ and work related stress in the Patterns of suicide by occupation in England and Wales: 2001– UK: practical development. Work Stress 18:113–136 2005. Br J Psychiatry 193:73–76 12. Crawford JR, Henry JD, Crombie C, Taylor EP (2001) Normative 35. Meltzer H, Lader D, Corbin T, Singleton N, Jenkins R, Brugha T data for the HADS from a large non-clinical sample. Br J Clin (2002) Non-fatal suicidal behaviour among adults aged 16 to 74 Psychol 40:429–434 in Great Britain. The Stationery Office, London 13. De Smet P, Sans S, Dramaix M, Boulenguez C, de Backer G, 36. Miller JM, Beaumont JJ (1995) Suicide, cancer, and other causes Ferrario M et al (2005) Gender and regional differences in per- of death among California veterinarians, 1960–1992. Am J Ind ceived job stress across Europe. Eur J Public Health 15:536–545 Med 27:37–49 14. Dennis M, Baillon S, Brugha T, Lindesay J, Stewart R, Meltzer H 37. Ostry A, Maggi S, Tansey J, Dunn J, Hershler R, Chen L et al (2007) The spectrum of suicidal ideation in Great Britain: com- (2007) The impact of psychosocial work conditions on attempted parisons across a 16–74 years age range. Psychol Med 37:795–805 and completed suicide among western Canadian sawmill work- 15. Etzersdorfer E, Vijayakumar L, Schony W, Grausgruber A, ers. Scand J Public Health 35:265–271 Sonneck G (1998) Attitudes towards suicide among medical 38. Paykel ES, Myers JK, Lindenthal JJ, Tanner J (1974) Suicidal students: comparison between Madras (India) and Vienna (Aus- feelings in the general population: a prevalence study. Br J tria). Soc Psychiatry Psychiatr Epidemiol 33:104–110 Psychiatry 24:460–469 16. Geurts SAE, Taris TW, Kompier MAJ, Dikkers JSE, Van Hoof 39. Sareen J, Cox BJ, Afifi TO, de Graaf R, Asmundson GJ, ten Have MLM, Kinnunen UM (2005) Work–home interaction from a M et al (2005) Anxiety disorders and risk for suicidal ideation work psychological perspective: development and validation of a and suicide attempts: a population-based longitudinal study of new questionnaire, the SWING. Work Stress 40:429–434 adults. Arch Gen Psychiatry 62:1249–1257 17. Gibb BE, Andover MS, Beach SR (2006) Suicidal ideation and 40. Schernhammer ES, Colditz GA (2004) Suicide rates among attitudes toward suicide. Suicide Life Threat Behav 36:12–18 physicians: a quantitative and gender assessment (meta-analysis). 18. Gual A, Segura L, Contel M, Heather N, Colom J (2002) Audit-3 Am J Psychiatry 161:2295–2302 and audit-4: effectiveness of two short forms of the alcohol use 41. Singleton N, Bumpstead R, O’Brien M, Lee A, Meltzer H (2001) disorders identification test. Alcohol Alcohol 37:591–596 Psychiatric morbidity among adults living in private households, 19. Gunnell D, Harbord R (2003) Suicidal Thoughts. In: Singleton N, 2000. The Stationery Office, London Lewis G (eds) Better or worse: a longitudinal study of the mental 42. Snaith RP (2003) The hospital anxiety and depression scale. health of adults living in private households in Great Britain. The Health Qual Life Outcomes 1:29 Stationery Office, London, pp 45–65 43. Stack S (2001) Occupation and suicide. Soc Sci Q 82:384–396 20. Hansez I, Schins F, Rollin F (2008) Occupational stress, work– 44. Stansfeld S, Candy B (2006) Psychosocial work environment and home interference and burnout among Belgian veterinary prac- mental health—a meta-analytic review. Scand J Work Environ titioners. Ir Vet J 61:233–241 Health 32:443–462

123 Soc Psychiat Epidemiol (2009) 44:1075–1085 1085

45. Stark C, Belbin A, Hopkins P, Gibbs D, Hay A, Gunnell D (2006) 49. Waldenstrom K, Ahlberg G, Bergman P, Forsell Y, Stoetzer U, Male suicide and occupation in Scotland. Health Stat Q 29:26–29 Waldenstrom M et al (2008) Externally assessed psychosocial 46. Stordal E, Morken G, Mykletun A, Neckelmann D, Dahl AA work characteristics and diagnoses of anxiety and depression. (2008) Monthly variation in prevalence rates of comorbid Occup Environ Med 65:90–96 depression and anxiety in the general population at 63–65 degrees 50. Wang JL (2006) Perceived work stress, imbalance between work North: the HUNT study. J Affect Disord 106:273–278 and family/personal lives, and mental disorders. Soc Psychiatry 47. Tennant R, Hiller L, Fishwick R, Platt S, Joseph S, Weich S et al Psychiatr Epidemiol 41:541–548 (2007) The Warwick–Edinburgh mental well-being scale (WE- 51. Webster S, Buckley P, Rose I (2007) Psychosocial working MWBS): development and UK validation. Health Qual Life conditions in Britain in 2007. Health and Safety Executive, Outcomes 5:63 London 48. Torre DM, Wang NY, Meoni LA, Young JH, Klag MJ, Ford DE 52. Zigmund AS, Snaith RP (1983) The hospital anxiety and (2005) Suicide compared to other causes of mortality in physi- depression scale. Acta Psychiatr Scand 67:361–370 cians. Suicide Life Threat Behav 35:146–153

123 The Veterinary Journal 187 (2011) 397–398

Contents lists available at ScienceDirect

The Veterinary Journal

journal homepage: www.elsevier.com/locate/tvjl

Short Communication Validation of the Warwick–Edinburgh Mental Well-being Scale (WEMWBS) as an overall indicator of population mental health and well-being in the UK veterinary profession

David J. Bartram a,*, Ghasem Yadegarfar b,1, Julia M.A. Sinclair a, David S. Baldwin a a University Department of Mental Health, Division of Clinical Neurosciences, School of Medicine, University of Southampton, Royal South Hants Hospital, Brintons Terrace, Southampton SO14 0YG, UK b Research and Development Support Unit, School of Medicine, University of Southampton, Level C, Southampton General Hospital, Tremona Road, Southampton SO16 6YD, UK article info abstract

Article history: The Warwick–Edinburgh Mental Well-being Scale (WEMWBS) was evaluated as an indicator of mental Accepted 15 February 2010 health and well-being within the veterinary profession in a cross-sectional study among a representative sample of 3200 veterinary surgeons practising in the UK. The WEMWBS mean score for the sample was 48.85 (95% conﬁdence interval 48.43–49.28). The score showed a negative correlation with anxiety and Keywords: depressive symptoms and a positive correlation with favourable psychosocial working conditions. A 1 Warwick–Edinburgh Mental Well-being unit increase in score was associated with reduced odds of reporting having experienced suicidal Scale thoughts in the previous 12 months, and reduced odds of reporting depressive or anxiety symptoms of Veterinary surgeons clinical signiﬁcance. The results support the validity of the scale as an overall indicator of population mental health and well-being for this occupational group. Ó 2010 Elsevier Ltd. All rights reserved.

Mental ill-health appears to be prevalent in the UK veterinary score, giving a possible summary score of 14–70; higher scores indi- profession (Bartram et al., 2009). The identification of an instru- cate higher levels of mental well-being (Tennant et al., 2007). Exam- ment with validity as an overall indicator of population mental ples of items used in the WEMWBS include: ‘I’ve been dealing with health and well-being, suitable for inclusion within, for example, problems well’ and ‘I’ve been feeling close to other people’. the Royal College of Veterinary Surgeons (RCVS) Survey of the Pro- The study was reviewed and approved by Southampton and fession, would facilitate the monitoring of trends within the pro- South West Hampshire Research Ethics Committee (B) (REC refer- fession (Bartram et al., in press). This paper investigates the ence number: 07/H0504/122). The WEMWBS was embedded in a potential for the Warwick–Edinburgh Mental Well-being Scale 120-item questionnaire which was mailed in 2007 to a stratified (WEMWBS) (Tennant et al., 2007), an instrument developed to random sample of 3200 veterinary surgeons practising in the UK, monitor positive mental well-being at a population level, to be comprising approximately 20% of the membership of the RCVS. Re- used for this purpose. Associations with other standardised mea- plies were anonymous. The questionnaire assessed multiple sures assessing different dimensions of mental health and working dimensions of mental health and well-being using valid and reli- conditions were examined in a cross-sectional study of veterinary able instruments. Anxiety and depressive symptoms were assessed surgeons practising in the UK. We hypothesised that WEMWBS using the Hospital Anxiety and Depression Scale (HADS) (Zigmund scores would show negative associations with scales assessing psy- and Snaith, 1983), psychosocial working conditions with the chological distress. Health and Safety Executive Management Standards Indicator Tool The WEMWBS comprises 14 positively phrased items which (HSE MSIT) (Cousins et al., 2004), suicidal ideation with questions measure positive affect (such as feelings of optimism, cheerfulness, developed by Paykel et al. (1974) and alcohol consumption with and relaxation), psychological functioning (for example, energy, the Alcohol Use Disorders Identification Test alcohol consumption clear thinking, self-acceptance, and competence) and interpersonal questions (AUDIT-C) (Bush et al., 1998). Further details of the ques- relationships. Each item is scored (based on experience over the pre- tionnaire are reported elsewhere (Bartram et al., 2009). For up to vious 2 weeks) on a 5-point Likert-style scale from ‘none of the time’ three missing WEMWBS responses, the missing scores were im- (1) to ‘all of the time’ (5). The overall score is the sum of each item puted using the mean of the remaining items. If more than 3/14 possible responses were missing, the scale was judged as invalid for that respondent and the case was excluded from the analysis. * Corresponding author. Tel.: +44 2380 825533; fax: +44 2380 234243. Usable questionnaires were returned by 1796 participants, rep- E-mail address: [email protected] (D.J. Bartram). 1 Present address: Department of Biostatistics and Epidemiology, School of Public resenting a response rate of 56.1%. The demographic and occupa- Health Sciences, Isfahan University of Medical Sciences, Isfahan 73461-81746, Iran. tional profile of study respondents was generally in close

I've been feeling optimistic about the future (n=1795) I've been feeling useful (n=1789) I've been feeling relaxed (n=1788) I've been feeling interested in other people (n=1788) I've had energy to spare (n=1787) I've been dealing with problems well (n=1793) I've been thinking clearly (n=1790) I've been feeling good about myself (n=1791) I've been feeling close to other people (n=1788) I've been feeling confident (n=1791) I've been able to make up my own mind about things (n=1789) I've been feeling loved (n=1790) I've been interested in new things (n=1793) I've been feeling cheerful (n=1796)

0% 20% 40% 60% 80% 100%

None of the time Rarely Some of the time Often All of the time

Fig. 1. Warwick–Edinburgh Mental Well-being Scale (WEMWBS) question responses for sample. Imputed values are excluded. alignment with RCVS membership and the original sample (Bar- from those in the current study (Tennant et al., 2007). Further re- tram et al., 2009). The WEMWBS mean score (± standard deviation, search is needed to evaluate the sensitivity of the scale to change SD) for the sample was 48.85 (± 9.06) (95% confidence interval, CI and to determine whether the associations can be generalised to 48.43–49.28) and the median score was 49 (inter-quartile range, other occupational groups. A shortened (7 item) version of the IQR, 43–55). The score distribution was near normal, with a slight WEMWBS, embedded within the 14 item scale, has robust mea- negative skew, and the scale did not show floor or ceiling effects. surement properties and offers an alternative for monitoring men- Item response frequencies showed little evidence of highly skewed tal well-being in populations (Stewart-Brown et al., 2009). distributions, with each of the response categories for all items being used by at least one respondent (Fig. 1). Conflict of interest statement Correlations between scores on the WEMWBS and other scales were calculated using Pearson correlation coefficients (Bonferroni None of the authors of this paper has any financial or personal correction applied). The WEMWBS score showed a high negative relationship with other people or organisations that could inappro- correlation with anxiety (r = À0.69, P < 0.001) and depressive priately bias the content of the paper. symptoms (r = À0.76, P < 0.001). There was a significant (P < 0.001) low-to-moderate positive correlation with subscales Acknowledgements measuring favourable psychosocial working conditions (low risk of work-related stress): demands (r = 0.32), control (r = 0.45), man- We thank the veterinary surgeons throughout the UK who par- agerial support (r = 0.48), peer support (r = 0.50), relationships ticipated in the survey. The study was supported by the Veterinary (r = 0.37), role (r = 0.45) and change (r = 0.41). Times and BUPA Giving. The Warwick–Edinburgh Mental Well- Multiple logistic regression was used to explore any association being Scale is jointly owned by NHS Scotland, the University of between the WEMWBS score and each of the other scales after Warwick and the University of Edinburgh. adjusting for age, gender and the other scales. A 1 unit increase in score was associated with a 7% reduction in the odds of report- References ing having experienced suicidal thoughts in the previous 12 months (odds ratio, OR 0.93, 95% CI 0.90–0.96, P < 0.001). There Bartram, D.J., Yadegarfar, G., Baldwin, D.S., 2009. A cross-sectional study of mental was no association with at-risk drinking (i.e. AUDIT-C score P5 for heath and well-being and their associations in the UK veterinary profession. Social Psychiatry and Psychiatric Epidemiology 44, 1075–1085. men or P4 for women) (OR 0.99, 95% CI 0.97–1.02, P = 0.525). A 1 Bartram, D.J., Sinclair, J.M.A., Baldwin, D.S., in press. Interventions with potential to unit increase in the score was associated with a 15% reduction in improve the mental health and wellbeing of UK veterinary surgeons. Veterinary the odds of reporting anxiety symptoms of possible or probable Record. Bush, K., Kivlahan, D.R., MCDonnell, M.B., Fihn, S.D., Bradley, K.A., 1998. The AUDIT clinical significance (i.e., HADS-A score > 8) or probable clinical sig- alcohol consumption questions (AUDIT-C). An effective screening test for nificance (HADS-A score > 11) (OR 0.85, 95% CI 0.83–0.87, problem drinking. Archives of Internal Medicine 158, 1789–1795. P < 0.001). A 1 unit increase in the score was also associated with Cousins, R., MacKay, C.J., Clarke, S.D., Kelly, C., Kelly, P.J., McCaig, R.H., 2004. ‘Management standards’ and work related stress in the UK: practical a 22% reduction in the odds of reporting depressive symptoms of development. Work and Stress 18, 113–136. possible or probable clinical significance (HADS-D score > 8) (OR Paykel, E.S., Myers, J.K., Lindenthal, J.J., Tanner, J., 1974. Suicidal feelings in the 0.78, 95% CI 0.76–0.81, P < 0.001) and a 25% reduction in the odds general population: a prevalence study. British Journal of Psychiatry 124, 460– 469. of reporting depressive symptoms of probable clinical significance Stewart-Brown, S., Tennant, A., Tennant, R., Platt, S., Parkinson, J., 2009. Internal (HADS-D score > 11) (OR 0.75, 95% CI 0.71–0.79, P < 0.001). construct validity of the Warwick–Edinburgh Mental Well-being Scale The associations of the WEMWBS score with other standardised (WEMWBS): a Rasch analysis using data from the Scottish Health Education measures in the current study support the validity of the scale as Population Survey. Health and Quality of Life Outcomes 7, 15. Tennant, R., Hiller, L., Fishwick, R., Platt, S., Joseph, S., Weich, S., Parkinson, J., Secker, an overall indicator of population mental health and well-being J., Stewart-Brown, S., 2007. The Warwick–Edinburgh Mental Well-being Scale for veterinary surgeons. The results are complementary to previous (WEMWBS): development and UK validation. Health and Quality of Life tests of the criterion validity of the WEMWBS, which are limited to Outcomes 5, 63. Zigmund, A.S., Snaith, R.P., 1983. The Hospital Anxiety and Depression Scale. Acta general population samples and examine correlations with scales Psychiatrica Scandinavica 67, 361–370. that assess dimensions of mental health and well-being different 269

Glossary

271

Ability The level of the underlying latent variable in a person, inferred from their item responses.

Caseness A clinically significant level of symptoms which exceeds a defined threshold and categorises an individual as possibly having a disorder. It is not a diagnosis.

Class interval RUMM2030 software divides the sample into a specified number of groups (or ‘class intervals’) of similar size, representing different levels of ability (estimated from the summed total score for all items) across the latent variable.

Classical test theory An observed test score comprises two hypothetical components: a true score and a random error component. It is classical in the sense that it is traditional.

Construct A single latent variable assumed to be underlying a set of items.

Construct validity The degree to which the scores of the rating scale are consistent with a priori hypotheses (e.g. with regard to internal relationships, relationships to scores on other instruments, or differences between relevant groups) based on the assumption that the rating scale validly measures the construct to be measured. See external construct validity and internal construct validity.

Content validity The degree to which the content (i.e. items) of the measurement instrument adequately reflects the construct to be measured.

Convergent construct validity The extent to which scores on an instrument are associated with other measures of the same or theoretically similar constructs. See external construct validity.

Differential item functioning The probability of affirming an item’s response categories is influenced by factors other than the level of the latent variable being measured (e.g. age, gender).

Difficulty The magnitude of the latent variable represented by an item.

Discriminant construct validity The extent to which scores are not associated with theoretically unrelated constructs or variables. See external construct validity.

272

External construct validity The association between the rating scale and external criteria such as other measures of similar or different dimensions of health. See convergent construct validity and discriminant construct validity.

Face validity The extent to which the variable which an instrument purports to measure is evident from its item content.

Fit statistics Indices which indicate the extent to which response data conform to Rasch model expectations.

Fundamental measurement The form of measurement used in the physical sciences in which units which are ordered and can be legitimately summed (concatenating units).

Instrument A means to capture data, e.g. a questionnaire, plus the information and documentation that supports its use (i.e. methods and instructions for administration or responding, scoring and interpretation of results).

Interval scale A measurement scale in which the value of the unit of measurement is maintained throughout the scale so that differences have equal values, regardless of location.

Invariance The maintenance of the identity of a variable from one occasion to the next, e.g. item estimates remain stable across population samples.

Item An individual question, statement, or task that is evaluated by the respondent to address a particular concept.

Latent variable An attribute (or construct) of interest that cannot be directly observed.

Logit The unit of measurement that results when the Rasch model is used to transform raw scores obtained from ordinal data to log odds ratios on an interval scale that is common to item difficulty and person ability. The value of 0.0 logits is routinely allocated to the mean of the item difficulty estimates.

Mental well-being A construct comprising both optimal psychological functioning and the experience of happiness and life satisfaction.

Modern test theory Includes item response theory and Rasch measurement theory. Differs from classical test theory in that the focus is the relationship 273

between a person’s measurement and his or her probability of responding to an item, rather than the relationship between a person’s measurement and his or her observed scale total score.

Multidimensional Measuring several latent variables simultaneously.

Person separation index An estimate of the spread or separation of persons on the measured variable. It statistic indicates the potential of a scale to discriminate between subgroups of persons with different levels of ability.

Polytomous The items in a rating scale have greater than two response options. By contrast, dichotomous items have two response options only.

Proportional mortality ratio A ratio of how more or less likely a death in a particular population group is to be from a particular cause (e.g. suicide) as opposed to other causes, than a death of someone of the same age and gender in the general population as a whole.

Reliability The degree to which a measurement instrument is free from measurement error. It is an estimate of the extent to which scores for people whose condition of interest has not changed are the same for repeated measurements, e.g. using different sets of items from the same instrument (internal consistency); or over time over time (test-retest).

Residual A summation of individual item or person deviations from Rasch model expectations, which are then standardised to form a z-score (observed minus expected = residual).

Responsiveness The ability of an instrument to detect change over time in the construct to be measured.

Targeting The extent to which the spectrum of health measured by a health rating scale (the range of item difficulty) matches the distribution of health in the study sample (the range of person ability).

Threshold The point between two adjacent response categories where there is an equal probability of scoring either category.

Unidimensional Measuring a single latent variable.

274

Validity The degree to which the instrument measures the construct(s) it is purported to measure. See Construct validity, Content validity, Face validity. 275

References

277

AGERBO, E., GUNNELL, D., BONDE, J.P., MORTENSEN, P.B. & NORDENTOFT, M. (2007) Suicide and occupation: the impact of socio-economic, demographic, and psychiatric differences. Psychological Medicine 37, 1131-1140

AGERBO, E., MORTENSEN, P.B., ERIKSSON, T., QIN, P. & WESTERGAARD- NIELSEN, N. (2001) Risk of suicide in relation to income level in people admitted to hospital with mental illness: nested case-control study. British Medical Journal 322, 334-335

ALEXANDER, R.E. (2001) Stress-related suicide by dentists and other health care workers: fact or folklore? Journal of the American Dental Association 132, 786-794

AMASS, S.F., DAVIS, K.S., SALISBURY, S.K. & WEISMAN, J.L. (2011) Impact of gender and race-ethnicity on reasons for pursuing a career in veterinary medicine and career aspirations. Journal of the American Veterinary Medical Association 238, 1435-1440

ANDERSON, J., HUPPERT, F. & ROSE, G. (1993) Normality, deviance and minor psychiatric morbidity in the community. A population-based approach to General Health Questionnaire data in the Health and Lifestyle Survey. Psychological Medicine 23, 475-485

ANDERSSON, L., ALLEBECK, P., GUSTAFSSON, J-E. & GUNNELL, D. (2008) Association of IQ scores and school achievement with suicide in a 40-year follow- up of a Swedish cohort. Acta Psychiatrica Scandinavica 118, 99-105

ANDRICH, D. (2004) Controversy and the Rasch model: a characteristic of incompatible paradigms? Medical Care 42, 1-7

ANDRICH, D. (2011) Rating scales and Rasch measurement. Expert Review of Pharmacoeconomics and Outcomes Research 11, 571-585

ANDRICH, D., SHERIDAN, B.S. & LUO, G. (2010) RUMM2030: a Windows program for the analysis of data according to Rasch unidimensional models for measurement. RUMM Laboratory, Perth, Australia

ANDRIESSEN, K. (2006) Do we need to be cautious in evaluating suicide statistics? [Letter] European Journal of Public Health 16, 445-447

ANON (2002) Survey details stress factors that influence Australian vets. Australian Veterinary Journal 80, 522-524

ANON (2010) Gold IIP status for Norfolk practice [News and reports]. Veterinary Record 167, 673

ARBUCKLE, J.L. (2008) Amos™ 17.0 User’s Guide. Chicago, IL, SPSS Inc.

ARYADOUST, S.V. (2009) The impact of Rasch item difficulty on confirmatory factor analysis. Rasch Measurement Transactions 23, 1207. Available at: http://www.rasch.org/rmt/rmt232c.htm Accessed: 04 October 2011

AUSTIN, E.J., EVANS, P., GOLDWATER, R. & POTTER, V. (2005) A preliminary study of emotional intelligence, empathy and exam performance in first year medical students. Personality and Individual Differences 39, 1395-1405

278

BACA-GARCIA, E., PARRA, C.P., PEREZ-RODRIGUEZ, M.M., DIAZ SASTRE, C., REYES TORRES, R., SAIZ-RUIZ, J. & DE LEON, J. (2007) Psychosocial stressors may be strongly associated with suicide attempts. Stress and Health 23, 191-198

BALDWIN, D.S. & RUDGE, S.E. (1995) Depression and suicide in doctors and medical students. In Health Risks to the Health Care Professional. Ed P. Litchfield. Royal College of Physicians. pp 77-90

BARTRAM, D. & GARDNER, D. (2008) Coping with stress. In Practice 30, 228-231

BARTRAM, D. (2010a) Revealing the parts previous surveys did not reach. RCVS News, November 2010. London, RCVS. pp 13-14. Available at: http://www.rcvs. org.uk/publications/rcvs-news-november-2010/. Accessed: 17 August 2011

BARTRAM, D.J. & BALDWIN, D.S. (2010) Veterinary surgeons and suicide: a structured review of possible influences on increased risk. Veterinary Record 166, 388-397

BARTRAM, D.J. (2009a) A cross-sectional study of mental heath and well-being and their associations in the UK veterinary profession. Thesis submitted in accordance with the requirements of the Royal College of Veterinary Surgeons for the Diploma of Fellowship, May 2009

BARTRAM, D.J. (2009b) The mental health of the UK veterinary profession: how do equine practitioners fare? [Invited oral presentation] Handbook of Presentations, 48th British Equine Veterinary Association Congress, 9-12 September, 2009. Birmingham, UK. pp 235-236

BARTRAM, D.J. (2009c) The mental health of the UK veterinary profession – how do farm animal practitioners fare? [Invited oral presentation] British Cattle Veterinary Association Congress, 26-28 November, 2009. Southport, UK. Cattle Practice [Journal of British Cattle Veterinary Association] 17 (2), 164-167

BARTRAM, D.J. (2010) The mental health of the UK veterinary profession: how do government vets fare? [Invited oral presentation] Abstracts. Veterinary Laboratories Agency and GVS/AGV National Conference 2010, ‘New Horizons - Working Together’, 22-24 September, 2010. University of Warwick, UK. p 9

BARTRAM, D.J., SINCLAIR, J.M.A. & BALDWIN, D.S. (2009a) Alcohol consumption among veterinary surgeons in the UK. Occupational Medicine 59, 323-326

BARTRAM, D.J., SINCLAIR, J.M.A. & BALDWIN, D.S. (2010) Interventions with potential to improve the mental health and well-being of UK veterinary surgeons. Veterinary Record 166, 518-523

BARTRAM, D.J., YADEGARFAR, G. & BALDWIN, D.S. (2009b) A cross-sectional study of mental heath and well-being and their associations in the UK veterinary profession. Social Psychiatry and Psychiatric Epidemiology 44, 1075-1085

BARTRAM, D.J., YADEGARFAR, G. & BALDWIN, D.S. (2009c) Psychosocial working conditions and work-related stressors among UK veterinary surgeons. Occupational Medicine 59, 334-341 279

BATCHELOR, C.E.M. & MCKEEGAN, D.E.F. (2011) Survey of the frequency and perceived stressfulness of ethical dilemmas encountered in UK veterinary practice. Veterinary Record 170, 19

BATCHO, C.S., DUREZ, P. & THONNARD, J-L. (2011) Responsiveness of the ABILHAND questionnaire in measuring changes in rheumatoid arthritis patients. Arthritis Care & Research 63, 135-141

BENNEWITH, O., NOWERS, M. & GUNNELL, D. (2007) Effect of barriers on the Clifton suspension bridge, England, on local patterns of suicide: implications for prevention. British Journal of Psychiatry 190, 266-267

BENTLER, P.M. & BONETT, D.G. (1980) Significance tests and goodness of fit in the analysis of covariance structures. Psychological Bulletin 88, 588-606

BERTOLOTE, J.M. (2004) Suicide prevention: at what level does it work? World Psychiatry 3, 147-151

BEVAN, A., HOUDMONT, J. & MENEAR, N. (2010) The Management Standards Indicator Tool and the estimation of risk. Occupational Medicine 60, 525-531

BHUGRA, D. (2006) Suicide and gender: cultural factors. Harvard Health Policy Review 7, 166-180

BIDDLE, L., BROCK, A., BROOKES, S.T. & GUNNELL, D. (2008) Suicide rates in young men in England and Wales in the 21st century: time trend study. British Medical Journal 336, 539-542

BJELLAND, I., DAHL, A.A., HAUG, T.T. & NECKELMANN, D. (2002) The validity of the Hospital Anxiety and Depression Scale: an updated literature review. Journal of Psychosomatic Research 52, 69-77

BLAIR, A. & HAYES, H.M. (1980) Cancer and other causes of death among US veterinarians, 1966-1977. International Journal of Cancer 25, 181-185

BLAIR, A. & HAYES, H.M. (1982) Mortality patterns among US veterinarians, 1947- 1977: an expanded study. International Journal of Epidemiology 11, 391-397

BLAND, J.M. & ALTMAN. D.G. (1995) Multiple significance tests: the Bonferroni method. British Medical Journal 310, 170

BOHLIG, M., FISHER, W.P., MASTERS, G.N. & BOND, T. (1998) Content validity and misfitting items. Rasch Measurement Transactions 12, 607. Available at: http://www.rasch.org/rmt/rmt121f.htm. Accessed: 13 October 2011

BOND, F.W. (2004) Getting the balance right: the need for a comprehensive approach to occupational health. Work & Stress 18, 146-148

BOND, T.G. (1994) Too many factors? Rasch Measurement Transactions 8, 347. Available at: http://www.rasch.org/rmt/rmt81p.htm Accessed: 04 October 2011

BONDE, J.P.E. (2008) Psychosocial factors at work and risk of depression: a systematic review of the epidemiological evidence. Occupational and Environmental Medicine 65, 438-445

280

BORGES, N.J. & SAVICKAS, M.L. (2002) Personality and medical specialty choice: a literature review and integration. Journal of Career Assessment 10, 362-380

BOWLING, A. (2005) Mode of questionnaire administration can have serious effects on data quality. Journal of Public Health 27, 281-291

BRADY, J. (2006) The association between alcohol misuse and suicidal behaviour. Alcohol & Alcoholism 41, 473-478

BRAUNHOLTZ, S., DAVIDSON, S., MYANT, K. & O’CONNOR, R. (2007) Well? What Do You Think? (2006): The Third National Scottish Survey of Public Attitudes to Mental Health, Mental Wellbeing and Mental Health Problems. Edinburgh, The Scottish Executive. http://www.scotland.gov.uk/Resource/Doc/197512/0052833.pdf. Accessed: 13 October 2011

BRAY, I. & GUNNELL, D. (2006) Suicide rates, life satisfaction and happiness as markers for population mental health. Social Psychiatry and Psychiatric Epidemiology 41, 333-337

BRENT, D.A. & MANN, J.J. (2005) Family genetic studies, suicide, and suicidal behavior. American Journal of Medical Genetics 133C, 13-24

BREZO, J., PARIS, J. & TURECKI, G. (2006) Personality traits as correlates of suicidal ideation, suicide attempts, and suicide completions: a systematic review. Acta Psychiatrica Scandinavica 113, 180-206

BRITISH VETERINARY ASSOCIATION & ASSOCIATION OF VETERINARY STUDENTS (2008) Survey results 2008. London, British Veterinary Association. http://www.bva.co.uk/public/documents/2008_AVS_Survey_Results.pdf. Accessed: 13 October 2011

BROCK, A. & GRIFFITHS, C. (2003) Trends in the mortality of young adults aged 15-44 in England and Wales, 1961 to 2001. Health Statistics Quarterly 19, 22-31

BROCK, A., BAKER, A., GRIFFITHS, C., JACKSON, G., FEGAN, G. & MARSHALL, D. (2006) Suicide trends and geographical variations in the United Kingdom, 1991- 2004. Health Statistics Quarterly 31, 6-22

BROOKS, S.K., CHALDER, T. & GERADA, C. (2011) Doctors vulnerable to psychological distress and addictions: treatment from the Practitioner Health Programme. Journal of Mental Health 20, 157-164

BROWN, S.L. (1994) Factor structure of a brief version of the Ways of Coping (WOC) questionnaire: a study with veterinary medicine students. Measurement and Evaluation in Counseling and Development 27, 308-315

BUSH, K., KIVLAHAN, D.R., MCDONNELL, M.B., FIHN, S.D. & BRADLEY, K.A. (1998) The AUDIT alcohol consumption questions (AUDIT-C). An effective screening test for problem drinking. Archives of Internal Medicine 158, 1789-1795

CALLANAN, V.J. & DAVIS, M.S. (2011) Gender differences in suicide methods. Social Psychiatry and Psychiatric Epidemiology. doi: 10.1007/s00127-011-0393-5 Published online: 21 May 2011 281

CAMPBELL, D.T. & FISKE, D.W. (1959) Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin 56, 81-105

CANO, S.J. & HOBART, J.C. (2011) The problem with health measurement. Patient Preference and Adherence 5, 279-290

CANO, S.J., BARRETT, L.E., ZAJICEK, J.P. & HOBART, J.C. (2011) Beyond the reach of traditional analyses: using Rasch to evaluate the DASH in people with multiple sclerosis. Multiple Sclerosis Journal 17, 214-222

CANO, S.J., BROWNE, J.P., LAMPING, D.L., ROBERTS, A.H.N., MCGROUTHER, D.A. & BLACK, N.A. (2004) The patient outcomes of surgery – hand/arm (POS- Hand/Arm): a new patient-based outcome measure. Journal of Hand Surgery 29, 477-485

CAPNER, C.A., LASCELLES, B.D.X. & WATERMAN-PEARSON, A.E. (1999) Current British veterinary attitudes to perioperative analgesia for dogs. Veterinary Record 145, 95-99

CATTELL, R.B. (1966) The scree test for the number of factors. Multivariate Behavioral Research 1, 245-276

CAVANAGH, J.T.O., CARSON, A.J., SHARPE, M. & LAWRIE, S.M. (2003) Psychological autopsy studies of suicide: a systematic review. Psychological Medicine 33, 395-405

CENTER, C., DAVIS, M., DETRE, T., FORD, D.E., HANSBROUGH, W., HENDIN, H., LASZLO, J., LITTS, D.A., MANN, J., MANSKY, P.A., MICHELS, R., MILES, S.H., PROUJANSKY, R., REYNOLDS, C.F. & SILVERMAN, M.M. (2003) Confronting depression and suicide in physicians: a consensus statement. Journal of the American Medical Association 289, 3161-3166

CHARLTON, J. (1995) Trends and patterns in suicide in England and Wales. International Journal of Epidemiology 24 (Suppl. 1), S45-S52

CHARLTON, J., KELLY, S., DUNNELL, K., EVANS, B. & JENKINS, R. (1993) Suicide deaths in England and Wales: trends in factors associated with suicide deaths. Population Trends 71, 34-42

CHEUNG, W.Y., GARRATT, A.M., RUSSELL, I.T. & WILLIAMS, J.G. (2000): The UK IBDQ – a British version of the inflammatory bowel disease questionnaire: development and validation. Journal of Clinical Epidemiology 53, 297-306

CHEUNG, Y-B., LIM, C., GOH, C., THUMBOO, J. & WEE, J. (2005) Order effects: a randomised study of three major cancer-specific quality of life instruments. Health and Quality of Life Outcomes 3, 37

CIEZA, A., HILFIKER, R., BOONEN, A., CHATTERJI, S., KOSTANJSEK, N., ÜSTÜN, B.T. & STUCKI, G. (2009) Items from patient-oriented instruments can be integrated into interval scales to operationalize categories of the International Classification of Functioning, Disability and Health. Journal of Clinical Epidemiology 62, 912-921

CLACK, G.B., ALLEN, J., COOPER, D. & HEAD, J.O. (2004) Personality differences between doctors and their patients: implications for the teaching of communication skills. Medical Education 38, 177-186

282

CLARK, L.A. & WATSON, D. (1995) Constructing validity: basic issues in objective scale development. Psychological Assessment 7, 309-319

CLARK, M.A. & JONES, J.W. (1979) Suicide by intravenous injection of a veterinary euthanasia agent: report of a case and toxicologic studies. Journal of Forensic Sciences 24, 762-767

CLARKE, A., FRIEDE, T., PUTZ, R., ASHDOWN, J., MARTIN, S., BLAKE, A., ADI, Y., PARKINSON, J., FLYNN, P., PLATT, S. & STEWART-BROWN, S. (2011) Warwick-Edinburgh Mental Well-being Scale (WEMWBS): validated for teenage school students in England and Scotland. A mixed methods assessment. BMC Public Health 11, 487

CLERY, E., MCLEAN, S. & PHILLIPS, M. (2007) Quickening death: the euthanasia debate. In British Social Attitudes: The 23rd Report – Perspectives on a changing society. Eds A. Park, J. Curtice, K. Thomson, M. Phillips, M. Johnson. Sage. pp 35-53

COHEN, S.P. (2007) Compassion fatigue and the veterinary health team. Veterinary Clinics of North America, Small Animal Practice 37, 123-134

COLE, D.A. (1987) Utility of confirmatory factor analysis in test validation research. Journal of Consulting and Clinical Psychology 55, 584-594

CONNOLLY, D.J. (2003) Stress in the veterinary profession: the nature, implications and consequences for veterinarians in the West of Ireland [MA Thesis]. National University of Ireland, Galway

CONNOLLY, N. (2002) An exploratory study of stress in the veterinary profession in the UK and Ireland. [MSc Thesis] University of Manchester Institute of Science and Technology, UK

CORDELL, W.H., CURRY, S.C., FURBEE, R.B. & MITCHELL-FLYNN, D.L. (1986) Veterinary drugs as suicide agents. Annals of Emergency Medicine 15, 939-943

CORTINA, J.M. (1993) What is coefficient alpha? An examination of theory and applications. Journal of Applied Psychology 78, 98-104

COUSINS, R., MACKAY, C.J., CLARKE, S.D., KELLY, C., KELLY, P.J. & MCCAIG, R.H. (2004) ‘Management standards’ and work related stress in the UK: practical development. Work & Stress 18, 113-136

CRONBACH, L.J. (1951) Coefficient alpha and the internal structure of tests. Psychometrika 16, 297-334

D’SOUZA, E., BARRACLOUGH, R., FISHWICK, D. & CURRAN, A. (2009) Management of occupational health risks in small-animal veterinary practices. Occupational Medicine 59, 316-322

DAIGLE, M.S. (2005) Suicide prevention through means restriction: assessing the risk of substitution. A critical review and synthesis. Accident Analysis & Prevention 37, 625-632

DE GRAAF, G. (2005) Veterinarians’ discourses on animals and clients. Journal of Agriculture and Environmental Ethics 18, 557-578 283

DE MENESES-GAYA, C., ZUARDI, A.W., LOUREIRO, S.R. & CRIPPA, J.A.S. (2009) Alcohol Use Disorders Identification Test (AUDIT): an updated systematic review of psychometric properties. Psychology and Neuroscience (Online) 2, 83- 97

DE MORTON, N.A. & NOLAN, J.S. (2011) Unidimensionality of the Elderly Mobility Scale in older acute medical patients: different methods, different answers. Journal of Clinical Epidemiology 64, 667-674

DEVELLIS, R.F. (2006) Classical test theory. Medical Care 44, S50-S59

DIESENHAMMER, E. A., ING, C. M., STRAUSS, R., KEMMLER, G., HINTERHUBER, H. & WEISS, E. M. (2009) The duration of the suicidal process: how much time is left for intervention between consideration and accomplishment of a suicide attempt? Journal of Clinical Psychiatry 70, 19-24

DINOS, S., STEVENS, S., SERFATY, M., WEICH, S. & KING, M. (2004) Stigma: the feelings and experiences of 46 people with mental illness: qualitative study. British Journal of Psychiatry 184, 176-181

DOMINO, G., MOORE, D., WESTLAKE, L. & GIBSON, L. (1982) Attitudes towards suicide: a factor analytic approach. Journal of Clinical Psychology 38, 257-262

DONALD, M., DOWER, J., CORREA-VELEZ, I. & JONES, M. (2005) Risk and protective factors for medically serious suicide attempts: a comparison of hospital- based with population-based samples of young adults. Australian and New Zealand Journal of Psychiatry 40, 87-96

DRUSS, B. & PINCUS, H. (2000) Suicidal ideation and suicide attempts in general medical illnesses. Archives of Internal Medicine 160, 1522-1526

DUBERSTEIN, P.R., CONWELL, Y., CONNER, K.R., EBERLY, S., EVINGER, J.S. & CAINE, E.D. (2004) Poor social integration and suicide: fact or artifact? A case- control study. Psychological Medicine 34, 1331-1337

DUBERSTEIN, P.R., CONWELL, Y., SEIDLITZ, L., DENNING, D.G., COX, C. & CAINE, E.D. (2000) Personality traits and suicidal behaviour and ideation in depressed inpatients 50 years of age and older. Journals of Gerontology, Series B: Psychological Sciences and Social Sciences 55, 18-26

DUDEK, F.J. (1979) The continuing misinterpretation of the standard error of measurement. Psychological Bulletin 86, 335-337

DURKHEIM, É. ([1897] 2006) On Suicide. Penguin Classics

EDWARDS, J.A., WEBSTER, S., VAN LAAR, D. & EASTON, S. (2008) Psychometric analysis of the UK Health and Safety Executive’s Management Standards work-related stress Indicator Tool. Work & Stress 22, 96-107

EDWARDS, P., ROBERTS, I., CLARKE, M., DIGUISEPPI, C., PRATAP, S., WENTZ, R., KWAN, I. & COOPER, R. (2007) Methods to increase response rates to postal questionnaires. Cochrane Database of Systematic Reviews, Issue 2

ELEJALDE, J.I., LOUIS, C.J., ELCUAZ, R. & PINILLOS, M.A. (2003) Drug abuse with inhalated xylazine. European Journal of Emergency Medicine 10, 252-253

284

ELKINS, A.D. & KEARNEY, M. (1992) Professional burnout among female veterinarians in the United States. Journal of the American Veterinary Medical Association 200, 604-608

ELLIOTT, D.M. & GUY, J.D. (1993) Mental health professionals versus non-mental- health professionals: childhood trauma and adult functioning. Professional Psychology: Research and Practice 24, 83-90

ELLIOTT, S.P. & HALE, K.A. (1999) Analysis of etorphine in postmortem samples by HPLC with UV diode-array detection. Forensic Science International 101, 9-16

ESKIN, M., VORACEK, M., STIEGER, S. & ALTINYAZAR, V. (2011) A cross- cultural investigation of suicidal behavior and attitudes in Austrian and Turkish medical students. Social Psychiatry and Psychiatric Epidemiology 46, 813-823

ETZERSDORFER, E., VIJAYAKUMAR, L., SCHÖNY, W., GRAUSGRUBER, A. & SONNECK, G. (1998) Attitudes towards suicide among medical students – comparison between Madras (India) and Vienna (Austria). Social Psychiatry and Psychiatric Epidemiology 33, 104-110

EVANS, O. & STEPTOE, A. (2002) The contribution of gender-role orientation, work factors and home stressors to psychological well-being and sickness absence in male- and female-dominated occupational groups. Social Science & Medicine 54, 481-492

FAIRNIE, H.M. (2005) Occupational injury, disease and stress in the veterinary profession [PhD Thesis]. Curtin University of Technology, Australia. Available at: http://adt.curtin.edu.au/theses/available/adt-WCU20070528.140327/. Accessed: 14 October 2011

FARMER, R. & ROHDE, J. (1980) Effect of availability and acceptability of lethal instruments on suicide mortality: an analysis of some international data. Acta Psychiatrica Scandinavica 62, 436-445

FAZEL, S., BENNING, R. & DANESH, J. (2005) Suicides in male prisoners in England and Wales, 1978-2003. Lancet 366, 1301-1302

FIRTH-COZENS, J. (1990) Sources of stress in women junior house officers. British Medical Journal 301, 89-91

FIRTH-COZENS, J. (2001) Interventions to improve physicians’ well-being and patient care. Social Science & Medicine 52, 215-222

FISHBAIN, D.A. (1986) Veterinarians with psychiatric impairment – a comparison with impaired physicians. Journal of the National Medical Association 78, 133-137

FISHER, W. (1992) Reliability statistics. Rasch Measurement Transactions 6, 238. Available at: http://www.rasch.org/rmt/rmt63i.htm. Accessed: 29 September 2011

FLORENTINE, J.B. & CRANE, C. (2010) Suicide prevention by limiting access to methods: a review of theory and practice. Social Science & Medicine 70, 1626- 1632

FLOYD, F.J. & WIDAMAN, K.F. (1995) Factor analysis in the development and refinement of clinical assessment instruments. Psychological Assessment 7, 286- 299 285

FOGLE, B. & ABRAHAMSON, D. (1990) Pet loss: a survey of the attitudes and feelings of practicing veterinarians. Anthrozoös 3, 143-150

FOOD AND DRUG ADMINISTRATION (2009) Patient reported outcome measures: use in medical product development to support labelling claims. Available at: www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guida nces/UCM193282.pdf. Accessed 19 September 2011

FORJAZ, M.J., RODRIGUEZ-BLÁZQUEZ, C. & MARTINEZ-MARTIN, P. (2009) Rasch analysis of the Hospital Anxiety and Depression Scale in Parkinson’s disease. Movement Disorders 24, 526-532

FRANK, E., BOSWELL, L., DICKSTEIN, L.J. & CHAPMAN, D.P. (2001) Characteristics of female psychiatrists. American Journal of Psychiatry 158, 205- 212

FRIDNER, A., BELKIC, K., MARINI, M., MINUCCI, D., PAVAN, L. & SCHENCK- GUSTAFSSON, K. (2009) Survey on recent suicidal ideation among female university hospital physicians in Sweden and Italy (The HOUPE Study): cross- sectional associations with work stressors. Gender Medicine 6, 314-328

FRITSCHI, L., MORRISON, D., SHIRANGI, A. & DAY, L. (2009) Psychological well- being of Australian veterinarians. Australian Veterinary Journal 87, 76-81

GAGNÉ, P., MOAMAI, J. & BOURGET, D. (2011) Psychopathology and suicide among Quebec physicians: a nested case control study. Depression Research and Treatment Article ID 936327, doi:10.1155/2011/936327

GARDNER, D.H. & HINI, D. (2006) Work-related stress in the veterinary profession in New Zealand. New Zealand Veterinary Journal 54, 119-124

GARNAAT, S.L. & NORTON, P.J. (2010) Factor structure and measurement invariance of the Yale-Brown Obsessive Compulsive Scale across four racial/ethnic groups. Journal of Anxiety Disorders 24, 723-728

GELBERG, S. & GELBERG, H. (2005) Stress management interventions for veterinary students. Journal of Veterinary Medical Education 32, 173-181

GENERAL REGISTER OFFICE FOR SCOTLAND (2007) Scotland’s population 2006. The Registrar General’s Annual Review of Demographic Trends. 152nd edition. Edinburgh, General Register Office for Scotland. p 32. Available at: http://www.gro-scotland.gov.uk/files1/stats/annual-report2006/annual-report- 2006.pdf. Accessed: 14 October 2011

GERRITY, M.S. (2001) Interventions to improve physicians’ well-being and patient care: a commentary. Social Science & Medicine 52, 223-225

GEURTS, S.A.E., TARIS, T.W., KOMPIER, M.A.J., DIKKERS, J.S.E., VAN HOOFF, M.L.M. & KINNUNEN, U.M. (2005) Work-home interaction from a work psychological perspective: development and validation of a new questionnaire, the SWING. Work & Stress 19, 319-339

GIBB, B.E., ANDOVER, M.S. & BEACH, S.R.H. (2006) Suicidal ideation and attitudes toward suicide. Suicide and Life-Threatening Behavior 36, 12-18

286

GIBBONS, C.J., MILLS, R.J., THORNTON, E.W., EALING, J., MITCHELL, J.D., SHAW, P.J., TALBOT, K., TENNANT, A. & YOUNG, C.A. (2011) Rasch analysis of the Hospital Anxiety and Depression Scale (HADS) for use in motor neurone disease. Health and Quality of Life Outcomes 9, 82

GIBBS, J.P. (2000) Status integration and suicide: occupational, marital, or both? Social Forces 79, 363-384

GILLING, M.G. & PARKINSON, T.J. (2009) The transition from veterinary student to practitioner: a ‘make or break’ period. Journal of Veterinary Medical Education 26, 209-215

GOLDNEY, R.D. (2005) Risk factors for suicidal behaviour. In Prevention and Treatment of Suicidal Behaviour: From Science to Practice. Ed K. Hawton. Oxford University Press. pp 161-182

GOSSOP, M., STEPHENS, S., STEWART, D., MARSHALL, J., BEARN, J. & STRANG, J. (2001) Health care professionals referred for treatment of alcohol and drug problems. Alcohol & Alcoholism 36, 160-164

GOULD, M.S., FISHER, P., PARIDES, M., FLORY, M. & SHAFFER, D. (1996) Psychosocial risk factors of child and adolescent completed suicide. Archives of General Psychiatry 53, 1155-1162

GRANGER, C. (2008) Rasch analysis is important to understand and use for measurement. Rasch Measurement Transactions 21, 1122-1123. Available at: http://www.rasch.org/rmt/rmt213d.htm. Accessed: 01 October 2011

GRAY, C. A., BLAXTER, A. C., JOHNSTON, P. A., LATHAM, C. E., MAY, S., PHILLIPS, C. A., TURNBULL, N. & YAMAGISHI, B. (2006) Communication education in veterinary education in the United Kingdom and Ireland: the NUVACS project coupled to progressive individual school endeavors. Journal of Veterinary Medical Education 33, 85-92

GREGORICH, S.E. (2006) Do self-report instruments allow meaningful comparisons across diverse population groups? Testing measurement invariance using the confirmatory factor analysis framework. Medical Care 44, S78-S94

GRIFFITHS, C., LADVA, G., BROCK, A. & BAKER, A. (2008) Trends in suicide by marital status in England and Wales, 1982-2005. Health Statistics Quarterly 37, 8- 14

GRIFFITHS, C., ROONEY, C. & BROCK, A. (2005) Leading causes of death in England and Wales – how should we group causes? Health Statistics Quarterly 28, 6-17

GROSS, E.B. (1997) Gender differences in physician stress: why the discrepant findings? Women and Health 26, 1-14

GUAL, A., SEGURA, L., CONTEL, M., HEATHER, N. & COLOM, J. (2002) AUDIT-3 and AUDIT-4: effectiveness of two short forms of the alcohol use disorders identification test. Alcohol & Alcoholism 37, 591-596

GUNNELL, D. & LEWIS, G. (2005) Studying suicide from the life course perspective: implications for prevention. British Journal of Psychiatry 187, 206-208 287

GUNNELL, D., HAWTON, K. & KAPUR, N. (2011) Coroners’ verdicts and suicide statistics in England and Wales. British Medical Journal 343, d6030

GUNNELL, D., MAGNUSSON, P.K.E. & RASMUSSEN, F. (2005) Low intelligence test scores in 18-year-old men and risk of suicide: cohort study. British Medical Journal 330, 167 doi: 10.1136/bmj.38310.473565.8F

GYAGENDA, I.S. & ENGELHARD, G. (2009) Using classical and modern measurement theories to explore rater, domain, and gender influences on student writing ability. Journal of Applied Measurement 10, 225-246

HAFEN, M., Jr, REISBIG, A.M.J., WHITE, M.B. & RUSH, B.R. (2006) Predictors of depression and anxiety in first-year veterinary students: a preliminary report. Journal of Veterinary Medical Education 33, 432-440

HAFEN, M., Jr, REISBIG, A.M.J., WHITE, M.B. & RUSH, B.R. (2008) The first-year veterinary student and mental health: the role of common stressors. Journal of Veterinary Medical Education 35, 102-109

HAGQUIST, C., BRUCE, M. & GUSTAVSSON, P. (2009) Using the Rasch model in nursing research: an introduction and illustrative example. International Journal of Nursing Studies 46, 380-393

HALLIWELL, R.E.W. & HOSKIN, B.D. (2005) Reducing the suicide rate among veterinary surgeons: how the profession can help. Veterinary Record 157, 397-398

HANSEZ, I., SCHINS, F. & ROLLIN, F. (2008) Occupational stress, work-home interference and burnout among Belgian veterinary practitioners. Irish Veterinary Journal 61, 233-241

HARDOUIN, J-B., CONROY, R. & SÉBILLE, V. (2011) Imputation by the mean score should be avoided when validating a patient reported outcomes questionnaire by a Rasch model in presence of informative missing data. BMC Medical Research Methodology 11, 105

HARLING, M., STREHMEL, P., SCHABLON, A. & NIENHAUS, A. (2009) Psychosocial stress, demoralization and the consumption of tobacco, alcohol and medical drugs by veterinarians. Journal of Occupational Medicine and Toxicology 4, 4. doi: 10.11186/1745-6673-4-4

HARMON-JONES, E. & HARMON-JONES, C. (2007) Cognitive dissonance theory after 50 years of development. Zeitschrift für Sozialpsychologie 38, 7-16

HARRIS, E.C. & BARRACLOUGH, B. (1997) Suicide as an outcome for mental disorders: a meta-analysis. British Journal of Psychiatry 170, 205-228

HATCH P.H., WINEFIELD, H.R., CHRISTIE, B.A. & LIEVAART, J.J. (2011) Workplace stress, mental health, and burnout of veterinarians in Australia. Australian Veterinary Journal 89, 460-468

HATCHER, S. (1994) Debt and deliberate self-poisoning. British Journal of Psychiatry 164, 111-114

HAWGOOD, J. & DE LEO, D. (2008) Anxiety disorders and suicidal behaviour: an update. Current Opinion in Psychiatry 21, 51-64

288

HAWTHORNE, G. & ELLIOTT, P. (2005) Imputing cross-sectional missing data: comparison of common techniques. Australian and New Zealand Journal of Psychiatry 39, 583-590

HAWTON, K. & VISLISEL, L. (1999) Suicide in nurses. Suicide and Life-Threatening Behavior 29, 86-95

HAWTON, K. (2007) Restricting access to methods of suicide: rationale and evaluation of this approach to suicide prevention. Crisis 28 (Suppl. 1), 4-9

HAWTON, K., AGERBO, E., SIMKIN, S., PLATT, B. & MELLANBY, R.J. (2011) Risk of suicide in medical and related occupational groups: a national study based on Danish case population-based registers. Journal of Affective Disorders 134, 320- 326

HAWTON, K., BERGEN, H., SIMKIN, S., BROCK, A., GRIFFITHS, C., ROMERI, E., SMITH, K. L., KAPUR, N. & GUNNELL, D. (2009) Effect of withdrawal of co- proxamol on prescribing and deaths from drug poisoning in England and Wales: time series analysis. British Medical Journal 338, b2270

HAWTON, K., CLEMENTS, A., SAKAROVITCH, C., SIMKIN, S. & DEEKS, J.J. (2001) Suicide in doctors: a study of risk according to gender, seniority and specialty in medical practitioners in England and Wales, 1979-1995. Journal of Epidemiology and Community Health 55, 296-300

HAWTON, K., CLEMENTS, A., SIMKIN, S. & MALMBERG, A. (2000) Doctors who kill themselves: a study of the methods used for suicide. Quarterly Journal of Medicine 93, 351-357

HAWTON, K., FAGG, J., SIMKIN, S., HARRISS, L. & MALMBERG, A. (1998) Methods used for suicide by farmers in England and Wales: the contribution of availability and its relevance to prevention. British Journal of Psychiatry 173, 320- 324

HAWTON, K., MALMBERG, A. & SIMKIN, S. (2004a) Suicide in doctors: a psychological autopsy study. Journal of Psychosomatic Research 57, 1-4

HAWTON, K., SIMKIN, S., DEEKS, J., COOPER, J., JOHNSTON, A., WATERS, K., ARUNDEL, M., BERNAL, W., GUNSON, B., HUDSON, M., SURI, D. & SIMPSON, K. (2004b) UK legislation on analgesic packs: before and after study of long term effect on poisonings. British Medical Journal 329, 1076

HAWTON, K., SIMKIN, S., RUE, J., HAW, C., BARBOUR, F., CLEMENTS, A., SAKAROVITCH, C. & DEEKS, J. (2002) Suicide in female nurses in England and Wales. Psychological Medicine 32, 239-250

HAWTON, K., ZAHL, D. & WEATHERALL, R. (2003) Suicide following deliberate self-harm: long-term follow-up of patients who presented to a general hospital. British Journal of Psychiatry 182, 537-542

HEATH, T. (1998) Matching personal attributes to those of the course and the workplace: examples from veterinary science. Higher Education Research & Development 17, 119-124

HEATH, T.J. (2007) Longitudinal study of veterinary students and veterinarians: the first 20 years. Australian Veterinary Journal 85, 281-289 289

HEINEMANN, A.W. & DEUTSCH, A. (2011) Commentary on ‘Past and present issues in Rasch analysis: the FIM revisited’. Journal of Rehabilitation Medicine 43, 958-960

HEM, E., HALDORSEN, T., AASLAND, O.G., TYSSEN, R., VAGLUM, P. & EKEBERG, Ø. (2005) Suicide rates according to education with a particular focus on physicians in Norway 1960-2000. Psychological Medicine 35, 873-880

HENNING, K., EY, S. & SHAW, D. (1998) Perfectionism, the impostor phenomenon and psychological adjustment in medical, dental, nursing and pharmacy students. Medical Education 32, 456-464

HERRMANN, C. (1997) International experiences with the Hospital Anxiety and Depression Scale – a review of validation data and clinical results. Journal of Psychosomatic Research 42, 17-41

HERZOG, H.A., Jr, VORE, T.L. & NEW, J.C., Jr. (1989) Conversations with veterinary students: attitudes, ethics, and animals. Anthrozoös 2, 181-188

HESKETH, B. & SHOUKSMITH, G. (1986) Job and non-job activities, job satisfaction and mental health among veterinarians. Journal of Occupational Behaviour 7, 325-339

HINTIKKA, J., KONTULA, O., SAARINEN, P., TANSKANEN, A., KOSKELA, K. & VIINAMÄKI, H. (1998) Debt and suicidal behaviour in the Finnish general population. Acta Psychiatrica Scandinavica 98, 493-496

HIRSCH, J.K. (2006) A review of the literature on rural suicide: risk and protective factors, incidence, and prevention. Crisis 27, 189-199

HOBART, J. & CANO, S. (2009) Improving the evaluation of therapeutic interventions in multiple sclerosis: the role of new psychometric methods. Health Technology Assessment 13, 12

HOBART, J. (2003) Rating scales for neurologists. Journal of Neurology, Neurosurgery and Psychiatry 74 (Suppl IV), iv22–iv26

HOBART, J., FREEMAN, J., LAMPING, D., FITZPATRICK, R. & THOMPSON, A. (2001a) The SF-36 in multiple sclerosis: why basic assumptions must be tested. Journal of Neurology, Neurosurgery and Psychiatry 71, 363-370

HOBART, J., LAMPING, D., FITZPATRICK, R., RIAZI, A. & THOMPSON, A. (2001b) The multiple sclerosis impact scale (MSIS-29): a new patient-based outcome measure. Brain 124, 962-973

HOBART, J.C., CANO, S.J. & THOMPSON, A.J. (2010) Effect sizes can be misleading: is it time to change the way we measure change? Journal of Neurology, Neurosurgery and Psychiatry 81, 1044-1048

HOBART, J.C., CANO, S.J., ZAJICEK, J.P. & THOMPSON, A.J. (2007) Rating scales as outcome measures for clinical trials in neurology: problems, solutions, and recommendations. Lancet Neurology 6, 1094-1105

HORN, J.L. (1965) A rationale and test for the number of factors in factor analysis. Psychometrika 30, 179-185

290

HUGONNARD, M., LEBLOND, A., KEROACK, S., CADORÉ, J., TRONCY, E. (2004) Attitudes and concerns of French veterinarians towards pain and analgesia in dogs and cats. Veterinary Anaesthesia and Analgesia 31, 154-163

JACCARD, J. & WAN, C.K. (1996) LISREL approaches to interaction effects in multiple regression. Sage

JAMIESON, S. (2004) Likert scales: how to (ab)use them. Medical Education 38, 1217-1218

JENKINS, R., BHUGRA, D., BEBBINGTON, P., BRUGHA, T., FARRELL, M., COID, J., FRYERS, T., WEICH, S., SINGLETON, N. & MELTZER, H. (2008) Debt, income and mental disorder in the general population. Psychological Medicine 38, 1485-1493

JEYARETNAM, J., JONES, H. & PHILLIPS, M. (2000) Disease and injury among veterinarians. Australian Veterinary Journal 78, 625-629

JOE, S., ROMER, D. & JAMIESON, P.E. (2007) Suicide acceptability is related to suicide planning in U.S. adolescents and young adults. Suicide and Life- Threatening Behavior 37, 165-178

JOHNSON, S., COOPER, C., CARTWRIGHT, S., DONALD, I., TAYLOR, P. & MILLET, C. (2005) The experience of work-related stress across occupations. Journal of Managerial Psychology 20, 178-187

JONES-FAIRNIE, H., FERRONI, P., SILBURN, S. & LAWRENCE, D. (2008) Suicide in Australian veterinarians. Australian Veterinary Journal 86, 114-116

JÖRNGÅRDEN, A., WETTERGEN, L. & VON ESSEN, L. (2006) Measuring health- related quality of life in adolescents and young adults: Swedish normative data for the SF-36 and the HADS, and the influence of age, gender, and method of administration. Health and Quality of Life Outcomes 4, 91

KAPLAN, D. (1990) Evaluating and modifying covariance structure models: a review and recommendation. Multivariate Behavioral Research 25, 137-155

KARABATSOS, G. (2001) The Rasch model, additive conjoint measurement, and new models of probabilistic measurement theory. Journal of Applied Measurement 2, 389-423

KARASEK, R. & THEORELL, T. (1990) Healthy Work: Stress, Productivity, and the Reconstruction of Working Life. Basic Books

KARASEK, R., BRISSON, C., KAWAKAMI, N., HOUTMAN, I., BONGERS, P. & AMICK, B. (1998) The Job Content Questionnaire (JCQ): an instrument for internationally comparative assessments of psychosocial job characteristics. Journal of Occupational Health Psychology 3, 322-355

KARASEK, R.A. (1979) Job demands, job decision latitude, and mental strain: implications for job redesign. Administrative Science Quarterly 24, 285-308

KELLOWAY, E.K., TEED, M. & KELLEY, E. (2008) The psychosocial environment: towards an agenda for research. International Journal of Workplace Health Management 1, 50-64 291

KELLY, S. & BUNTING, J. (1998) Trends in suicide in England and Wales, 1982-96. Population Trends 92, 29-41

KELLY, S., CHARLTON, J. & JENKINS, R. (1995) Suicide deaths in England and Wales, 1982-1992: the contribution of occupation and geography. Population Trends 80, 16-25

KERSTEN, P., WHITE, P.J. & TENNANT, A. (2011) The consultation and relational empathy measure: an investigation of its scaling structure. Disability & Rehabilitation. doi:10.3109/09638288.2011.610493. Published online: 08 October 2011

KIEFFER, J.M., VERRIPS, G.H.W. & HOOGSTRATEN J. (2011) Instrument-order effects: using the Oral Health Impact Profile 49 and the Short Form 12. European Journal of Oral Sciences 119, 69-72

KING, E.A. (2001) The Wessex Suicide Audit 1988-1993: A study of 1457 suicides with and without a recent psychiatric contact. International Journal of Psychiatry in Clinical Practice 5, 111-118

KING, E. & FROST, N. (2005) The New Forest Suicide Prevention Initiative (NFSPI). Crisis 26, 25-33

KINLEN, L.J. (1983) Mortality among British veterinary surgeons. British Medical Journal 287, 1017-1019

KINSELLA, M. (2006) Suicide in the veterinary profession: the hidden reality – Part Two. Irish Veterinary Journal 59, 704-706

KINTZ, P., CIRIMELE, V. & LUDES, B. (2002) Blood investigation in a fatality involving the veterinary drug T-61. Journal Analytical Toxicology 26, 529-531

KIRWAN, A.P. (2005) Mankind and other animals [MA Thesis]. Open University, Milton Keynes, UK

KLEPPA, E., SANNE, B. & TELL, G.S. (2008) Working overtime is associated with anxiety and depression: The Hordaland Health Study. Journal of Occupational and Environmental Medicine 50, 658-666

KNOX, K.L., LITTS, D.A., TALCOTT, G.W., FEIG, J.C. & CAINE, E.D. (2003) Risk of suicide and related adverse outcomes after exposure to a suicide prevention programme in the US Air Force: cohort study. British Medical Journal 327, 1376- 1378

KNOX, K.L., PFLANZ, S., TALCOTT, G.W., CAMPISE, R.L., LAVIGNE, J.E., BAJORSKA, A., TU, X. & CAINE, E.D. (2010) The US Air Force suicide prevention program: implications for public health policy. American Journal of Public Health 100, 2457-2463

KOGAN, L.R., MCCONNELL, S.L. & SCHOENFELD-TACHER, R. (2005) Veterinary students and non-academic stressors. Journal of Veterinary Medical Education 32, 193-200

KOHN, M.L. & SCHOOLER, C. (1982) Job conditions and personality: a longitudinal assessment of their reciprocal effects. American Journal of Sociology 87, 1257- 1286

292

KÜÇÜKDEVECI, A.A., TENNANT, A., GRIMBY, G. & FRANCHIGNONI, F. (2011) Strategies for assessment and outcome measurement in physical and rehabilitation medicine: an educational review. Journal of Rehabilitation Medicine 43, 661-672

LA PORTA, F., FRANCESCHINI, M., CASELLI, S., CAVALLINI, P., SUSASSI, S. & TENNANT, A. (2011) Unified balance scale: an activity-based, bed to community, and aetiology-independent measure of balance calibrated with Rasch analysis. Journal of Rehabilitation Medicine 43, 435-444

LABOVITZ, S. & HAGEDORN, R. (1971) An analysis of suicide rates among occupational categories. Sociological Inquiry 41, 67-72

LAMBERT, S., PALLANT, J.F. & GIRGIS, A. (2011) Rasch analysis of the Hospital Anxiety and Depresssion Scale among caregivers of cancer survivors: implications for its use in psycho-oncology. Psycho-oncology 20, 919-925

LAMONTAGNE, A.D., KEEGEL, T., LOUIE, A.M., OSTRY, A. & LANDBERGIS, P.A. (2007) A systematic review of the job-stress intervention evaluation literature, 1990-2005. International Journal of Occupational and Environmental Health 13, 268-280

LAW, C.K., YIP, P.S.F. & CAINE, E.D. (2011) The contribution of charcoal burning to the rise and decline of suicides in Hong Kong from 1997-2007. Social Psychiatry and Psychiatric Epidemiology 46, 797-803

LAWTON, G., LUNDGREN-NILSSON, Å., BIERING-SØRENSEN, F., TESIO, L., SLADE, A., PENTA, M., GRIMBY, G., RING, H. & TENNANT, A. (2006) Cross- cultural validity of FIM in spinal cord injury. Spinal Cord 44, 746-752

LEENTJENS, A.F.G., DUJARDIN, K., MARSH, L., RICHARD, I.H., STARKSTEIN, S.E., & MARTINEZ-MARTIN, P. (2011) Anxiety rating scales in Parkinson’s disease: a validation study of the Hamilton Anxiety Rating Scale, the Beck Anxiety Inventory, and the Hospital Anxiety and Depression Scale. Movement Disorders 26, 407-415

LEITER, M.P. & SCHAUFELI, W.B. (1996) Consistency of the burnout structure across occupations. Anxiety, Stress and Coping 9, 229-243

LEVIN, K.A. & LEYLAND, A.H. (2005) Urban/rural inequalities in suicide in Scotland, 1981-1999. Social Science & Medicine 60, 2877-2890

LINACRE, J.M. (1994) Sample size and item calibration stability. Rasch Measurement Transactions 7, 328. Available at: http://www.rasch.org/rmt/rmt 74m.htm. Accessed: 24 August 2011

LINACRE, J.M. (1998) Do correlations prove scores linear? Rasch Measurement Transactions 12, 605-606. Available at: http://www.rasch.org/rmt/rmt121b.htm. Accessed: 14 October 2011

LINDEMAN, S., LÄÄRÄ, E., HAKKO, H. & LÖNNQVIST, J. (1996) A systematic review on gender-specific suicide mortality in medical doctors. British Journal of Psychiatry 168, 274-279

LINDSAY, G., STRAND, S., CULLEN, M.A., CULLEN, S., BAND, S., DAVIS, H., CONLON, G., BARLOW, J. & EVANS, R. (2011) Parenting Early Intervention 293

Programme Evaluation. Research report DFE-RR121(a). Department for Education, London. Available at: https://www.education.gov.uk/publications/ eOrderingDownload/DFE-RR121A.pdf. Accessed: 27 September 2011

LINSLEY, K.R., SCHAPIRA, K. & KELLY, T.P. (2001) Open verdict v. suicide – importance to research. British Journal of Psychiatry 178, 465-468

LIU, K.Y., BEAUTRAIS, A., CAINE, E., CHAN, K., CHAO, A., CONWELL, Y., LAW, C., LEE, D., LI, P. & YIP, P. (2007) Charcoal burning suicides in Hong Kong and urban Taiwan: an illustration of the impact of a novel suicide method on overall regional rates. Journal of Epidemiology and Community Health 61, 248-253

LOHR, K.N. (2002) Assessing health status and quality-of-life instruments: attributes and review criteria. Quality of Life Research 11, 193-205

LORETTO, W., POPHAM, F., PLATT, S., PAVIS, S., HARDY, G., MACLEOD, L. & GIBBS, J. (2005) Assessing psychological well-being: a holistic investigation of NHS employees. International Review of Psychiatry 17, 329-336

LUCAS, C.P. (1992) The order effect: reflections on the validity of multiple test presentations. Psychological Medicine 22, 197-202

LUNDGREN-NILSSON, Å.L. & TENNANT, A. (2011) Past and present issues in Rasch analysis: the Functional Independence Measure (FIM™) revisited. Journal of Rehabilitation Medicine 43, 884-891

MACKAY, C.J., COUSINS, R., KELLY, P.J., LEE, S. & MCCAIG, R.H. (2004) ‘Management standards’ and work related stress in the UK: policy background and science. Work & Stress 18, 91-112

MAHON, M.J., TOBIN, J.P., CUSAK, D.A., KELLEHER, C. & MALONE, K.M. (2005) Suicide among regular-duty military personnel: a retrospective case-control study of occupation-specific risk factors for workplace suicide. American Journal of Psychiatry 162, 1688-1696

MALMBERG, A., SIMKIN, S. & HAWTON, K. (1999) Suicide in farmers. British Journal of Psychiatry 175, 103-105

MALONE, K.M., OQUENDO, M.A., HAAS, G.L., ELLIS, S.P., LI, S. & MANN, J.J. (2000) Protective factors against suicidal acts in major depression: reasons for living. American Journal of Psychiatry 157, 1084-1088

MANN, J.J. (2003) Neurobiology of suicidal behaviour. Nature Reviews Neuroscience 4, 819-828

MANN, J.J., APTER, A., BERTOLOTE, J., BEAUTRAIS, A., CURRIER, D., HAAS, A., HEGERL, U., LONNQVIST, J., MALONE, K., MARUSIC, A., MEHLUM, L., PATTON, G., PHILLIPS, M., RUTZ, W., RIHMER, Z., SCHMIDTKE, A., SHAFFER, D., SILVERMAN, M., TAKAHASHI, Y., VARNIK, A., WASSERMAN, D., YIP, P. & HENDIN, H. (2005) Suicide prevention strategies – a systematic review. Journal of the American Medical Association 294, 2064-2074

MANN, J.J., WATERNAUX, C., HAAS, G.L. & MALONE, K.M. (1999) Toward a clinical model of suicidal behaviour in psychiatric patients. American Journal of Psychiatry 156, 181-189

294

MARIS, R.W., BERMAN, A.L. & SILVERMAN, M.M. (2000) The social relations of suicides. In Comprehensive Textbook of Suicidology. Guilford Publications. pp 240-265

MARTIN, F. & TAUNTON, A. (2005) Perceptions of the human-animal bond in veterinary education of veterinarians in Washington State: structured versus experiential learning. Journal of Veterinary Medical Education 32, 523-530

MARTIN, F., RUBY, K. & FARNUM, J. (2003) Importance of the human-animal bond for pre-veterinary, first-year, and fourth-year veterinary students in relation to their career choice. Journal of Veterinary Medical Education 30, 67-72

MARTIN, L.T. & KUBZANSKY, L.D. (2005) Childhood cognitive performance and risk of mortality: a prospective cohort study of gifted individuals. American Journal of Epidemiology 162, 887-890

MASSOF, R.W. (2002) The measurement of vision disability. Optometry and Vision Science 79, 516-552

MASSOF, R.W. (2011) Understanding Rasch and item response theory models: applications to the estimation and validation of interval latent trait measures from responses to rating scale questionnaires. Ophthalmic Epidemiology 18, 1-19

MAYER, J.D. & SALOVEY, P. (1997) What is emotional intelligence? In Emotional Development and Emotional Intelligence: Educational Implications. Eds P. Salovey, D. Sluyter. New York, Basic Books. pp 3-31

MCALLISTER, P., GREENBERG, N. & HENDERSON, M. (2011) Occupational psychiatry in the armed forces: should depressed soldiers carry guns? Advances in Psychiatric Treatment 17, 350-356

MCDONALD, R.P. & HO, M-H.R. (2002) Principles and practice in reporting structural equation analyses. Psychological Methods 7, 64-82

MCHORNEY, C.A. & TARLOV, A.R. (1995) Individual-patient monitoring in clinical practice: are available health status surveys adequate? Quality of Life Research 4, 293-307

MCHORNEY, C.A., HALEY, S.M. & WARE, J.E.J. (1997) Evaluation of the MOS SF-36 Physical Functioning Scale (PF-10): II. Comparison of relative precision using Likert and Rasch scoring methods. Journal of Clinical Epidemiology 50, 451- 461

MCHORNEY, C.A., WARE, J.E., LU, J.F.R. & SHERBOURNE, C.D. (1994) The MOS 36-item short-form health survey (SF-36): III. Tests of data quality, scaling assumptions, and reliability across diverse patient groups. Medical Care 32, 40-66

MCKENZIE, K., SERFATY, M. & CRAWFORD, M. (2003) Suicide in ethnic minority groups. British Journal of Psychiatry 183, 100-101

MCLEAN, J., MAXWELL, M., PLATT, S., HARRIS, F. & JEPSON, R. (2008) Risk factors for suicide and suicidal behaviour: a literature review. Scottish Government Social Research. Available at: http://www.scotland.gov.uk/Publications/2008/11 /28141444/23. Accessed: 26 October 2011 295

MCLEAN, J., SCHINKEL, M., WOODHOUSE, A., PYNNONEN A-M. & MCBRYDE, L. (2007) Evaluation of the Scottish safeTALK Pilot. Scottish Development Centre for Mental Health, Edinburgh. Available at: http://www.chooselife.net/uploads/ documents/18safeTALKPilotEvaluationFullReport.pdf. Accessed: 04 October 2011

MCMANUS, I.C., KEELING, A. & PAICE, E. (2004) Stress, burnout and doctors' attitudes to work are determined by personality and learning style: a twelve year longitudinal study of UK medical graduates. BMC Medicine 2, 29

MCMANUS, S., MELTZER, H., BRUGHA, T.S., BEBBINGTON, P.E. & JENKINS, R. (2009) Adult psychiatric morbidity in England, 2007: results of a household survey. National Centre for Social Research, London. Available at: http://www.ic.nhs.uk/ cmsincludes/_process_document.asp?sPublicationID=1231750469828&sDocID=5 446. Accessed: 13 October 2011

MEEHAN, M.P. & BRADLEY, L. (2007) Identifying and evaluating job stress within the Australian small animal veterinary profession. Australian Veterinary Practitioner 37, 70-83

MEIT, S.S., BORGES, N.J. & EARLY, L.A. (2007) Personality profiles of incoming male and female medical students: results of a multi-site 9-year study. Medical Education Online 12, 7

MELCHIOR, M., CASPI, A., MILNE, B.J., DANESE, A., POULTON, R. & MOFFITT, T.E. (2007) Work stress precipitates depression and anxiety in young, working women and men. Psychological Medicine 37, 1119-1129

MELLANBY, R.J. & HERRTAGE, M.E. (2004) Survey of mistakes made by recent veterinary graduates. Veterinary Record 155, 761-765

MELLANBY, R.J. (2005) Incidence of suicide in the veterinary profession in England and Wales. Veterinary Record 157, 415-417

MELLANBY, R.J., HUDSON, N.P.H., ALLISTER, R., BELL, C.E., ELSE, R.W., GUNN-MOORE, D.A., BYRNE, C., STRAITON, S. & RHIND, S.M. (2010) Evaluation of suicide awareness programmes delivered to veterinary undergraduates and academic staff. Veterinary Record 167, 730-734

MELLANBY, R.J., PLATT, B., SIMKIN, S. & HAWTON K. (2009) Incidence of alcohol-related deaths in the veterinary profession in England and Wales, 1993- 2005. The Veterinary Journal 181, 332-335

MELTZER, H., GRIFFITHS, C., BROCK, A., ROONEY, C. & JENKINS, R. (2008) Patterns of suicide by occupation in England and Wales: 2001-2005. British Journal of Psychiatry 193, 73-76

MEREDITH, W. & TERESI, J.A. (2006) An essay on measurement and factorial invariance. Medical Care 44, S69-S77

MICHIE, S. & WILLIAMS, S. (2003) Reducing work related psychological ill health and sickness absence: a systematic literature review. Occupational and Environmental Medicine 60, 3-9

MILHAM, S. & OSSIANDER, E. (2001) Occupational mortality database, Washington State Department of Health, 1950-1999. Available at: https://fortress.wa.gov/doh/occmort/. Accessed: 14 October 2011

296

MILLER, I.W., NORMAN, W.H., BISHOP, S.B. & DOW, M.G. (1986) The modified scale for suicidal ideation: reliability and validity. Journal of Consulting and Clinical Psychology 54, 724-725

MILLER, J.M. & BEAUMONT, J.J. (1995) Suicide, cancer and other causes of death among California veterinarians, 1960-1992. American Journal of Industrial Medicine 27, 37-49

MILLER, M., AZRAEL, D. & HEMENWAY, D. (2004) The epidemiology of case fatality rates for suicide in the Northeast. Annals of Emergency Medicine 43, 723- 730

MITCHENER, K.L. & OGILVIE, G.K. (2002) Understanding compassion fatigue: keys for the caring veterinary healthcare team. Journal of the American Animal Hospital Association 38, 307-310

MOKKINK, L.B., TERWEE, C.B., KNOL, D.L., STRATFORD, P.W., ALONSO, J., PATRICK, D.L., BOUTER, L.M. & DE VET, H.C.W. (2010a) The COSMIN checklist for evaluating the methodological quality of studies on measurement properties: a clarification of its content. BMC Medical Research Methodology 10, 22

MOKKINK, L.B., TERWEE, C.B., PATRICK, D.L., ALONSO, J., STRATFORD, P.W., KNOL, D.L., BOUTER, L.M. & DE VET, H.C.W. (2010b) The COSMIN checklist for assessing the methodological quality of studies on measurement properties of health status measurement instruments: an international Delphi study. Quality of Life Research 19, 539-549

MOKKINK, L.B., TERWEE, C.B., PATRICK, D.L., ALONSO, J., STRATFORD, P.W., KNOL, D.L., BOUTER, L.M. & DE VET, H.C.W. (2010c) The COSMIN study reached international consensus on taxonomy, terminology, and definitions of measurement properties for health-related patient-reported outcomes. Journal of Clinical Epidemiology 63, 737-745

MOŚCICKI, E.K. (1997) Identification of suicide risk factors using epidemiologic studies. Psychiatric Clinics of North America 20, 499-517

MOUNT, M.K., BARRICK, M.R., SCULLEN, S.M & ROUNDS, J. (2005) Higher- order dimensions of the big five personality traits and the big six vocational interest types. Personnel Psychology 58, 447-478

MULLER, S. & RODDY, E. (2009) A Rasch analysis of the Manchester Foot Pain and Disability Index. Journal of Foot and Ankle Research 2, 29

MURPHY, G.E. (1998) Why women are less likely than men to commit suicide. Comprehensive Psychiatry 39, 165-175

NEELEMAN, J. (1998) Regional suicide rates in the Netherlands: does religion still play a role? International Journal of Epidemiology 27, 466-472

NEELEMAN, J., HAPLERN, D., LEON, D. & LEWIS, G. (1997) Tolerance of suicide, religion and suicide rates: an ecological and individual study in 19 Western countries. Psychological Medicine 27, 1165-1171 297

NETTERSTRØM, B., CONRAD, N., BECH, P., FINK, P., OLSEN, O., RUGULIES, R. & STANSFELD, S. (2008) The relation between work-related psychosocial factors and the development of depression. Epidemiologic Reviews 30, 118-132

NIEDERKROTENTHALER, T., VORACEK, M., HERBERTH, A., TILL, B., STRAUSS, M., ETZERSDORFER, E., EISENWORT, B. & SONNECK, G. (2010) Role of media reports in completed and prevented suicide: Werther v. Papageno effects. British Journal of Psychiatry 197, 234-243

NORQUIST, J.M., FITZPATRICK, R., DAWSON, J. & JENKINSON, C. (2004) Comparing alternative Rasch-based methods vs. raw scores in measuring change in health. Medical Care 42, I-25

NUSBAUM, K.E., WENZEL, J.G.W. & EVERLY, G.S., Jr. (2007) Psychologic first aid and veterinarians in rural communities undergoing livestock depopulation. Journal of the American Veterinary Medical Association 231, 692-694

O'CONNOR, R.C. & NOYCE, R. (2008) Personality and cognitive processes: self- criticism and different types of rumination as predictors of suicidal ideation. Behaviour Research and Therapy 46, 392-401

O'CONNOR, R.C. (2007) The relations between perfectionism and suicidality: a systematic review. Suicide and Life-Threatening Behavior 37, 698-714

OFFICE FOR NATIONAL STATISTICS (ONS) (2011) Suicide rates in the United Kingdom, 2000-2009. Statistical Bulletin. 27 January 2011. Available at: http://www.ons.gov.uk/ons/rel/subnational-health4/suicides-in-the-united- kingdom/2009/suicide-rates-in-the-united-kingdom--2000-2009.pdf. Accessed: 14 October 2011

OOMENS, S., GEURTS, S. & SCHEEPERS, P. (2007) Combining work and family in the Netherlands: blessing or burden for one’s mental health? International Journal of Law and Psychiatry 30, 369-384

OWENS, C. OWEN, G., LAMBERT, H., DONOVAN, J., BELAM, J., RAPPORT, F. & LLOYD, K. (2009) Public involvement in suicide prevention: understanding and strengthening lay responses to distress. BMC Public Health 9, 308

OWENS, C., OWEN, G., LLOYD, K., RAPPORT, F., DONOVAN, J. & LAMBERT, H. (2011) Recognising and responding to suicidal crisis within family and social networks: qualitative study. British Medical Journal 343, d5801. doi: 10.1136/bmj.d5801

PALLANT, J. (2007) SPSS Survival Manual. Third edition. Open University Press

PALLANT, J.F., MILLER, R.L. & TENNANT, A. (2006) Evaluation of the Edinburgh Post Natal Depression Scale using Rasch analysis. BMC Psychiatry 6, 28

PALLANT, J.F. & TENNANT, A. (2007) An introduction to the Rasch measurement model: an example using the Hospital Anxiety and Depression Scale (HADS). British Journal of Clinical Psychology 46, 1-18

PARKER, J.D.A., CREQUE, R.E., Sr, BARNHART, D.L., HARRIS, J.I., MAJESKI, S.A., WOOD, L.M., BOND, B.J. & HOGAN, M.J. (2004) Academic achievement in high school: does emotional intelligence matter? Personality and Individual Differences 37, 1321-1330

298

PAUL, E.S. & PODBERSCEK, A.L. (2000) Veterinary education and students' attitudes towards animal welfare. Veterinary Record 146, 269-272

PAULHUS, D.L. & REID, D.B. (1991) Enhancement and denial in socially desirable responding. Journal of Personality and Social Psychology 60, 307-317

PAYKEL, E.S., MYERS, J.K., LINDENTHAL, J.J. & TANNER, J. (1974) Suicidal feelings in the general population: a prevalence study. British Journal of Psychiatry 124, 460-469

PECK, D.F. (2005) Foot and mouth outbreak: lessons for mental health services. Advances in Psychiatric Treatment 11, 270-276

PECK, D.F., GRANT, S., MCARTHUR, W. & GODDEN, D. (2002) Psychological impact of foot-and-mouth disease on farmers. Journal of Mental Health 11, 523- 531

PELL, G. (2005) Use and misuse of Likert scales. Medical Education 39, 970

PERROUD, N., UHER, R., HAUSER, J., RIETSCHEL, M., HENIGSBERG, N., PLACENTINO, A., KOZEL, D., MAIER, W., MORS, O., SOUERY, D., DMITRZAK- WEGLARZ, M., JORGENSEN, L., KOVACIC, Z., GIOVANNINI, C., MENDLEWICZ, J., ZOBEL, A., STROHMAIER, J., MCGUFFIN, P., AITCHISON, K.J. & FARMER, A. (2010) History of suicide attempts among patients with depression in the GENDEP project. Journal of Affective Disorders 123, 131-137

PERSSON, C., WENNMAN-LARSEN, A., SUNDIN, K. & GUSTAVSSON, P. (2008) Assessing informal caregivers’ experiences: a qualitative and psychometric evaluation of the Caregiver Reaction Assessment Scale. European Journal of Cancer Care 17, 189–199

PETERSEN, M.R. & BURNETT, C.A. (2008) The suicide mortality of working physicians and dentists. Occupational Medicine 58, 25-29

PETRIDES, K.V., FREDERICKSON, N. & FURNHAM, A. (2004) The role of trait emotional intelligence in academic performance and deviant behavior at school. Personality and Individual Differences 36, 277-293

PHILLIPS, D.P. (1974) The influence of suggestion on suicide: substantive and theoretical implications of the Werther effect. American Sociological Review 39, 340-354

PHILLIPS-MILLER, D., MORRISON, C. & CAMPBELL, N.J. (2001) Same profession, different career: a study of men and women in veterinary medicine. In Advances in Psychology Research, Volume 5. Ed F. Columbus. Nova Science Publishers. pp 1-53

PHILLIPS-MILLER, D.L., CAMPBELL, N.J. & MORRISON, C.R. (2000) Work and family: satisfaction, stress, and spousal support. Journal of Employment Counseling 37, 16-30

PIRKIS, J. (2009) Suicide and the media. Psychiatry 8, 269-271

PLATT, B., HAWTON, K., SIMKIN, S. & MELLANBY, R.J. (2010a) Suicidal behaviour and psychosocial problems in veterinary surgeons: a systematic review. 299

Social Psychiatry and Psychiatric Epidemiology. doi: 10.1007/s00127-010-0328-6 Published online: 23 December 2010

PLATT, B., HAWTON, K., SIMKIN, S. & MELLANBY, R.J. (2010b) Systematic review of the prevalence of suicide in veterinary surgeons. Occupational Medicine 60, 436-446

PLATT, S. & HAWTON, K. (2000) Suicidal behaviour and the labour market. In The International Handbook of Suicide and Attempted Suicide. Eds K. Hawton, K. van Heeringen. Wiley. pp 309-384

PLATT, S., PAVIS, S. & AKRAM, G. (1999) Changing labour market conditions and health: a systematic literature review 1993-98. Report for European Foundation for the Improvement of Living and Working Conditions. Office for Official Publications of the European Communities, Luxembourg. Available at: http://www.eurofound. europa.eu/pubdocs/1999/15/en/1/ef9915en.pdf. Accessed: 14 October 2011

PTACEK, J.T., LEONARD, K. & MCKEE, T.L. (2004) ‘I’ve got some bad news…’: veterinarians’ recollections of communicating bad news to clients. Journal of Applied Social Psychology 34, 366-390

RAEKALLIO, M., HEINONEN, K.M., KUUSSAARI, J. & VAINIO, O. (2003) Pain alleviation in animals: attitudes and practices of Finnish veterinarians. Veterinary Journal 165, 131-135

RASCH, G. (1960) Probabilistic models for some intelligence and attainment tests. Danish Institute for Educational Research

RCVS (2007) Annual Report 2007. Royal College of Veterinary Surgeons, London

REEVE, B.B., HAYS, R.D., BJORNER, J.B., COOK, K.F., CRANE, P.K., TERESI J.A., THISSEN, D., REVICKI, D.A., WEISS, D.J., HAMBLETON, R.K., LIU, H., GERSHON, R., REISE, S.P., LAI, J-S. & CELLA, D. (2007) Psychometric evaluation and calibration of health-related quality of life item banks: plans for the Patient-Reported Outcomes Measurement Information System (PROMIS). Medical Care 45 (Suppl 1), S22-S31

REEVE, C.L., ROGELBERG, S.G., SPITZMÜLLER, C. & DIGIACOMO, N. (2005) The caring-killing paradox: euthanasia-related strain among animal-shelter workers. Journal of Applied Social Psychology 35, 119-143

REICHENBERG, A. & MACCABE, J.H. (2007) Feeling the pressure: work stress and mental health. Psychological Medicine 37, 1073-1074

REIJULA, K., RÄSÄNEN, K., HÄMÄLÄINEN, M., JUNTUNEN, K., LINDBOHM, M., TASKINEN, H., BERGBOM, B. & RINTA-JOUPPI, M. (2003) Work environment and occupational health of Finnish veterinarians. American Journal of Industrial Medicine 44, 46-57

REINERT, D.F. & ALLEN, J.P. (2007) The alcohol use disorders identification test: an update of research findings. Alcoholism: Clinical and Experimental Research 31, 185-199

RÉSIÈRE, D., MÉGARBANE, B., MANET, P., BUISINE, A. & HILPERT, F. (2001) Acute intoxication by pentobarbital. Presse Medicale 30, 269-270

300

REVICKI, D., HAYS, R.D., CELLA, D. & SLOAN, J. (2008) Recommended methods for determining responsiveness and minimally important differences for patient- reported outcomes. Journal of Clinical Epidemiology 61, 102-109

RHIND, S.M., BAILLIE, S., KINNISON, T., SHAW, D.J., BELL, C.E., MELLANBY, R.J., HAMMOND, J., HUDSON, N.P.H., WHITTINGTON, R.E. & DONNELLY, R. (2011) The transition into veterinary practice: opinions of recent graduates and final year students. BMC Medical Education 11, 64

RIBERIO, J.D. & JOINER, T.E. (2009) The interpersonal-psychological theory of suicidal behaviour: current theory and future directions. Journal of Clinical Psychology 65, 1219-1299

RICE, D. (2008) Suicide by veterinary surgeons [Letter]. Veterinary Record 162, 355

ROBERTSON-SMITH, G., ROBINSON, D., HICKS, B., KHAMBHAITA, P. & HAYDAY, S. (2010) The 2010 RCVS Survey of the UK Veterinary and Veterinary Nursing Professions. Brighton, Institute for Employment Studies. Available at: http://www.rcvs.org.uk/publications/rcvs-survey-of-the-professions- 2010/surveyprofessions2010.pdf. Accessed: 17 August 2011

ROBINSON, D. & HOOKER, H. (2006) The UK Veterinary Profession in 2006: The Findings of a Survey of the Profession Conducted by the Royal College of Veterinary Surgeons. Royal College of Veterinary Surgeons, London

ROGELBERG, S.G., DIGIACOMO, N., REEVE, C.L., SPITZMULLER, C., CLARK, O.L., TEETER, L., WALKER, A.G., CARTER, N.T. & STARLING, P.G. (2007) What shelters can do about euthanasia-related stress: an examination of recommendations from those on the front line. Journal of Applied Animal Welfare Science 10, 331-347

ROHLF, V. & BENNETT, P. (2005) Perpetration-induced traumatic stress in persons who euthanize nonhuman animals in surgeries, animal shelters, and laboratories. Society & Animals 13, 201-219

ROLLIN, B.E. (1987) Euthanasia and moral stress. Loss, Grief & Care 1, 115-126

ROLLIN, B.E. (2006) An Introduction to Veterinary Medical Ethics: Theory and Cases. Second edition. Blackwell

ROMAIN, N., GIROUD, C., MICHAUD, K. & MANGIN, P. (2003) Suicide by injection of a veterinarian barbiturate euthanasia agent: report of a case and toxicological analysis. Forensic Science International 131, 103-107

ROSE, G. ([1992] 2008) Rose’s strategy of preventive medicine. Oxford, Oxford University Press

ROUSSON, V., GASSER, T. & SIEFERT, B. (2002) Assessing intrarater, interrater and test-retest reliability of continuous measurements. Statistics in Medicine 21, 3431-3446

ROUTLY, J.E., TAYLOR, I.R., TURNER, R., MCKERNAN, E.J. & DOBSON, H. (2002) Support needs of veterinary surgeons during the first few years of practice: perceptions of recent graduates and senior partners. Veterinary Record 150, 167- 171 301

RUDD, M.D., BERMAN, A.L., JOINER, T.E., NOCK, M.K., SILVERMAN, M.M., MANDRUSIAK, M., VAN ORDEN, K. & WITTE, T. (2006) Warning signs for suicide: theory, research, and clinical applications. Suicide and Life-Threatening Behavior 36, 255-262

SANDERS, C.R. (1995) Killing with kindness: veterinary euthanasia and the social construction of personhood. Sociological Forum 10, 195-214

SAREEN, J., COX, B.J., AFIFI, T.O., DE GRAAF, R., ASMUNDSON, G.J.G., TEN HAVE, M. & STEIN, M.B. (2005) Anxiety disorders and risk for suicidal ideation and suicide attempts: a population-based longitudinal study of adults. Archives of General Psychiatry 62, 1249-1257

SARTORIUS, N. (2007) Stigma and mental health. Lancet 370, 810-811

SAUNDERS, K.E.A. & HAWTON, K. (2006) Suicidal behaviour and the menstrual cycle. Psychological Medicine 36, 901-912

SCHERNHAMMER, E.S. & COLDITZ, G.A. (2004) Suicide rates among physicians: a quantitative and gender assessment (meta-analysis). American Journal of Psychiatry 161, 2295-2302

SCHNEIDER, B., WETTERLING, T., SARGK, D., SCHNEIDER, F., SCHNABEL, A., MAURER, K. & FRITZE, J. (2006) Axis I disorders and personality disorders as risk factors for suicide. European Archives of Psychiatry and Clinical Neuroscience 256, 17-27

SCHOLTES, V.A., TERWEE, C.B. & POOLMAN, R.W. (2011) What makes a measurement instrument valid and reliable? Injury 42, 236-240

SCHUTTE, N.S., MALOUFF, J.M., THORSTEINSSON, E.B., BHULLAR, N. & ROOKE, S.E. (2007) A meta-analytic investigation of the relationship between emotional intelligence and health. Personality and Individual Differences 42, 921- 933

SCHWENK, T.L., DAVIS, L & WIMSATT, L.A. (2010) Depression, stigma, and suicidal ideation in medical students. Journal of the American Medical Association 304, 1181-1190

SEALE, C. (2009) Legalisation of euthanasia or physician-assisted suicide: survey of doctors’ attitudes. Palliative Medicine 23, 205-212

SERPELL, J.A. (2005) Factors influencing veterinary students’ career choices and attitudes to animals. Journal of Veterinary Medical Education 32, 491-496

SHARMA, S., MUKHERJEE, S., KUMAR, A. & DILLON, W.R. (2005) A simulation study to investigate the use of cutoff values for assessing model fit in covariance structure models. Journal of Business Research 58, 935-943

SHEEHY, N. & O’CONNOR, R.C. (2008) Cognitive style and suicidal behaviour. In Suicide: Strategies and Interventions for Reduction and Prevention. Ed S. Palmer. Hove, Routledge. pp 69-83

SIEGRIST, J. (1996) Adverse health effects of high-effort/low-reward conditions. Journal of Occupational Health Psychology 1, 27-41

302

SIEGRIST, J. (2008) Chronic psychosocial stress at work and risk of depression: evidence from prospective studies. European Archives of Psychiatry and Clinical Neuroscience 258 (Suppl. 5), 115-119

SINGLETON, N., BUMPSTEAD, R., O’BRIEN, M., LEE, A. & MELTZER, H. (2001) Psychiatric Morbidity among Adults Living in Private Households, 2000. The Stationery Office, London

SINOKKI, M., HINKKA, K., AHOLA, K., KOSKINEN, S., KLAUKKA, T., KIVIMÄKI, M., PUUKKA, P., LÖNNQVIST, J. & VIRTANEN, M. (2009) The association between team climate at work and mental health in the Finnish Health 2000 Study. Occupational and Environmental Medicine 66, 523-528

SKEGG, K., FIRTH, H., GRAY, A. & COX, B. (2010) Suicide by occupation: does access to means increase the risk? Australian and New Zealand Journal of Psychiatry 44, 429-434

SMITH, A.B., RUSH, R., FALLOWFIELD, L.J., VELIKOVA, G. & SHARPE, M. (2008) Rasch fit statistics and sample size considerations for polytomous data. BMC Medical Research Methodology 8, 33

SMITH, A.B., WRIGHT, E.P., RUSH, R., STARK, D.P., VELIKOVA, G. & SELBY, P.J. (2006) Rasch analysis of the dimensional structure of the Hospital Anxiety and Depression Scale. Psycho-oncology 15, 817-827

SMITH, D.R., LEGGAT, P.A., SPEARE, R. & TOWNLEY-JONES, M. (2009) Examining the dimensions and correlates of workplace stress among Australian veterinarians. Journal of Occupational Medicine and Toxicology 4, 32

SMITH, E.V. (2002) Understanding Rasch measurement: detecting and evaluating the impact of multidimensionality using item fit statistics and principal component analysis of residuals. Journal of Applied Measurement 3, 205-231

SMITH, R. M. (2000) Fit analysis in latent trait measurement models. Journal of Applied Measurement 2, 199-218

SMITH, R.A. & LEWIS, D. (1989) Suicide by ingestion of T-61. Veterinary and Human Toxicology 31, 319-320

SMITH, R.M., LINACRE, J.M. & SMITH, E.V. (2003) Guidelines for manuscripts. Journal of Applied Measurement 4, 198-204

SMITH, S.C., LAMPING, D.L., BANARJEE, S., HARWOOD, R., FOLEY, B., SMITH, P., COOK, J.C., MURRAY, J., PRINCE, M., LEVIN, E., MANN, A. & KNAPP, M. (2005) Measurement of health-related quality of life for people with dementia: development of a new instrument (DEMQOL) and an evaluation of current methodology. Health Technology Assessment 9, 16-19

SNAITH, R.P. (2003) The Hospital Anxiety and Depression Scale. Health and Quality of Life Outcomes 1, 29

STACK, S. (2001) Occupation and suicide. Social Science Quarterly 82, 384-396

STANSFELD, S. & CANDY, B. (2006) Psychosocial work environment and mental health – a meta-analytic review. Scandinavian Journal of Work, Environment and Health 32, 443-462 303

STANSFELD, S. (2002) Work, personality and mental health. British Journal of Psychiatry 181, 96-98

STANSFELD, S.A., FUHRER, R., SHIPLEY, M.J. & MARMOT, M.G. (2002) Psychological distress as a risk factor for coronary heart disease in the Whitehall II study. International Journal of Epidemiology 31, 248-255

STARK, C., BELBIN, A., HOPKINS, P., GIBBS, D., HAY, A. & GUNNELL, D. (2006a) Male suicide and occupation in Scotland. Health Statistics Quarterly 29, 26-29

STARK, C., GIBBS, D., HOPKINS, P., BELBIN, A., HAY, A. & SELVARAJ, S. (2006b) Suicide in farmers in Scotland. Rural and Remote Health 6, 509

STENNER, A.J. & SMITH M. (1982) Testing construct theories. Perceptual and Motor Skills 55, 415-426

STENNER, A.J., BURDICK, H., SANDFORD, E.E. & BURDICK D.S. (2006) How accurate are Lexile text measures? Journal of Applied Measurement. 7, 307-322

STENNER, A.J., SMITH, M. & BURDICK, D.S. (1983) Toward a theory of construct definition. Journal of Educational Measurement 20, 305-316

STERKEN, J., TROUBLEYN, J., GASTHUYS, F., MAES, V., DILTOER, M. & VERBORGH, C. (2004) Intentional overdose of Large Animal Immobilon. European Journal of Emergency Medicine 11, 298-301

STEVENS, S.S. (1946) On the theory of scales of measurement. Science 103, 677- 680

STEWART-BROWN, S. & JANMOHAMED, K. (2008) Warwick-Edinburgh Mental Well-being Scale. User Guide Version 1, June 2008. Ed J. Parkinson. Available at: http://www.healthscotland.com/documents/2702.aspx. Accessed: 14 October 2011

STEWART-BROWN, S. (2005) Interpersonal relationships and the origins of mental health. Journal of Public Mental Health 4, 24-29

STEWART-BROWN, S., TENNANT, A., TENNANT, R., PLATT, S. & PARKINSON, J. (2009) Internal construct validity of the Warwick-Edinburgh Mental Well-being Scale (WEMWBS): a Rasch analysis using data from the Scottish Health Education Population Survey. Health and Quality of Life Outcomes 7, 15

STORDAL, E., MORKEN, G., MYKLETUN, A., NECKELMANN, D. & DAHL, A.V. (2008) Monthly variation in prevalence rates of comorbid depression and anxiety in the general population at 63-65° North: the HUNT study. Journal of Affective Disorders 106, 273-278

STOUT, W.F. (1990) A new item response theory modeling approach with applications to unidimensionality assessment and ability estimation. Psychometrika 55, 293-325

STOWELL, L.I. (1998) Suicide with the veterinary drug acepromazine. Journal of Analytical Toxicology 22, 166-168

STRAUSS, M.E. & SMITH, G.T. (2009) Construct validity: advances in theory and methodology. Annual Review of Clinical Psychology 5, 1-25

304

STUCKI, G., DALTROY, L., KATZ, J.N., JOHANNESSON, M & LIANG, M.H. (1996) Interpretation of change scores in ordinal clinical scales and health status measures: the whole may not equal the sum of the parts. Journal of Clinical Epidemiology 49, 711-717

TANNENBAUM, J. (1993) Veterinary medical ethics: a focus of conflicting interests. Journal of Social Issues 49, 143-156

TARIS, T.W. & KOMPIER, M. (2003) Challenges in longitudinal designs in occupational health psychology. Scandinavian Journal of Work, Environment and Health 29, 1-4

TAYLOR, C., GRAHAM, J., POTTS, H., CANDY, J., RICHARDS, M. & RAMIREZ, A. (2007) Impact of hospital consultants’ poor mental health on patient care. British Journal of Psychiatry 190, 268-269

TAYLOR, C., GRAHAM, J., POTTS, H.W.W., RICHARDS, M.A. & RAMIREZ, A.J. (2005) Changes in mental health of UK hospital consultants since the mid-1990s. Lancet 366, 742-744

TENNANT, A. & CONAGHAN, P.G. (2007) The Rasch measurement model in rheumatology: What is it and why use it? When should it be applied, and what should one look for in a Rasch paper. Arthritis and Rheumatism 57, 1358-1362

TENNANT, A. & PALLANT, J.F. (2006) Unidimensionality matters! (A tale of two Smiths?) Rasch Measurement Transactions 20, 1048-1051. Available at: http://www.rasch.org/rmt/rmt201c.htm. Accessed: 14 October 2011

TENNANT, R., HILLER, L., FISHWICK, R., PLATT, S., JOSEPH, S., WEICH, S., PARKINSON, J., SECKER, J. & STEWART-BROWN, S. (2007) The Warwick- Edinburgh Mental Well-being Scale (WEMWBS): development and UK validation. Health and Quality of Life Outcomes 5, 63

TERESI, J.A. & FLEISHMAN, J.A. (2007) Differential item functioning and health assessment. Quality of Life Research 16, 33-42

TERWEE, C.B., BOT, S.D.M., DE BOER, M.R., VAN DER WINDT, D.A.W.M., KNOL, D.L., DEKKER, J., BOUTER, L.M. & DE VET, H.C.W (2007) Quality criteria were proposed for measurement properties of health status questionnaires. Journal of Clinical Epidemiology 60, 34-42

THOMAS, K. & GUNNELL, D. (2010) Suicide in England and Wales 1861-2007: a time-trends analysis. International Journal of Epidemiology 39, 1464-1475

THOMAS, K., CHANG, S-S. & GUNNELL, D. (2011) Suicide epidemics: the impact of newly emerging methods on overall suicide rates – a time trends study. BMC Public Health 11, 314

THOMPSON, A.H. (2008) Younger onset of depression is associated with greater suicidal intent. Social Psychiatry and Psychiatric Epidemiology 43, 538-544

THURSTONE, L.L. (1928) Attitudes can be measured. American Journal of Sociology 33, 529–554

TIGHE, J., MCMANUS, I.C., DEWHURST, N.G., CHIS, L. & MUCKLOW, J. (2010) The standard error of measurement is a more appropriate measure of quality for 305

postgraduate assessments than is reliability: an analysis of MRCP (UK) examinations. BMC Medical Education 10, 40

TIMMINS, R.P. (2006) How does emotional intelligence fit into the paradigm of veterinary medical education? Journal of Veterinary Medical Education 33, 71-75

TOMLIN, J.L., BRODBELT, D.C. & MAY, S.A. (2010a) Influences on the decision to study veterinary medicine: variation with gender and background. Veterinary Record 166, 744-748

TOMLIN, J.L., BRODBELT, D.C. & MAY, S.A. (2010b) Veterinary students’ understanding of a career in practice. Veterinary Record 166, 781-786

TORRE, D.M., WANG, N.Y., MEONI, L.A., YOUNG, J.H., KLAG, M.J. & FORD, D.E. (2005) Suicide compared to other causes of mortality in physicians. Suicide and Life-Threatening Behavior 35, 146-153

TRIMPOP, R., AUSTIN, E.J. & KIRKCALDY, B.D. (2000a) Occupational and traffic accidents among veterinary surgeons. Stress Medicine 16, 243-257

TRIMPOP, R., KIRKCALDY, B., ATHANASOU, J. & COOPER, C. (2000b) Individual differences in working hours, work perceptions and accident rates in veterinary surgeries. Work & Stress 14, 181-188

TYSSEN, R. & VAGLUM, P. (2002) Mental health problems among young doctors: an updated review of prospective studies. Harvard Review of Psychiatry 10, 154- 165

TYSSEN, R., DOLATOWSKI, F.C., RØVIK, J.O., THORKILDSEN, R.F., EKEBERG, Ø., HEM, E., GUDE, T., GRØNVOLD, N.T. & VAGLUM, P. (2007) Personality traits and types predict medical school stress: a six-year longitudinal and nationwide study. Medical Education 41, 781-787

TYSSEN, R., HEM, E., VAGLUM, P., GRØNVOLD, N.T. & EKEBERG, Ø. (2004) The process of suicidal planning among medical doctors: predictors in a longitudinal Norwegian sample. Journal of Affective Disorders 80, 191-198

TYSSEN, R., VAGLUM, P., GRØNVOLD, N.T. & EKEBERG, Ø. (2001a) Factors in medical school that predict postgraduate mental health problems in need of treatment: a nationwide and longitudinal study. Medical Education 35, 110-120

TYSSEN, R., VAGLUM, P., GRØNVOLD, N.T. & EKEBERG, Ø. (2001b) Suicidal ideation among medical students and young physicians: a nationwide and prospective study of prevalence and predictors. Journal of Affective Disorders 64, 69-79

VAN BAALEN, B., ODDING, E., VAN WOENSEL, M.P.C., VAN KESSEL, M.A., ROEBROECK, M.E. & STAM, H.J. (2006) Reliability and sensitivity to change of measurement instruments used in a traumatic brain injury population. Clinical Rehabilitation 20, 686-700

VAN DER VEN, A.H.G.S. & ELLIS, J.L. (2000) A Rasch analysis of Raven's standard progressive matrices. Personality and Individual Differences 29, 45-64

VAN DER ZEE, C.H., PRIESTERBACH, A.R., VAN DER DUSSEN, L., KAP, A., SCHEPERS, V.P.M., VISSER-MEILY, J.M.A. & POST, M.W.M. (2010)

306

Reproducibility of three self-report participation measures: the ICF measure of participation and activities screener, the participation scale, and the Utrecht scale for evaluation of rehabilitation-participation. Journal of Rehabilitation Medicine 42, 752-757

VAN HEERINGEN, K. (2002) Towards a psychobiological model of the suicidal process. In Understanding Suicidal Behaviour: The Suicidal Process Approach to Research, Treatment and Prevention. Ed K. van Heeringen. Wiley. pp 136-159

VAN ORDEN, K.A., JOINER, T.E., HOLLAR, D., RUDD, M.D., MANDRUSIAK, M. & SILVERMAN, M.M. (2006) A test of the effectiveness of a list of suicide warning signs for the public. Suicide and Life-Threatening Behavior 36, 272-287

VANDENBERG, R.J. & LANCE, C.E. (2000) A review and synthesis of the measurement invariance literature: suggestions, practices, and recommendations for organizational research. Organizational Research Methods 3, 4-70

VERROTTI, A., CICCONETTI, A., SCORRANO, B., DE BERARDIS, D., COTELLESSA, C., CHIARELLI, F. & FERRO, F.M. (2008) Epilepsy and suicide: pathogenesis, risk factors and prevention. Neuropsychiatric Disease and Treatment 4, 365-370

VETERINARY BENENEVOLENT FUND (2007) Annual Report and Accounts 2006. London, Veterinary Benevolent Fund

VORACEK, M. (2004) National intelligence and suicide rate: an ecological study of 85 countries. Personality and Individual Differences 37, 543-553

VOYDANOFF, P. (2004) The effects of work demands and resources on work-to- family conflict and facilitation. Journal of Marriage and Family 66, 398-412

WALDENSTRÖM, K., AHLBERG, G., BERGMAN, P., FORSELL, Y., STOETZER, U., WALDENSTRÖM, M. & LUNDBERG, I. (2008) Externally assessed psychosocial work characteristics and diagnoses of anxiety and depression. Occupational and Environmental Medicine 65, 90-96

WALLACE, P. (2001) Patients with alcohol problems – simple questioning is the key to effective identification and management. British Journal of General Practice 51, 172-173

WANG, P.S., SIMON, G.E., AVORN, J., AZOCAR, F., LUDMAN, E.J., MCCULLOCH, J., PETUKHOVA, M.Z. & KESSLER, R.C. (2007) Telephone screening, outreach, and care management for depressed workers and impact on clinical and work productivity outcomes: a randomized controlled trial. Journal of the American Medical Association 298, 1401-1411

WARE, J.E. & GANDEK, B. (1998) Methods for testing data quality, scaling assumptions, and reliability: the IQOLA Project approach. International Quality of Life Assessment. Journal of Clinical Epidemiology 51, 945-952.

WAUGH, R.F. & CHAPMAN, E.S. (2005) An analysis of dimensionality using factor analysis (true-score theory) and Rasch measurement: What is the difference? Which method is better? Journal of Applied Measurement 6, 80-99 307

WHITE, A., SHIRALKAR, P., HASSAN, T., GALBRAITH, N. & CALLAGHAN, R. (2006) Barriers to mental healthcare for psychiatrists. Psychiatric Bulletin 30, 382- 384

WHITTINGTON, J.E. & HUPPERT, F.A. (1996) Changes in the prevalence of psychiatric disorder in a community are related to changes in the mean level of psychiatric symptoms. Psychological Medicine 26, 1253-1260

WIECLAW, J., AGERBO, E., MORTENSEN, P.B. & BONDE, J.P. (2005) Occupational risk of affective and stress-related disorders in the Danish workforce. Scandinavian Journal of Work, Environment and Health 31, 343-351

WIECLAW, J., AGERBO, E., MORTENSEN, P.B. & BONDE, J.P. (2006a) Risk of affective and stress-related disorders among employees in human service professions. Occupational and Environmental Medicine 63, 314-319

WILHELM, K., KOVESS, V., RIOS-SEIDEL, C. & FINCH, A. (2004) Work and mental health. Social Psychiatry and Psychiatric Epidemiology 39, 866-873

WILLIAMS, D. (2006) Pressures of student debt [Letter]. Veterinary Record 158, 383

WILLIAMS, J.M.G. & WELLS, J. (1991) Suicidal patients. In Cognitive Therapy in Clinical Practice: An Illustrative Casebook. Eds J. Scott, J.M.G. Williams, A.T. Beck. Routledge. pp 118-128

WILLIAMS, J.M.G. & BROADBENT, K. (1986) Autobiographical memory in suicide attempters. Journal of Abnormal Psychology 95, 144-149

WILLIAMS, J.M.G. & POLLOCK, L.R. (2002) Psychological aspects of the suicidal process. In Understanding Suicidal Behaviour: The Suicidal Process Approach to Research, Treatment and Prevention. Ed K. van Heeringen. Wiley. pp 76-93

WILLIAMS, J.M.G. (2001) Suicide and attempted suicide. Penguin

WILLIAMS, S.M., ARNOLD, P.K. & MILLS, J.N. (2005) Coping with stress: a survey of Murdoch University veterinary students. Journal of Veterinary Medical Education 32, 201-212

WOLF, T.M. (1994) Stress, coping and health: enhancing well-being during medical school. Journal of Medical Education 28, 8-17

WORLEY, L.L.M. (2008) Our fallen peers: a mandate for change. Academic Psychiatry 32, 8-12

WRIGHT, B.D. (1996) Comparing Rasch measurement and factor analysis. Structural Equation Modeling 3, 3-24

WRIGHT, B.D. & LINACRE, J.M. (1989) Observations are always ordinal; measurements, however, must be interval. Archives of Physical Medicine and Rehabilitation 70, 857-860

WRIGHT, B.D. & LINACRE, J.M. (1994) Reasonable mean-square fit values. Rasch Measurement Transactions 8, 370. Available at: http://www.rasch.org/rmt/ rmt83b.htm. Accessed: 24 October 2011

308

YIP, P.S.F. & LEE, D.T. (2007) Charcoal-burning suicides and strategies for prevention. Crisis 28, 21-27

YIP, P.S.F., YANG, K.C.T., IP, B.Y.T., LAW, Y.W. & WATSON, R. (2007) Financial debt and suicide in Hong Kong SAR. Journal of Applied Social Psychology 37, 2788-2799

ZEMAITIENE, N. & ZABORSKIS, A. (2005) Suicidal tendencies and attitude towards freedom to choose suicide among Lithuanian schoolchildren: results from three cross-sectional studies in 1994, 1998 and 2002. BMC Public Health 5, 83

ZENNER, D., BURNS, G.A., RUBY, K.L., DEBOWES, R.M. & STOLL, S.K. (2005) Veterinary students as elite performers: preliminary insights. Journal of Veterinary Medical Education 32, 242-248

ZIGMOND, A.S. & SNAITH, R.P. (1983) The Hospital Anxiety and Depression Scale. Acta Psychiatrica Scandinavica 67, 361-370

ZWICK, W.R. & VELICER, W.F. (1986) Comparison of five rules for determining the number of components to retain. Psychological Bulletin 99, 432-442