Combining Age Related and Additional Needs (CARAN) Report

2007 review of the needs formulae for hospital services and prescribing activity in

Final report

30 November 2007

Stephen Morris* Health Economics Research Group, Brunel University

Roy Carr-Hill Centre for Health Economics, University of York

Paul Dixon Centre for Health Economics, University of York

Malcolm Law Wolfson Institute of Preventive Medicine, Barts and The London, Queen Mary University of London

Nigel Rice Centre for Health Economics, University of York

Matthew Sutton Health Economics Research Unit, University of Aberdeen

Laura Vallejo-Torres Health Economics Research Group, Brunel University

* Correspondence to: Dr. S. Morris, Health Economics Research Group, Brunel University, Uxbridge, Middlesex, UB8 3PH, UK. Email: [email protected].

1 Acknowledgements

We would like to thank Francis Dickinson, Keith Derbyshire, Ravi Baghirathan and Martin Campbell at the Department of Health for their careful and considerate advice and support during this work.

Throughout the project we have benefited from the advice and suggestions provided by the Advisory Committee on Resource Allocation (ACRA) and its Technical Advisory Group (TAG).

We are grateful to the organisations who supplied us with data for the project, sometimes at very short notice: the Office of National Statistics; the Prescribing Support Unit; the HES team at the Department of Health; and, the Information Centre for Health and Social Care.

We would also like to thank the following individuals for their assistance and advice:

Bo Burstrom (Karolinska Institute) David Clucas (Prescribing Support Unit) Alison Crawford (The Information Centre for Health and Social Care) Sue Davies (Office for National Statistics) Chris Dibben (University of St. Andrews) Emma Gordon (Office for National Statistics) Alberto Holly (University of Lausanne) Helen Kendall (Prescription Pricing Division, NHS Business Services Authority) David Lloyd (Prescribing Support Unit) John Reed (The Information Centre for Health and Social Care) David Roberts (Prescribing Support Unit) Eileen Robertson (Department of Health) Chris Roebuck (The Information Centre for Health and Social Care) Steve Rowan (Office for National Statistics) Louise Turner (The Information Centre for Health and Social Care) Wynand van der Ven (Erasmus University) Alan Webber (Department of Health) Sue Westlake (Office for National Statistics) Frank Windmeijer (University of Bristol)

2 Contents

Glossary of terms 9

1. Aims and objectives of the Review 12 1.1 Background to the current Review 12 1.2 Aims and Objectives of the Current Review 13 1.2.1 Stage 1 13 1.2.2 Stage 2 13 1.2.3 Stage 3 13 1.2.4 Contributing to the reduction in avoidable health inequalities 14 1.3 Structure of the report 14

Part I. Review of the current formula and alternatives to it 16

2. Review of the AREA Report 16 2.1 The AREA Report 16 2.1.1 Basic approach 16 2.1.2 Brief description of methods and results 16 2.2 Critiques of the AREA Report 21 2.2.1 Types of critique 21 2.2.2 Approach to modelling 22 2.2.3. NHS deficits 24 2.2.4 Conclusions 26

3. Review of international literature on methods of capitations 27 3.1 Introduction 27 3.2 Types of scheme 27 3.3 Individual capitation schemes 28 3.3.1 Basic ideas 28 3.3.2 Medicare 29 3.3.3 The Netherlands 30 3.4 Area capitation schemes 31 3.4.1 Utilisation approach 32 3.4.2 Epidemiological approach 33 3.5 Conclusions 34

4. Review of the epidemiological approach 36 4.1. Basic ideas 36 4.2. Data sources 37 4.3 Testing the proportionality assumption 40 4.4 Conclusions 45

5. Data and variables 48 5.1 Units of analysis 48 5.1.1 Hospital activity 48 5.1.2 Prescribing activity 48 5.1.3 Individual level modelling 49 5.2 Population data 49

3 5.3 Hospital activity 50 5.3.1 Admitted patient data 50 5.3.2 Outpatient data 51 5.3.3 Costing hospital activity 51 5.4 Prescribing activity 53 5.5 Community health services 54 5.6 Supply variables 55 5.6.1 Approaches to computing the supply of admitted patient 55 facilities 5.6.2 Supply variables for different types of hospital activity 56 5.6.3 Outpatient supply 56 5.6.4 Supply of care home places 57 5.6.5 Primary care 57 5.6.6 Supply variables for practice level modelling 57 5.7 Practice characteristics 57 5.8 Needs indicators 58 5.8.1 Vital events from ONS 58 5.8.2 2001 Census 59 5.8.3 Index of Deprivation 2004 (ID2004) 59 5.8.4 Administrative datasets 60 5.8.5 Practice level needs indicators 61 5.8.6 Reducing the starting set of needs indicators 62 5.9 Quality and Outcomes Framework 62 5.9.1 QOF prevalence data 62 5.9.2 Standardisation 63 5.9.3 External validity 63 5.9.4 Conclusions 66 5.10 Regional indicators 66

6. Individual level models 68 6.1 Introduction 68 6.2 Creation of the individual level dataset 68 6.2.1 Individuals with no activity 68 6.2.2 Tests of sample stability 69 6.2.3 Information on previous diagnoses 69 6.2.4 Unit of measurement of covariates 69 6.3 Summary statistics 69 6.4 Two-part models 71 6.5. Analysis of the role of historical diagnoses 74 6.6 Conclusions 75

7. Addressing unmet needs 76 7.1 Background 76 7.2 Unmet needs in the previous Review 76 7.2.1 Treatment of variables with counter-intuitively signed 76 coefficients 7.2.2 Addition of predicted morbidity indices constructed using 76 individual self-reported morbidity data from the HSE 7.2.3 Investigating regional variations in the effects of needs 77 variables

4 7.3. Unmet needs in the current Review 77 7.3.1 Treatment of variables with counterintuitively signed 77 coefficients 7.3.2 Addition of morbidity-related data from QOF and from HES 78 7.3.3 Investigating regional variations in the effects of needs 79 variables 7.4 Treatment of rurality 79 7.5 Modelling met needs 79 7.6 Conclusions 79

8. Replicating the AREA Report 80 8.1. Background 80 8.2 Acute and maternity activity 80 8.2.1 Dependent variable 80 8.2.2 Independent variables 80 8.2.3 Results 82 8.3 Prescribing activity 83 8.3.1 Dependent variable 83 8.3.2 Independent variables 84 8.3.3 Results 84 8.4 Mental health activity 85 8.4.1 Dependent variable 85 8.4.2 Independent variables 86 8.4.3 Results 86 8.5 Conclusions 87

Part II. Modelling acute activity 88

9. Two stage approach 88 9.1 Background 88 9.2 Summary statistics 88 9.3 Outpatient activity 91 9.4 Age–gender cost curves 93 9.5 Additional needs 97 9.5.1 Deriving models for acute activity 97 9.5.2 Additive and multiplicative models 99 9.5.3 Transforming potential needs indicators into z-scores 100 9.5.4 Indicative additional needs indices 101 9.5.5 Results 101 9.6 Conclusions 106

10. Impact of rurality indicators and the sterilisation of supply 109 10.1 Background 109 10.2 Methods 109 10.3 Results 110 10.4 Conclusions 111

11. Models based on responsive areas 112 11.1 Background 112 11.2 Methods 112

5 11.3 Results 113 11.4 Conclusions 115

12. One stage approach 116 12.1 Background 116 12.2 Two stage approach to modelling age related and additional needs 116 12.3 One stage approaches to modelling age related and additional needs 117 12.3.1 Additive one stage approach 117 12.3.2 Stratified one stage approach 117 12.4 Comparing the approaches 118 12.4.1 Redistribution 118 12.4.2 Explanatory power 119 12.4.3 Estimation 120 12.5 Deriving models using a stratified one stage approach 123 12.6 Conclusions 124

13. Uncertainty intervals associated with estimated needs coefficients 129 13.1 Background 129 13.2 Methods 129 13.2.1 Two stage approach 129 13.2.2 Stratified one stage approach 130 13.3 Results 131 13.4 Conclusions 131

14. Analysis of acute activity data for 2005/6 132 14.1 Background 132 14.2 Methods 132 14.3 Results 133 14.4 Conclusions 146

Part III. Modelling prescribing, mental health and maternity activity 147

15. Modelling prescribing activity 147 15.1 Approach 147 15.1.1 Units of analysis 147 15.1.2 Two stage versus one stage approaches 147 15.1.3 Independent variables 147 15.1.4 Variable selection 148 15.1.5 Analysis 148 15.1.6 Indicative additional needs indices 148 15.2 Results 148 15.2.1 Summary statistics 148 15.2.2 Regression results 149 15.3 Conclusions 150

16. Modelling mental health activity 154 16.1 Approach 154 16.1.1 Units of analysis 154 16.1.2 Hospital versus community activity 154 16.1.3 Two stage versus one stage approaches 154

6 16.1.4 Independent variables and variable selection 155 16.1.5 Analysis 155 16.2 Results 155 16.2.1 Summary statistics 155 16.2.2 Age gender cost curves 158 16.2.3 Additional needs 160 16.2.4 Stratified one stage approach 168 16.2.5 Modelling antidepressant prescribing activity 169 16.3 Conclusions 170

17. Modelling maternity activity 171 17.1 Background 171 17.2 Two stage approach 171 17.2.1 Units of analysis 171 17.2.2 Two stage versus one stage approaches 171 17.2.3 Independent variables and variable selection 171 17.2.4 Analysis 172 17.2.5 Summary statistics 172 17.2.6 Age gender cost curves 175 17.2.7 Additional needs 178 17.3 Cost per birth approach 178 17.3.1 Basic approach 178 17.3.2 Units of analysis 179 17.3.3 Births data 179 17.3.4 Independent variables and variable selection 180 17.3.5 Results 181 17.4 Conclusions 186

Part IV. Practice-based allocations 187

18. Practice-based allocations 187 18.1 Background 187 18.2 Methods 187 18.2.1 Approaches to modelling 187 18.2.2 Practice data 188 18.3 Results 189 18.4 Conclusions 190

Part V. Summary and recommendations 192

19. Summary 192

20. Recommendations and issues for further research 199

Appendices 203 Appendix 1: Schematic diagram of weighted capitation formulae in England 204 Appendix 2: Applying the epidemiological approach: an illustrative example 205 Appendix 3: Updating mid-year super output area population estimates 212 Appendix 4: Supply variables 213 Appendix 5: Practice characteristics 215

7 Appendix 6: Components of the ID2004 216 Appendix 7: Preliminary sifting of needs indicators 218 Appendix 8: Needs indicators 219 Appendix 9: Provider types 221 Appendix 10: HES morbidity indices 222 Appendix 11: Base model for combined admitted patient and outpatient 223 activity with coefficients on regional indicators reported Appendix 12: Stratified one stage models removing inconsistent variables 225

References 228

8 Glossary of terms

ACRA Advisory Committee on Resource Allocation. Additional needs Needs for health care over and above those pertaining to age and gender. AREA Report Report published following the last comprehensive review of the needs formulae in England, entitled Allocation of Resources to English Areas. ASTRO-PU Age, Sex and Temporary Resident Prescribing Unit; a set of weights designed to weight individual GP practice populations for age, sex and temporary residents. Attribution Dataset A dataset that records the home postcode of people registered with every GP practice. Backward stepwise and Statistical procedures for identifying a subset of variables forward stepwise selection that are significantly correlated with the dependent procedures variable in a regression model. Cost-weighted activity Utilisation of health care services multiplied by the unit cost of those services; the dependent variable in the hospital activity models. Counter-intuitive signs Refers to variables that have coefficients with unexpected signs in a regression model (e.g., we might expect areas with more employment deprivation to have higher use all else equal when in fact they may have lower use); these may be due to unmet needs but may be due to other factors as well. Epidemiological approach An approach to allocating health care resources based on the assumption that health care needs in an area are proportional to the numbers of cases of disease in an area. EQ5D A standardised instrument for use as a measure of health outcome; measured on a scale from zero (death) to one (full health). Face validity Whether or not a variable or