Hamid Reza Ravanbod Thesis

Home , Baricitinib, Pegsunercept, Tofacitinib

School of Biomedical Sciences Charles Sturt University

Rheumatoid Arthritis and Risk of Infection: The Role of Disease-Modifying Anti-inflammatory Drugs

Hamid Reza Ravanbod MBBS, M.Sc. (Public Health), M.Sc. (Podiatric Surgery)

Submitted in partial fulfilment of the requirements for the degree of Doctor of Philosophy First submitted August 2019 Revised March 2020

CERTIFICATE OF AUTHORSHIP

Hamidreza Ravanbod

PUBLICATIONS FROM THIS WORK

1. Hamid Reza Ravanbod, H.R., Jazayeri, J.A., Russell, K.G., and Carroll, G.J. (2017). Serious infections in rheumatoid arthritis and strategies for their prevention - A review and discussion of implications for clinical practice. Journal of Immunology, Infection & Inflammatory Diseases, 2(3). https://scientonline.org/open-access/serious-infections- in-rheumatoid-arthritis-and-strategies-for-their-prevention-a-review-and-discussion- of-implications-for-clinical-practice.pdf 2. Serious and non-serious infections in recipients of conventional synthetic and biologic DMARDs in rheumatoid arthritis; an examination of self-reported data from the ARAD registry (in preparation).

ETHICS APPROVAL

This research was approved by the Human Research Ethics Committee (HREC), Charles Sturt University. Protocol number: 2014/080 (Appendix L).

iii

ACKNOWLEDGEMENTS

I would like to express my sincere thanks and gratitude to my supervisors, Dr. Jalal Jazayeri (CSU, principal supervisor) and Dr. Graeme Carroll (UWA) for their guidance and supports during this research. I also wish to express my gratitude to late Professor Kenneth Russell for his expert advice and contributions to the statistics in the fifth chapter. While this PhD research has never been easy, it has always been a privilege to undertake it. Thank you, both, for helping me along the way.

I would also like to thank Dr Sandra Savocchia, Dr Christopher Scott, Ms. Vibhasha Chand, and Mr Abishek Santhakumar for their various advice or assistance with technical matters.

I would like to acknowledge Kara Gilbert for proofreading this thesis, in accordance with the ethical standards for editing and proofreading contained in the Australian Standards for Editing Practices (2nd ed.) (2013) as set out by the Institute of Professional Editors (IPEd) in relation to editing and proofreading research these.

Special thanks to my family, parents, and my children, who supported me and gave me time to finish this project, and to my employers, who provided me with an ongoing income to support my family during my university studies.

LIST OF ABBREVIATIONS

ABT: Abatacept ACPA: Anti-citrullinated peptide antibody ACR: American College of Rheumatology AD: Anno Domini ADA: Adalimumab AIC: Akaike information criterion AIRR: Annualised internal rate of return ANC: Absolute neutrophil count APC: Antigen-presenting cells APRIL: A proliferation-inducing ligand ARAD: Australian Rheumatology Association Database; bDMARDs: Biologic disease-modifying anti-rheumatic drugs BJM: Bone, joint, muscle BMI: Body mass index BSRBR: British Society for Rheumatology Biologics Register CABG: Coronary artery bypass grafting CCP: Cyclic citrullinated peptide, CHD: Coronary heart disease CMV: Cytomegalovirus CNS: Central nervous system COPD: Chronic obstructive pulmonary disease CoQ10: Coenzyme Q 10 CRP: C-reactive protein CS: Corticosteroid csDMARDs: Conventional synthetic biologic disease-modifying anti-rheumatic drugs CSF: Colony-stimulating factor CSU: Charles Sturt University CV: Cardiovascular CVID: Common variable immunodeficiency CYA: Cyclosporine A DF: Degree of freedom DM: Diabetes mellitus

DMARD: Disease-modifying anti-rheumatic drugs DREAM: Dutch Rheumatoid Arthritis Monitoring Registry (Netherlands) EENT: Eye, ear, nose, throat, EOW: Every other week ESR: Erythrocyte sedimentation rate ETN: Etanercept GAG: Glycosaminoglycans GCONV: Global convergence variable GDR: German RABBIT Registry Review GISEA: Registry (Italian Group for the Study of Early Arthritis) GIT: Gastrointestinal tract (GIT) GM CSF: granulocyte-macrophage colony-stimulating factor HAQ Score: Health assessment questionnaire score HB: Hepatitis B HCQ: Hydroxychloroquine HLA: Human leukocyte antigen HREC: Human Research Ethics Committee T1DM: Insulin-dependent diabetes mellitus IHD: Ischemic heart disease ILD: Interstitial lung disease IM: Intramuscular IMIDs: Immune-mediated inflammatory diseases INX: Infliximab IR: Incidence rate IRR: Incidence rate ratio IUIS: International Union of Immunological Societies IV: Intravenous JAK: Janus kinase inhibitors -2 Log L: Deviance in the model LDA: Low disease activity LEF: Leflunomide lr: Likelihood ratio LRTI: Lower respiratory tract infection MBDA: Multi biomarker disease activity

MBL: Mannose binding lectin (MBL) MCP: Metacarpophalangeal MCSF: Macrophage colony-stimulating factor, MHDA: Moderate to high disease activity MI: Myocardial infarction MMPs: Matrix metalloproteinases MSK: Musculoskeletal MTP: Metatarsophalangeal MTX: Methotrexate NF-KB: Nuclear factor kappa-Β, T2DM: Non-insulin dependent diabetes mellitus NK: Natural killer cell NSAIDs: Non-steroidal, anti-inflammatory drug NTM: Non-tuberculous mycobacterial OCP: Oral contraceptive pill OIs: Opportunistic infections PG: Proteoglycan PIP: Proximal interphalangeal PML: Leukoencephalopathy PRISMA: Preferred reporting items for systematic reviews PYs: Person-years RA: Rheumatic arthritis RABBIT: Rheumatoid arthritis (RA) observation of biologic therapy RANKL: Receptor activator of nuclear factor kappa-Β ligand RCT: Randomised control (or controlled) trial RF: Rheumatic factor RTX: Rituximab RX: A medical prescription SAS software: Statistical Analysis System software SC: Schwarz criterion SD: Standard deviation SERENE: Study evaluating rituximab’s efficacy in MTX iNadequate rEsponders SI: Serious infection SIE: Serious infection event

vii

SSTIs: Skin and soft tissue infections TB: Tuberculosis TCZ: Tocilizumab TKI: Tyrosine kinase inhibitor TNF-α: Tumour necrosis factor-α TNFI: TNF inhibitor TOF: Tofacitinib UK: United Kingdom US: Ultrasound USA: United States of America UTI: Urinary tract infection UWA: University of Western Australia

viii

WHOLE THESIS ABSTRACT

The development of infection is far more common in rheumatoid arthritis (RA) patients than in the general population. It is probably one of the most important consequences of RA. It is shown that RA can also increase the rate of serious infection (SI), from less than one per hundred patient years (100PYs) in the normal population to around five per 100PYs in the RA population. The risk of infection in RA increases due to several factors. Some of these include (i) the nature of RA disease and the pathophysiological changes in the immune system, (ii) RA medications, a number of which suppress the immune system, and (iii) coexisting genetic factors, such as Mannose binding lectin (MBL) deficiency, which increases the risk of immunodeficiency through well-known or unknown mechanism(s).

In this project, data were collected from the Australian Rheumatology Association Database (ARAD), in which a cohort of 3569 RA patients (960 males and 2609 females), who had completed related questionnaires 28176 times (during 200 to 2014) were investigated for the development of infections. Among the 3569 patients, 459 patients were eliminated because they had filled out the questionnaire only once, after which 3110 patients remained. Eight duplicates were eliminated, leaving 27709 visits from 3110 patients. All these visits were examined, to capture self-reported infections in different organs and the medications that were being taken at the time. ARAD reports were statistically analysed using the Chi-square test, Fisher’s exact test and logistic or multinomial logistical regression modelling. The thesis is divided into five chapters:

 Chapter 1 provides a detailed overview of the entire thesis, including a comprehensive background of the topic and the project hypothesis, goals, objectives, and strategies.

 Chapter 2 outlines a comprehensive systematic review in which the implications of the development of infection in RA patients and strategies for the prevention of infection are discussed. This chapter was published as a review article in 2017. This chapter provides a background on the subject of this thesis and provides a comprehensive review of the relevant studies that have been undertaken in this area.

 Chapter 3 outlines a descriptive analysis of the infection status of RA patients, in which the role of disease modifying anti-inflammatory drugs (DMARDs) are investigated. ARAD

reports are examined with respect to demographic and treatment categories. Observed differences were then subjected to descriptive statistical appraisal. This chapter is an introduction to the more complex inferential analysis outlined in chapter 4.

 Chapter 4 outlines an inferential analysis of the association between the risk of infection and each anti-RA medication. The analysis provides valuable information concerning the relative frequency of self-reported infections in users of diverse anti-rheumatic therapies. Various organs, including eyes, ears, nose, throat, lungs, urinary tract, heart, gastrointestinal tract, and the central nervous system (CNS) are examined, as well as systemic infections of a viral and pyogenic nature (sepsis /septicaemia). This provides an introduction to the use of adjusted equations for predicting the risk of infection, presented in the next chapter.

 Chapter 5 presents more complex assessments around the incidence of serious infection, its demographic characteristics, and potential risk factors. Patient reports taken from 27709 visits by 3110 patients during 2001 to 2014 were searched for evidence of hospitalisation or intravenous (IV) infusion for infection. Resultant data were tested using inferential and descriptive analyses, and odds ratios for potential risk factors were calculated. A few studies indicate that RA disease and anti RA medication can specifically increase the risk of serious infections. Serious infection (SI) is still the number one cause of death in RA, globally, and so investigating the basis for SIs is important because of the risk of immediate mortality, ongoing morbidity, and health economic burdens. Moreover, an increased understanding of SIs may lead to the development of improved strategies for the prevention of infection. In Chapter 5, serious infection, with all its potential risk factors, is discussed and analysed in detail.

Based on the systematic literature research, we have found that SI is far more common in RA than in the general population. In addition, anti RA medications have different impacts on serious infections, with corticosteroids demonstrating a huge impact on infection followed by bDMARDs and csDMARDs. The time of prescribing bDMARDs in the first year or after, higher dosage of bDMARDs, and combination therapy with bDMARDs all increase the risk of infection. Although it seems that, in the Australian database, csDMARDs alone, during prescription, can evoke higher rates of infection than bDMARDs alone; this difference is statistically significant in self-reports of heart infection, lung infection (p-value = 0.0156), urinary system infection (p-value = 0.0002), and GIT infection. Both csDMARDs and

bDMARDs are associated with a higher risk of infection in RA. All in all, without isolating the first year of taking bDMARDs, it seems that bDMARDs causes less infection but more serious infection. The impact of various medications on infection depends on the type and severity of infection.

xi TABLE OF CONTENTS

Certificate of authorship ...... i

Publications from this work ...... ii

Ethics approval ...... iii

List of Abbreviations ...... v

Whole Thesis Abstract ...... ix

Table of Contents ...... xiii chapter 1 ...... 1 Abstract ...... 2

1. Introduction ...... 3 1.1 Overview ...... 3 1.2 Overview of the thesis rationale ...... 3 1.3. Background information ...... 3 1.3.1. Rheumatoid arthritis ...... 3 1.3.2. Diagnosis and prevalence in RA ...... 4 1.3.3. Consequences and medication in RA ...... 5 1.3.4. Pathophysiology in RA ...... 5 1.4. Molecular pathogenesis ...... 6 1.4.1. Mechanism of actions of bDMARDs and csDMARDs ...... 7 1.4.2. Mechanism of action of bDMARDs ...... 8 1.4.3. TNFα ...... 9 1.4.4. TNFα inhibitors ...... 9 1.5. Major risk factors ...... 11 1.6. Signs and symptoms and laboratory tests ...... 11 1.7. Complications ...... 13 1.8. Moderate and serious infections ...... 14 1.9. Medical treatment ...... 15 1.9.1. Medication and risk of infection in the literature ...... 17 1.10 Discussion ...... 19 1.11 Organisation of this thesis ...... 23 1.12 Hypotheses to be examined in this thesis ...... 24 1.13 Significance of undertaking this review ...... 25 2. Methods ...... 25

3. Summary of the Results ...... 26 3.1. Strengths of this research ...... 27 xii

3.2. Limitations ...... 28 4. Conclusion ...... 28

References ...... 30

Chapter 2 ...... 37 Abstract ...... 38

1. Introduction ...... 39

2. Methods ...... 40 2.1. Search strategy and selection criteria ...... 40 3. Results and discussions ...... 40 3.1. Study selection ...... 40 3.3. Risk factor categories ...... 44 3.4. The impact of medications (non‐biologics) ...... 45 3.5. Corticosteroids ...... 46 3.6 Synthetic DMARDS ...... 46 3.7 The impact of medications (biologics) ...... 48 3.8. TNF‐α Inhibitors ...... 48 3.9. Abatacept (ABT), Rituximab, Anakinra, Tofacitinib and Tocilizumab ...... 49 3.10. Risks associated with combination therapies ...... 51 3.11. Tuberculosis (TB) and non‐tuberculous mycobacterial (NTM) infections ...... 52 3.12. Serological and other laboratory parameters that influence SI risk ...... 52 3.13. Mannose Binding Lectin (MBL) and other immune deficiencies ...... 52 3.14. Implications for Clinical Practice ...... 54 3.14.1. Age ...... 54 3.14.2. Corticosteroid (CS) Use and Dosage ...... 54 3.14.3. Doses of biologic agents ...... 54 3.14.4. Vaccination ...... 55 3.14.5. Comorbidities related and unrelated to RA ...... 55

4. Conclusion ...... 55

References ...... 57

Chapter 3 ...... 63 Abstract ...... 64

1. Introduction ...... 66 1.1. DMARDs ...... 68 1.2. bDMARDs ...... 68 1.3. Aims and Objectives ...... 69 2. Methods ...... 70 2.1. Data Collection ...... 70 2.2. Statistical Analysis ...... 70 3. Results and discussions ...... 70

xiv

3.1. Demography of whole RA population ...... 70 3.2. Demography of patients taking purely bDMARDs ...... 71 3.3. Demography of patients receiving csDMARDs alone ...... 74 3.4. Comparison of patients receiving bDMARDs and patients on csDMARDs alone ...... 77 3.4.1. Prednisolone comparison ...... 78 3.4.2. Alcohol comparison ...... 79 3.4.3. Smoking comparison ...... 80 3.4.4. Sex distribution comparison ...... 82 3.4.5. T2DM comparison ...... 85 3.4.6. T1DM comparison ...... 86 3.4.7. Skin and nail infections comparison ...... 87 3.4.8. Eyes, Ears, nose, Throat (EENT) Infections – a comparison ...... 89 3.4.9. Heart infections comparison ...... 91 3.4.10. Lung infections comparison ...... 92 3.4.11. Gasterointestinal tract (GIT) infections ...... 95 3.4.12. Urinary tract infections (UTI) ...... 97 3.4.13. Musculoskeletal infections (MSK) ...... 99 3.4.14. Artificial joint infections ...... 102 3.4.15. Nervous system infections ...... 103 3.4.16. Tuberculosis (TB) infection ...... 103 3.3.17. Blood infections ...... 104 3.4.18. Viral Infections ...... 106 3.5. Chapter discussion and conclusion ...... 108 References: ...... 117

Chapter 4 ...... 125 Abstract ...... 126

1. Introduction ...... 128 1.1. Aims ...... 131 1.2. Hypothesis ...... 131 2. Methods ...... 132 2.1 Data Collection ...... 132 2.2. Statistical Analysis ...... 132 3. Results and Discussions ...... 132 3.1. Different organ infections ...... 135 3.2. Eye, Ears, Nose and Throat (EENT) infection ‐ analysis of Anti‐RA medicines ...... 136 3.2.1. Wald Chi-square, Likelihood ratio test and Score test to test significance of differences ...... 137 3.2.2. Effects of medications on Eye Ear Nose and Throat (EENT) infection ...... 137 3.3. Chest or lung infection ‐ analysis of anti‐RA medicines ...... 142 3.3.1. Wald Chi-square, likelihood ratio test and score test to test significance of differences ...... 143 3.3.2. Effects of different medications on lung infection ...... 143

3.4. Skin and Nail infection ‐ analysis of Anti‐RA medicines ...... 150 3.4.1. Effects of different medications on skin and nail infection ...... 151 3.5. Artificial (Prosthetic) Joint infection ‐ analysis of Anti‐RA medicines ...... 156 3.5.1. Wald Chi-square, Likelihood ratio test and Score test to test significance of differences ...... 158 3.5.2. Effects of different medications on artificial (prosthetic) joint infection ...... 158 3.6. Bone, joint and muscle (BJM) infection ‐ analysis of anti‐RA medicines ...... 162 3.6.1. Wald Chi-squared, Likelihood ratio test and Score test to test significance of differences ...... 163 3.6.2. Effects of different medications on bone, joint and muscle infection ...... 163 3.7. Blood infection ‐ analysis of Anti‐RA medicines ...... 170 3.7.1. Wald Chi-square, Likelihood ratio test and Score test to test the significance of differences ...... 171 3.7.2. Effects of different medications on blood infection ...... 171 3.8. Gastro‐intestinal tract infection ‐ analysis of medication confounders ...... 176 3.8.1. Wald Chi-square, Likelihood ratio test and Score test ...... 178 3.8.2. Effects of different medications on GIT infections ...... 178 3.9. Nervous System infection ‐ analysis of medication confounders ...... 182 3.10. TB infection ‐ analysis of medication confounders ...... 183 3.11. Urinary tract infection ‐ analysis of medication confounders ...... 184 3.11.1. Wald Chi-square, Likelihood ratio test and Score test to test significance of differences ...... 186 3.11.2. Effects of medications on Urinary tract infection ...... 186 3.12. Viral infection ‐ analysis of medication confounders ...... 193 3.12.3 Chapter Conclusion ...... 198

References: ...... 203

Chapter 5 ...... 206 Abstract ...... 207

1. Introduction ...... 208 1.1. Aims ...... 209 1.2. Hypothesis ...... 210 2. Methods ...... 210 2.1. Data Collection ...... 210 2.2. Statistical Analysis ...... 211 3.0 Results and discussion ...... 212 3.1. Analysis of Rheumatoid Arthritis (RA) and Serious Infections (SIs) in Australia ...... 212 3.2. Age and gender ...... 215 3.3. Length of time in the program ...... 217 3.4. Time in the program as a function of Gender ...... 218 3.5. Distribution of age groups ...... 218 3.6. Incidence and rate of SIs...... 219 3.7. Incidence of SIs ...... 220

xvi

3.7.1. Rates of serious infections ...... 221 3.7.2. Predictor variables ...... 222 3.8. Prediction of Serious infection ...... 224 4. Discussion ...... 227

5. Chapter conclusion ...... 230

Thesis summary and Remarks ...... 232 Summary of main findings ...... 233 Concluding remarks ...... 236 References: ...... 238

Appendices ...... 241 Description of data in appendix ...... 243 Taking different medication levels ...... 243 Response levels ...... 243

Appendix A: Output of SAS for EENT Infection ...... 244

Appendix B: OUTPUT of SAS for Lung Infection ...... 268

Appendix C: Output of SAS for Nail and skin infection ...... 301

Appendix D: Output of SAS for artificial joint infection ...... 328

Appendix E: Output of SAS for bone muscle joint infection ...... 351

Appendix F: Output of SAS for blood infection ...... 385

Appendix G: Output of SAS for GIT Infection ...... 411

Appendix H: Output of SAS for Nervous system infection ...... 433

Appendix I: Output of SAS for TB infection ...... 461

Appendix J: Output of SAS for Urinary Tract Infection ...... 485

Appendix K: Output of SAS for viral infection ...... 509

Appendix L: Ethical approval for the thesis ...... 535

APPENDIX M: Sample of ARAD questionnaire ...... 536

xvii

CHAPTER 1

Introduction and overview

Abstract

Objective: To provide a comprehensive background to the project and to summarise the goals and approaches of this thesis.

Methods: After a systematic review, ARAD patients’ records from 2001to 2014 were tested using a series of descriptive and inferential statistical analysis. Initially the data was once divided to (i) those with serious infection and those with non-serious infection, Then the development of serious infection was evaluated in patients taking bDMARDs and compared with those who were taking csDMARDs. Afterward in each section, these groups were compared for their features and risk of infection.

Results: In the systematic review 31 articles met the criteria for further analysis and showed increased association of serious infection with taking prednisolone, bDMARDs and to a lesser extent csDMARDs. The risk of infection is reported to be higher in the first year of taking bDMARDs compared to the following years. ARAD data is analyzed by a series of descriptive and inferential analyses. In the descriptive analysis the mean age for RA patients was found to be 61.47; for the group taking csDMARDs it was 59.24 and for those taking bDMARDs it was 62.62 years respectively. ENT infections, with a frequency of 14.75%, were the most common infection type in RA. Heart infection, lung infection, urinary tract infection, and GIT infection were statistically more frequent in users of csDMARDs compared to bDMARDs. Cyclosporine and Prednisolone were almost associated with all types of infections in RA. Age, gender, alcohol consumption, etc. are potentially associated with increased risk of SIs.

Conclusion: Based on the systematic research, SI is far more common in RA than in the general population. Based on ARAD data, for most types of infection, csDMARDs alone are associated with higher rates of diverse infection, whereas bDMARDs alone are more strongly associated with serious infections. According to the ARAD analysis, the most common infection in RA in Australia is EENT infection (14.75%). The risk of any serious infection is almost 2.92% in ARAD and for females this risk starts at younger ages. Among various risk factors, smoking is linked to serious infections.

1. Introduction

1.1 Overview This chapter presents a summary of key information important to the rationale for the thesis. Information about rheumatoid arthritis (RA) is presented, followed by background information about infection in RA. Biologic DMARDs and csDMARDs are defined and their role in the treatment of RA, together with their capacity to predispose to infection, is outlined. The role of different risk factors in increasing the risk of infection is briefly reviewed, followed by background information about medications. The aims, hypotheses, and significance of the study conclude the introduction.

1.2 Overview of the thesis rationale Linkages between RA and serious infection have been hypothesised and continue to be refined as our understanding of RA, its pathogenesis and methods of treatment continue to evolve. A growing body of research indicates that sometimes using effective treatments, such as bDMARDs, is associated with unwanted effects, including minor and major infections, some of which are serious and can be life-threatening or fatal[1].

1.3. Background information

1.3.1. Rheumatoid arthritis

According to Arthritis Australia, RA is an autoimmune disease which causes swelling and pain of the joints. RA disease causes inflammation and joint damage in the smaller joints in the hands and feet, through damage to the lining of the joints. Rarely, in RA, larger joints, such as the knees and hip joints, can also be affected, too [1]. Symptoms vary from person to person and may include symmetric joint pain, swelling and tenderness, with morning stiffness [2].

RA is usually diagnosed from its symptoms, a physical examination and testsm such as blood tests for inflammatory factors and antibodies (anti-CCP), including rheumatoid factor. X-rays can also help to see if joints are damaged[3].

1.3.2. Diagnosis and prevalence in RA

Historically, the precise time at which RA emerged is difficult to determine and is mainly based on both assumptions and empirical analysis; however, it seems that RA is mainly a disease of the modern world. Probably the earliest evidence of RA start from portraits by artists of the Flemish school, during the mid-15th to early 16th centuries. These depictions hint at the existence of rheumatoid-like deformities in the European models used by these artists[4]. According to Australian guidelines, the diagnosis of RA is made on the basis of clinical presentation, in association with autoantibodies and evidence of systemic inflammation. Common features of RA are discussed in the following sections.

Common features of rheumatoid arthritis [5]:  Early morning stiffness for longer than one hour

 Family history of inflammatory arthritis

 Joint swelling in more than five joints and symmetry of the affected area

 Rheumatoid factor positivity, compression tenderness in hands and feet

 Anti CCP positivity, chronicity of symptoms for more than 6 weeks

 Bony erosion apparent in X-rays of the hands, wrists and feet

 Presence of rheumatoid nodules and raised inflammatory markers (ESR and CRP)

 Systemic features, such as fatigue and weight loss, are relatively common

The incidence of RA is generally variable and, overall, the number of people affected by RA is not large. Among all countries, Japan and France have the lowest incidence rates of 8 per 100,000 and 8.8 per 100,000, respectively, while the highest incidence rate is reported to occur in the United States, with 44.6 per 100,000. Rheumatoid arthritis incidence rates may alter marginally, as they are exaggerated by time of reporting and the gap between symptom onset and report to a population-based registry[6].

The prevalence of RA among blacks is lower compared to whites. Although the prevalence of RA among the black population is lower, there is no evidence that the disease manifestations differ. However, there is even some evidence to suggest that RA in blacks is less severe, in

terms of extra-articular manifestations and disability. Rheumatoid arthritis is also more common in women and is relatively uncommon in young men (<35 years). The French Afro- Caribbean population have specific manifestations of the disease, such as high female predominance, high immune seropositivity and low tobacco use. As these data are extracted from France, which has a mixed population, it seems that geographical varieties play a small role in these differences[7]. In another study in North America, the prevalence of RA among Eskimos has been also reported as 0.8 % (the method of diagnosing RA in this study was based on clinical signs and symptoms plus serology, without the benefit of x-rays)[8]. Also, the prevalence rate of RA among Chinese people is different. In a study by Zeng et. al. (2002) the prevalence rate of RA in China is almost 0.2-0.3% of the population[9].

1.3.3. Consequences and medication in RA

Rheumatoid arthritis (RA) has been a major cause of disability and loss of productivity among different populations, including Australians[10]. Multiple comorbidities and extra-articular disease manifestations may accompany rheumatoid arthritis (RA) [10].

Prior to 1990 or thereabouts, a conventional multi-disciplinary management approach to this chronic disease was usually followed. Increasingly, in the last three decades, the emphasis has shifted to the use of multiple and increasingly sophisticated pharmacological measures, with early and aggressive management strategies advocated. One of the disadvantages of this approach has been a rise in the rates of non-serious and also serious infections in a not inconsiderable subset of RA patients receiving these new treatments [10].

1.3.4. Pathophysiology in RA

In rheumatoid arthritis (RA), the site of the initial inflammatory process is the synovial lining of diarthrodial joints. In these joints, usually synovial fluid is the source of food for the articular cartilage and lubricates the cartilage matrix. During the inflammatory process, the synovial tissue undergoes increased vascularization and infiltration by activated macrophages, lymphocytes, and plasma cells. As the disease advances, a pannus forms from the progressive overgrowth of this tissue, which then threatens the adjacent cartilage and bone (6).

Although the aetiology and pathogenesis of RA have yet to be completely elucidated, several factors have been identified that contribute importantly to the disease process. These factors

include genetic contributors, environmental factors, the interaction of genes and environment, and cellular abnormalities. A series of immune system factors such as tumour necrosis factor (TNF), have been identified which, when induced, is elaborated and then interacts with target cells, and appears to drive inflammation and tissue damage. Accordingly, medications, such as infliximab (Remicade), adalimumab (Humira), etanercept (Etanercept), certolizumab pegol (Cimzia), and golimumab (Simponi), have been produced which can prevent the interaction of TNF with its endogenous receptors or bind and neutralize their activity in the extracellular environment [12].

1.4. Molecular pathogenesis The mechanism of action by which progressive inflammation and damage occurs is a complex cellular interplay between several key cell types and processes. Usually RA initiates with abnormal presentation of self-antigen by antigen-presenting cells (APC), such as B cells, dendritic cells, or macrophages, which leads, in turn, to the activation of autoreactive T lymphocytes[13]. As the disease progresses, the sub lining of the synovium is infiltrated by T cells, B cells, macrophages, and plasma cells. T cells, once activated, build up in the affected joint and secrete lymphokines such as interleukin-2 and interferon, and other pro-inflammatory cytokines. In addition to acting as APC, B cells produce RF and other autoantibodies, secrete pro-inflammatory cytokines, such as tumour necrosis factor (TNF)-α, and activate T cells. In addition, macrophages secret cytokines and stimulate synoviocytes to release enzymes, which may damage cartilage and bone[14].

Several other cell types accumulate and stimulate in the synovial membrane of RA patients via activated endothelial cells, including synovial fibroblasts and osteoclasts, both of which can promote bone degradation. Synovial fibroblasts contribute to cartilage and joint destruction through the expression of matrix-degrading enzymes, such as matrix metalloproteinases (MMPs), and are activated by a variety of cytokines, including TNF-α and interleukin-1. The identification and understanding of this process has led to the development of several novel therapeutic strategies that target these cytokines[15]. Osteoclasts resorb bone matrix and are complemented by osteoblasts that produce bone matrix. Macrophage colony-stimulating factor (MCSF) and the receptor antagonist of NF-KB ligand (RANKL) are required for the growth and differentiation needed by osteoclasts to become fully developed. An abnormal activation of osteoclasts leads to the bone destruction observed in RA patients, in whom osteoclast 6

formation in inflamed joints is promoted by pro-inflammatory cytokines through their influence on RANKL expression. Figure 1.1 represents a current model of the hypothesized pathogenesis of RA[15].

Figure 1.1 Schematic picture of pathogenesis in RA.

1.4.1. Mechanism of actions of bDMARDs and csDMARDs

These drugs are immune-suppressive and are designed to slow cartilage damage. There are two types of DMARDs; (i) conventional synthetic DMARD commonly reoffered to as csDMARDs, examples of which include methotrexate, sulfasalazine, hydroxychloroquine, or leflunomide and (ii) Biologic DMARDs (bDMARDs) which only came to market in the early 1990s[16]. These include Lenercept, etanercept, abatacept, infliximab, rituximab, tocilizumab etc. some of these drugs are monoclonal antibody based are produced in prokaryotic or eukaryotic cells using hybridoma technology (Table 2.1). Such monoclonal antibodies are engineered to have specific targets and pharmacodynamic properties. In addition, these drugs are engineered to improve their pharmacokinetics and pharmacodynamics properties such as long stability and serum half-life to reduce their frequency of administration. Such modifications include addition of Fc potion of human IgG antibodies or by PEGylation, addition of polyethylene glycol[16].

1.4.2. Mechanism of action of bDMARDs

There are a range of bDMARDs, which have been developed as anti-inflammatory drugs, targeting a range of proinflammatory cytokines (Figure 1.2). Generally, cytokines are targeted in four ways[17]: 1. Anti-cytokine antibodies 2. Receptor-blocking antibodies 3. Soluble receptors: TNF-α soluble receptors bind to and inactivate TNF-α, thus reducing the TNF-α pool available for membrane-bound receptors and signal transduction 4. Receptor antagonists

A B C D 5. 6. 7. 8. 9. 10. Figure 1.2 Mode of action of anti-cytokines (A) normal cytokine-receptor interactions (B) neutralization of cytokines with either soluble receptors or monoclonal antibodies (C) receptor antagonist block receptor so that no inflammatory signal is sent (D) suppression of inflammatory cytokines by activating anti-inflammatory pathways[18],[19] .

Anti-inflammatory drugs each have their own mechanism of action leading to interfering/blockage of the critical pathways in the inflammatory cascade[16]. For example,  Methotrexate stimulates adenosine release from fibroblasts[16]

 Anti -TNFα inhibitors all bind to the cytokine TNF and inhibit its interaction with the TNF receptors [20]  Hydroxychloroquine has mild immunomodulatory action that inhibits intracellular toll- like receptor TLR9 [21].

1.4.3. TNFα

Tumour necrosis factor-α (TNF-α) is a 26 kDa membrane bound cell signalling cytokine, which has several roles in the immune system. These include: (i) antitumor activity (ii) immune system modulation (iii) inflammation (iv) anorexia, (v) cachexia, (vi) septic shock, (vii) viral replication and (viii) haematopoiesis. In arthritis, these cytokines collectively induce chondrocytes to produce metalloproteinases (MMPs), which contribute to cartilage and bone erosion[22]. Overexpression of TNFα plays a key role in the pathogenesis of many chronic inflammatory and rheumatic diseases, including rheumatoid arthritis, ankylosing spondylitis, psoriatic arthritis, Crohn’s disease, as well in pulmonary inflammation and emphysema and myocarditis etc[22].

1.4.4. TNFα inhibitors

Several biologics have been designed to block the proinflammatory activity of TNFα (Figure 1.3). These include etanercept, infliximab and adalimumab. Such biologics have shown to reduce symptoms and improve function and quality of life [23]. There are two main strategies for inhibiting TNF: 1. Monoclonal anti-TNF antibodies 2. Soluble TNF receptors (sTNF-R) -recombinant protein

Figure 1.3 Structure of some of the TNFα inhibitors: etanercept, recombinant fusion protein with two p75 TNF receptors that is solubilised by linking to the Fc portion of human IgG1; pegsunercept, a soluble tumour necrosis factor receptor which is PEGylated; onerecept, recombinant human TNFα binding protein-1; adalimumab; infliximab, an IgG1 monoclonal antibody; and, CDP571, a humanised monoclonal antibody to TNFα[24] , [22].

1.5. Major risk factors There are several risk factors which may ignite the above-mentioned molecular pathways in RA. Environmental risk factors is one of them. The primary known environmental risk factor for RA is cigarette smoking, however, an unanticipated finding also shows that taking the oral contraceptive for 7 or more consecutive years is associated with a lower risk of RA [25]. Smoking usually is associated with sero-positive, not seronegative, RA [25]. An increase in the duration of smoking years increases the risk of developing seropositive RA [26]. Former smokers are also at risk. Studies show that former smokers remain at risk of RA for anywhere between 10 and 19 years after smoking cessation. Another risk factor is air pollution. In a study by Hart et.al. (2009), the prevalence of RA is higher in the regions of the USA which have greater air pollution [27]. By gathering the results of these two studies, Hart et.al., in their study, concluded that inhaled particulate matter from traffic pollution might contribute to the risk of developing RA [27].

Alcohol consumption, birth weight, and early life hygiene are other well-known risk factors[28]. It has been suggested that there is a dose-dependent inverse risk associated with alcohol consumption and RA [28]. Also, in an analysis of women, it was revealed that women with a higher birth weight (>4.54 kg) had a two- fold increased risk of adult onset RA[28]. In addition, in a number of studies, oral contraceptive pill (OCP) consumption is associated with a lower risk for RA (Kłodziński & Wisłowska, 2018)[28]. A comprehensive list of risk protective and causative risk factors is presented in Table 1.1[8].

1.6. Signs and symptoms and laboratory tests The main symptoms of RA are pain and stiffness. There are usually four distinct phases in RA: an initial phase (no clinical manifestations), an early inflammatory phase (clinical manifestations); a destructive phase (erosions and disease progression); and an ongoing phase (irreversible joint destruction). Two major overlapping subpopulations in RA include individuals who are positive for the presence of rheumatoid factor (RF) and individuals who are positive for the presence of antibodies that can bind cyclic citrullinated peptides (CCP). Patients with neither of these biomarkers tend to have a more benign course and are referred to as having “seronegative” RA [6].

Table 1.1 List of risk factors in RA [29]

Risk Factors Protein tyrosine phosphatase, non-receptor type 22 (PTPN22) Peptidyl arginine deiminase 4 (PADI4) DNA methylation changes

CD40, CC chemokine ligand 21 (CCL21), CC chemokine receptor

6 (CCR6) Tumour necrosis factor receptor-associated factor-1 (TRAF1/C5) Interleukin-6 receptor (IL6R) Increase Chance of MHC regions especially amino acids at positions 70 and 71. disease Fc gamma receptor (FCGR) Tumour necrosis factor receptor-associated factor-1 (TRAF1/C5) Signal transducer and activator of transcription 4 (STAT4) Exposure to tobacco smoke Female sex Low vitamin D intake and levels Obesity Occupational dust (silica) High sodium, red meat and iron consumption Air pollution HLA DRB1*1301 (decreased risk for ACPA positive RA) Statin use Possible protective Healthy diet effect: Consumption of fish Consumption of alcohol Hormone replacement

Rheumatoid factor is a type of antibody present in around 80% of RA patients. It is believed that this antibody attacks healthy tissue and causes inflammation. This factor is assessed and measured in the blood stream and once it passes a certain amount, RF is reported positive. Previously, RF was the only way to diagnose RA but, nowadays, other antibodies, such as anti CCP and antinuclear antibodies, are also being used[29]. Anti-CCP is another destructive antibody in RA causing inflammation and damage to the joints and they may be positive long before symptoms manifest in RA[29]. Antinuclear antibodies, such as ANA, are also antibodies which are circulating normally in the body, and when their amount increases, they can attack normal tissue and are indicators of autoimmune diseases[6]. ESR and CRP are mainly useful to measure the level of inflammation in a particular patient and cannot be used to diagnose RA (table 1.2) [30].

Table 1.2 A list of signs and symptoms and diagnostic laboratory tests in RA[6].

Signs and symptoms Disrupted sleep Low grade fever Fatigue Depression and mood changes Dry eyes and mouth Weight loss Joint pain (more small joints in hands and feet) Joint swelling (more small joints in hands and feet) Joint stiffness (more small joints in hands and feet) Red joint (more small joints in hands and feet) Warm joints (more small joints in hands and feet) Joint deformity (more small joints in hands and feet) Numbness and tingling (feet and hands) Subcutaneous Nodules Dry eyes and mouth Depression and mood changes Muscle aches Lack of appetite Loss of energy Limping Hoarseness Painful walking Laboratory tests ESR CRP RF Anti-CCP Antinuclear Antibody (ANA)

1.7. Complications Although RA is not a terminal disease, related information indicates a gap in mortality between individuals with RA and the general population. For example, RA increases the prevalence of ischemic heart disease (IHD) and pulmonary disease, particularly interstitial lung disease (ILD), type 1 diabetes, obesity, infection in different organs, serious infection, and hypertension [29][5][31]. Rheumatoid arthritis (RA) may require special attention due to the particularly

devastating effects in organs such as lung, heart, CNS, or lymphatic system that sometimes ensue[29].

1.8. Moderate and serious infections Probably one of the most important consequences in RA is the development of infections. Medicine Net has a well worded definition for infection: “Infections may be localized, or may become systemic (body wide)”[32]. Table 1.3 lists a series of most common microbes which frequently cause infection in RA.

Table1.3 Common RA-associated microbes[33].

Name of microbe Proteus mirabilis Epstein-Barr Virus Mycoplasma spp. Prophyromonas spp.

Periodontal disease (PD) is probably one of the most commonly infections associated with RA. This association has been considered since the early 1820s. Almost twenty different bacterial species can cause PD. P. gingivalis, Prevotella intermedia, Tannerella forsythia, and Aggregatibacter actinomycetemcomitans are the most common ones. There is, however, another possibility that PD can increase the incidence of RA[33].

It has been shown that a range of bacterial and viral infections can manifest rheumatic disease symptoms, including reactive arthritis. These infections include gastrointestinal or genitourinary infections with Salmonella, Shigella, Campylobacter, Yersinia, and Chlamydia trachomatis, HIV, parvovirus and hepatitis viruses B and C [34].

Table 1.4 Incidence cohort of RA patients, followed from 1955 to 1994 at the Mayo Clinic [32]

Infection Rate ratio Urinary tract infection 1.1 Septicaemia 1.5 Pneumonia 1.6 Lower respiratory tract infection 1.9 Other 2.0 Intra-abdominal 2.8 Skin or soft tissue 3.3 Osteomyelitis 10.6 Septic arthritis 14.9

RA can also increase the rate of serious infection (SI), from less than one per hundred patient years (100PYs) in the normal population to around 5 per 100PYs in RA, overall (Table 1.4) [5]. The risk of infection in RA increases due to several changes. Some of these changes include RA disease and pathophysiology of changes in the immune system, RA medications, a number of which suppress the immune system, and, finally, sometimes there are coexisting genetic factors, such as Mannose Binding Lectin (MBL) deficiency, which increases the risk of immunodeficiency through well-known or unknown mechanisms [35][33].

The risk for the development of serious infection (SI) can also increase in RA. In the literature, the term serious infection is usually used for an infection which requires specific interventions, such as hospitalisation or intravenous antibiotics or both, or any infection which results in death or severe disability. In this study, data have been collected from participants in the ARAD database, who have self-reported details of their illness, treatment, and course over time, including complications, such as infections. While infection can happen in any organ, based on the literature, the most predominant infections in RA include (i) bronchopulmonary (ii) urogenital (iii) soft tissue and (iv) skin, bone/joint sepsis and gastrointestinal infections [31]. Less common infections in RA include the CNS, the cardiovascular system and the lymphatic system[31].

1.9. Medical treatment The primary goals in treating patients with rheumatoid arthritis (RA) are to reduce pain and stiffness, slow disease development and improve function. Medications, such as non-biologic

and biologic disease-modifying antirheumatic drugs (DMARDs), can reduce pain, retard disease progression, and improve functional outcomes[36] whenever it proves possible to reduce dosages or eliminate these medications. However, studies show that there are probably some positive associations when taking biological medications if there should be a serious infection. Richter et.al., in a study published in 2015, performed an observational cohort study of 947 patients with serious infection in a total cohort of 11,150 participants in the German registry (RABBIT) [35]. He and his colleagues observed that persons exposed to bDMARDs at the time of an SI had a reduced risk of sepsis (septicaemia) and mortality[37].

Wherever DMARDs are discussed in this study, DMARDs is divided to csDMARDs and bDMARDs. csDMARDs, or Conventional synthetic DMARDs, include: 1- IM Methotrexate 2- Hydroxychloroquine, 3- Sulphasalazine, 4- Arava (Leflunomide), 5- Azathioprine, 6- Cyclosporin. bDMARDs, or biologics or biological DMARDs, include: 1- Humera/Adalimumab, 2- Etanercept/Etanercept, 3- Kineret/Anakinra, 4- Remicade/Infliximab, 5- Mabthera/Rituximab, 6- Orencia/Abatacept, 7- Actemra/Tocilizumab, 8- Simponi/Golimumab, 9- Cimzia/Certolizumab Pegol. Prednisolone, IM gold and penicillamine do not belong to any group and are studied separately.

Mechanism of action of csDMARDs Methotrexate (MTX), usually the first drug of choice for people with RA, stimulates adenosine release from fibroblasts. When csDMARDs, such as MTX, are ineffective or partially ineffective, other treatments options will involve bDMARDs.

Mechanism of actions of bDMARDs and csDMARDs These drugs are immune-suppressive and are designed to slow cartilage damage [38]. There are two types of DMARDs; (i) conventional synthetic DMARD, commonly reoffered to as csDMARDs, examples of which include methotrexate, sulfasalazine, hydroxychloroquine and leflunomide, and (ii) biologic DMARDs (bDMARDs), which only came to market in the early 1990s. These include lenercept, etanercept, abatacept, infliximab, rituximab and tocilizumab. Some of these drugs are monoclonal antibody-based and are produced in prokaryotic or eukaryotic cells using hybridoma technology (Table 2.1). Such monoclonal antibodies are engineered to have specific targets and pharmacodynamic properties[38].

In addition, these drugs are engineered to improve their pharmacokinetics and pharmacodynamics properties, such as long stability and serum half-life, to reduce their frequency of administration. Such modifications include the addition of the Fc portion of human IgG antibodies or the addition of polyethylene by glycol PEGylation [39].

Mechanism of action of bDMARDs Cytokines are therapeutic targets for a range of bDMARDs, which are designed to reduce their production (overexpression) or function. There are four ways in which cytokines are targeted. These include the application of (i) anti-cytokine antibodies (ii) receptor-blocking antibodies (iii) soluble receptors and (iv) receptor antagonists. Most bDMARDs fall into one of these categories. Examples include infliximab, lenercept and etanercept and adalimumab among others inhibiting the “second signal” required for T-cell activation, and depleting B-cells or inhibiting factors that active B-cells (rituximab and belimumab) [17].

Considering the difference between anti-RA medication, different countries have developed different therapeutic guidelines, based on factors such as the availability of medication and the health economy. Table 1.5 shows a comparison between the different treatment modalities in Australia, the United States and Canada.

1.9.1. Medication and risk of infection in the literature

Richter et al, in their study in 2015, concluded that bDMARDs supress the immune system [37]. In some studies, bDMARDs are very safe and adding or not adding human antibodies will not change this safety. For example, in a study by Wong Pack published in 2016, the authors examined the incidence of serious infections in RA patients treated either with the combination of denosumab and an immunosuppressive biologic DMARD or with an immunosuppressive biologic DMARD alone [37]. Denosumab is a human monoclonal antibody which is used to treat osteoporosis arising from multiple different causes. The sample included patients over 18 years of age with RA, registered in the practice 3 months before and after the index date, and who had received 1 injection/infusion or filled a prescription for an immunosuppressive biologic DMARD therapy for RA. Among all 308 patients in the sample, the authors concluded that there is a low incidence of SIs in RA patients receiving bDMARDs, including patients who currently are taking bDMARDs (Table 1.5) [43].

Table 1.5 Cross-comparison of RA therapy between Australia, the United States and Canada

Countries Therapeutic Guidelines Initially, start treatment with simple analgesics, such as paracetamol, and supplements, such as Omega-3. Also, patient education and referral to physiotherapist and podiatrist are essential. Pharmaceutical therapy starts Australian initially with NSAIDs and COX-2 inhibitors. If, in spite of using these medicines, swelling is persistent beyond six weeks, the patient needs to be referred to a rheumatologist to start DMARDs or low dose glucocorticoids. Referral to a rheumatologist can happen initially after multiple swollen joints are detected or if six weeks of NSAID therapy does not improve signs and symptoms. Advanced therapy in RA includes combination of DMARDs, leflunomide or cyclosporin or taking biologic agents, anti-TNFs, anakinra and rituximab[40]. American Use a treat-to-target strategy. Start with monotherapy (with MTX) rather than double therapy or triple therapy. In moderate or high disease activity without previous DMARDs, patient should take DMARD monotherapy, which is better than double or triple therapy. If the disease is still active, a combination of DMARDs or a TNFi or a non-TNF biologic, in no particular order, is preferred. If the disease activity remains moderate or high, use TNFI monotherapy or TNFi plus MTX. If disease activity persists and is moderate to high, add low dose glucocorticoid. Depending on the activity of disease, the dosage of glucocorticoid can be increased but it should remain as low as possible[41]. Canadian Start DMARD as soon as possible, through combination with methotrexate (MTX) or monotherapy with MTX. If response is inadequate, then switch to DMARDs. Usually first choice is anti-TNF with MTX, then ABAT/RTX or TCZ. If there is still an inadequate response, switch to any biologic or switch to traditional DMARDs. Inadequate response is defined as not reaching targets by 3 to 6 months[42]. Notes. MTX: Methotrexate; Anti TNF: Tumor necrosis factor inhibitor; ABAT: Abatacept, RTX: Rituximab, TCZ: Tocilizumab

Table 1.6 Samples of bDMARDs and their molecular structure[44] Type Name Description Infliximab Mouse-human chimeric anti-human TNF (Remicade®) mAb Adalimumab Fully human anti-human TNF mAb mAb based (HumiraTM) a humanised monoclonal antibody to TNF- CDP571 α Tumor necrosis factor alpha (TNF-alpha) Golimumab inhibitors Recombinant Fusin Etanercept (Enbrel®) p75sTNF-RII-Fc (dimeric) protein Lenercept p55sTNF-RI-IgG1 (dimeric)

Pegylated Certolizumab (Cimzia) PEGylated anti-TNFα biologic

1.10 Discussion Zamora-leoff et.al. (2016), in a retrospective study among 181 patients suffering from RA, found that the risk of serious infection is the highest in the first year after diagnosis of interstitial lung disease (ILD). They found that the most common types of infection among this group included pneumonia, septicaemia, and opportunistic infections. It was also revealed that prednisolone in doses more than 10 mg, with or without DMARDs, was associated with the highest rate of infection. The authors of this study concluded that the underlying autoimmune process and use of immunosuppressive drugs or both are potential risk factors for higher infection rates among patients with RA-ILD [45]. Curtis et.al. (2018) investigated a sample of 17433 RA patients with hospitalised pneumonia/sepsis SIs and 16796 with myocardial infarction (MI) and coronary heart disease (CHD). They found that higher multi-biomarker disease activity (MBDA) scores were associated with hospitalised infections, predominantly in the older, US RA population [31].

Morel et. al.(2017), in a study of 1491 patients with RA who were treated with tocilizumab, found that a high absolute neutrophil count (ANC) (above 5.0 × 109 at baseline), a negative anti-citrullinated peptide antibody (ACPA) and concomitant therapy with leflunomide (LEF) are predictive factors of serious infection [46].

Accortt et.al. (2018), in a study of patients over 18 years old and with a disease activity index score of two or more than two, found that, compared to low RA disease activity (LDA), moderate‐to‐high RA disease activity (MHDA) had a greater number of serious infections. The authors concluded that lower RA disease activity was associated with lower serious infection rates and recommended that treating physicians strive for remission of RA rather than accept an LDA [31].

Salmon et. al. (2015) revealed that, in practice, usually patients with rheumatoid arthritis treated with abatacept (ABT) have more comorbidities and serious infections are slightly more frequently observed. In the Orencia and Rheumatoid Arthritis (ORA) registry, predictive risk factors for serious infections included age and a proceeding history of serious infections [47]. Hashimoto et. al. (2015), in a study of 370 patients with RA, demonstrated that, although the current disease activity was similar in patients with SIs, patients with multiple SI had greater radiographic joint damage and more advanced physical dysfunction. [48]. Rutherford et.al. (2017) examined 19282 patients with rheumatoid arthritis for 46771 patient-years and they found that the incidence of serious infection was lowest with certolizumab. Rituximab and tocilizumab both have higher rates of infection and there is a possibility that patient factors as opposed to the drug itself were responsible for the observed difference [49].

In another study by Tarp et al. (2017), the crude incidence rate (IR) per 100 patient-years for serious infections was calculated for the sustained remission, low disease activity (LDA), and moderate to high disease activity (MHDA) groups. [31]. Baradat et. al. (2017) published a systematic review of 16 RCTs. In this systematic review, the rates of serious infection and death were compared between patients with RA who were treated with a combination therapy of methotrexate and biological disease-modifying antirheumatic drug (bDMARDs) and patients with RA who were using biological disease-modifying antirheumatic drug (bDMARDs) monotherapy. The authors in this study concluded that there was no significant difference between the two groups. They confirmed that using methotrexate and bDMARDs combination therapy in RA does not cause an increased risk of serious adverse events [50].

De Andrade (2017) published a study in which he concluded that there is no difference in the rate of SIs between patients who were taking rituximab (RTX), on one hand, or bDMARDs (such as TNF inhibitors), on the other. Silva-Fernandez et. al. (2016) presented another study 20

in which the authors demonstrate that there is no difference at all in the risk of SIs over the first year of treatment in patients treated with RTX compared with those treated with a second TNFα after discontinuing a first TNFα [51]. Subesinghe et.al (2018) published a report concerning the recurrence rate of SI among RA patients registered with the British Society for Rheumatology Biologics Register. Among 5289 subjects with at least one serious infection, contributing to 19 431 patient-years follow-up, the first SI rate was 4.6% (95% CI: 4.5, 4.7), increasing to 14.1% (95% CI: 13.5, 14.8) [49].

Pappas et.al. (2017) conducted an extended observation analysis in clinical trials and showed that rituximab does not increase the risk of serious infection events (SIE) in patients with rheumatoid arthritis (RA). They describe characteristics of rituximab-treated patients who experienced a SIE versus those who did not. In this study, they concluded that retreatment with rituximab infusions was not associated with a higher rate of SIEs. [52]. Henry et. al. (2017) showed that, among a sample of 1278 RA patients who were treated with standard vs reduced doses of rituximab for 5 years, the SI rate was lower in those who received reduced doses. [53].

Zhang et.al. (2017) investigated 688 patients with pure RA and examined the association between the infections and disease outcome. The authors concluded that repeated exposure to infectious agents during the disease duration might lead to poor outcome for RA. They advised paying more attention to those patients who have repeatedly infectious agents during their disease duration in order to improve their prognosis [33]. Jinno et.al. (2017) examined 792,921 hospitalisations for infection where there was a secondary diagnosis of RA and concluded that the proportion of hospitalisations for infections among RA patients appeared to decline over time for pneumonia and opportunistic infections (OIs). They also observed a slight decrease in UTIs, a slight increase in skin and soft tissue infections (SSTIs), and an increase in hospitalisations with sepsis. [54]. Bortoluzzi et.al. (2016) studied the databases of the Lombardy Region in the period between 1/1/2004 and 31/12/2013. They concluded that among 4656 RA patients recorded in the database, treatment with bDMARDs was not associated with an increase in hospitalised infection. The risk was lower with abatacept, which accords with the perception that it has a better safety profile [55].

In a meta-analysis by Singh, Cameron, Noorbaloochi et.al. (2015), 525 serious infections were identified in 59 studies. A total of 342 infections occurred with biologic therapy with or without DMARD adjunct therapy and 183 infections occurred with conventional DMARD therapy. They concluded that therapy with standard dose or increased dose biologics or DMARD combination biologics increases the risk of serious infections. Therefore, they advised that practitioners should consider risk factors, such as corticosteroid therapy, increased age and comorbidities to estimate the individual risk of infection when undertaking treatment with biologic DMARDs[56]. Unfortunately, from this Meta-analysis, it is not clear if traditional DMARDs had included corticosteroid as well or whether the data were limited to Methotrexate only.

In an article published by Salmon et.al. (2016), the authors concluded that factors predictive of serious infections include: age, history of previous serious infections, diseases such as diabetes and a lower number of previous anti-TNFα therapies. However, on multivariate analysis, only a history of previous serious or recurrent infections (HR 1.94, 95% CI 1.18 to 3.20, p=0.009) and age (HR per 10-year increase 1.44, 95% CI 1.17 to 1.76, p=0.001) were significantly associated with a higher risk of serious infections. [47]. Subesinghe et.al.(2018) used data from the British Society for Rheumatology Biologics Register -Rheumatoid Arthritis, to follow up 5289 subjects with at least one SI 19 431 patient-years. [49].

Tarp et.al. (2015), in a meta-analysis of 106 trials found that, compared with traditional DMARDs, standard-dose biological drugs and high-dose biological drugs were associated with an increased risk of serious infections, although low-dose biological drugs were not. [57]. In another study by Singh et.al. (2015), the authors performed a systematic review and meta- analysis of patients with RA recorded in Copenhagen University Hospital. They identified 106 trials that reported serious infection among patients who were taking biologics. They concluded that, of traditional DMARDs, standard-dose biological drugs and high-dose biological drugs, only high-dose biological drugs were associated with an increased risk of serious infections. In their analysis, low-dose biological drugs and csDMARDs were not. [57].

In a study by Kawashima et. al. (2017), the impacts of the long-term use of biologic agents on serious infection were investigated. The authors showed that the incidence rate of serious infections was not significantly different between biologics-treated and non-biologic or 22

csDMARDs-treated patients [48]. Prednisolone usage (1-4 mg/da) was significantly associated with serious infections [48]. Curtis et.al. (2016) in their study among 3355 RA patients in the sustained remission group and 3912 in the sustained LDA group, found that patients in sustained remission have a lower risk of serious infections compared to those in sustained LDA [31]. Diederik et.al (2017) in their study compared the effects of TNF inhibitors (TNFi) and rituximab (RTX) on SI rate. The analysis included patients registered in the British Society for Rheumatology Biologics Register (BSRBR)-RA. A total of 3419 patients who were taking tumour necrosis factor inhibitor (TNFi) and 1396 patients who were taking rituximab (RTX) were compared. Patients contributed almost 2765 and 1224 person-years (pyrs), respectively. The risk of SIs was comparable in RA patients using rituximab or a TNFα, in the first year [52].

Altogether, it is difficult to reconcile the conflicting information that has arisen from many disparate studies of infections complicating RA, since the patient cohorts are not always comparable and the drugs under evaluation differ in respect to class or family, dose, duration of therapy and adjunctive agents. Furthermore, the infections are not always well defined. Data pertaining to deaths as a result of SIs tend to be limited or scanty. Assuming very little difference in RA cohorts and that SIs can be relied upon, despite somewhat differing definitions, the following conclusions can be drawn:  Higher rates of SIs are encountered in RA per se, irrespective of treatment.  SIs are a function of RA disease activity.  Corticosteroids are a potent risk factor for SIs. No safe dose has been defined.  Current widely used csDMARDs, such as HCQ, SAS, MTX and LEF, confer only a modest risk for SIs.  SIs are increased in the first year of therapy with a bDMARD and rates taper, thereafter, but probably remain above background risk throughout treatment.

1.11 Organisation of this thesis This thesis interpolates materials from one paper by the authors Dr. Hamid Ravanbod, Dr Jalal Jazayeri, Dr Graeme Carroll and Professor Ken Russell. In Chapter 2, the medical literature concerning serious infections in rheumatoid arthritis is reviewed. In this chapter strategies to prevent serious infection in RA are discussed. In total, 3,324 articles were reviewed to form

this chapter. Among these articles, 31 studies met the selection criteria such as large population size, heterogeneous populations, and English language with adolescence RA.

In Chapters 3 and 4, there are descriptive and inferential analyses of the impact of all available anti-RA medications on serious infections in different organs. The data for this analysis have been gathered from the Australian Rheumatology Association Database (ARAD). Participants with rheumatoid arthritis provided information for the database in response to structured questionnaires. Also, some equations are generated to predict the risk of infection based on some well-known cofactors. An example of the questionnaire is provided in the appendix. Chapter 4 presents the inferential analysis of the RA and organ infections with some of the potential risk factors among the Australian population.

In chapter 5, serious infection with all its potential risk factors is discussed and analysed in detail. Globally, serious infection (SI) is still the main cause of death in RA and investigating the basis for SIs is important because of the risk of immediate mortality, ongoing morbidity, and health economic burdens. Moreover, an increased understanding of SIs may lead to the development of improved strategies for prevention.

The risk of serious infection pertains to most if not all organs in RA is estimated to be in the order of 5 per 100PYs in RA overall (35). It is important to know the most common sites and the most common pathogens for these infections.

A thorough review has been undertaken to identify arrange of risk factors, which include pathophysiology of RA, medications and immunodeficiencies. There is also a discussion of the risk of SIs in RA, potential risk factors and a concise summary concerning the contributions of csDMARDs and bDMARDs to SIs.

1.12 Hypotheses to be examined in this thesis

 Infection (including serious infection) occurs commonly in RA worldwide.

 Regarding medication-induced infection and serious infections in RA, bDMARDs are the safest available anti-RA medications

 There are several detectable risk factors for SI in RA and anti-RA medications are amongst these risk factors. Genetic factors can also impact the risk of serious infection among patients with RA.  Infections and Sis among patients registered with ARAD are common and taking anti- RA medicines can impact the frequency of these infections.

 The risk of infection is different between different patients and it is possible to predict this risk based on cofactors.

 The frequency of self-reported infection in different organs varies in ARAD participants.

 Commonly used anti-RA medications have different impacts on the risk of infection in different organs.

1.13 Significance of undertaking this review Treatment in RA targets pain relief, reduction of joint damage and improved joint function. A growing number of medications are available for the treatment of RA. They are categorised as csDMARDs and bDMARDs. Usually treatment plans can change depending on the disease activity, severity of symptoms, signs and prognosis and sometimes expected medication side effects (25). Conventional synthetic disease-modifying anti-rheumatic drugs (csDMARDs) interfere with the immune system to suppress it, indirectly and non-specifically. On the other hand, biologic DMARDs specifically suppress a pathway in the immune system. There is an increasing trend to use bDMARDs rather than rely on csDMARDs alone in RA (25) (38). In this study, the contributions of csDMARDs and bDMARDs to infections, including Sis, are evaluated and compared. We further categorised infections according to organ type and severity and compared the impact of medication on the development of infection in each organ, separately.

2. Methods A systematic review of all available articles concerning serious infection was conducted. ARAD reports (patients’ responses to questionnaires) from 2001to 2014 were tested under several descriptive analyses and results were analysed statistically by Chi-square test, and Fisher test where appropriate or by means of logistic or multinomial logistical regression

modelling. Demographic, disease specific, treatment and infection record data were extracted from the ARA Database which contains details of a cohort of 3569 RA patients (960 males and 2609 females), who had completed related questionnaires 28176 times (during 2001-2014). Among the 3569 patients, 459 patients were eliminated because they had filled out the questionnaire only once. We were therefore left with 3110 patients. After deducing eight duplications and eliminating accordingly, there remained 27709 visits from 3110 patients. In Chapter 3, using filtering procedure in SPSS, the whole data was divided into two groups; those who were just taking bDMARDs without any csDMARDs and patients and those who were taking just csDMARDs without taking any bDMARDs. The demographic distributions of the risk factors were assessed and compared utilizing SAS software.

Furthermore, the whole dataset was tested by SAS software to work out the impact of each anti- RA medicine on the frequency of different types of organ infection and the results are reported in Chapter 4. For this purpose, data were fitted in the logistic regression model and results were tested by using Chi square and Fisher tests. The odds ratio of effectiveness is also calculated for the medicines that have significant effects on infection. Furthermore, each type of organ infection was categorised based on the severity of the infection. Chapter 5 presents more complex assessments around the incidence of serious infection, its demographic characteristics, and potential risk factors. Patients’ reports among 27709 visits from 3110 patients during 2001 to 2014 were searched for evidence of hospitalisation or IV infusion for infection. Resultant data were tested by inferential and descriptive analyses and odds ratios for potential risk factors were calculated. In this section, a few equations were also created to help predict the likelihood of serious infection in a single patient based on known risk factors.

3. Summary of the Results In the systematic review of 3324 articles, only 31 articles met the criteria for the review. Descriptive analysis of ARAD revealed the mean age amongst participants with RA was 61.47. In the group taking csDMARDs the mean age was 59.24 years and, in the group taking bDMARDs, the mean age was 62.62 years. The Wald P-value of the differences between both groups of Ras, based on risk factors, is very large. Taking csDMARDs alone and bDMARDs

alone were associated with statistically significant difference in the rates of heart infection, lung infection, urinary tract infection, and GIT infection.

During 2001 to 2014, the most frequent infections amongst RA participants was related to the eye, ear, nose and throat or EENT (14.75%). Cyclosporine and prednisolone were associated with increased rates for all types of infections, whereas bDMARDs, such as adalimumab, were associated with a reduction in the frequency of nail/skin infection. Just under 3% of the ARAD cohort reported SIs. Adalimumab and etanercept were the most commonly used bDMARDs in patients who reported SIs, but they were also the most frequently prescribed agents in this category. Age, gender, alcohol consumption, medication, diabetes, kidney disease, liver disease, heart attack and, sometimes, previous coronary artery bypass grafting (CABG) were each implicated in higher rates of SI.

3.1. Strengths of this research The research reported here takes a more comprehensive approach to infections in RA, because it does not focus exclusively on serious infections but, rather, includes self-reported infections of diverse severity and categorises these infections according to internally defined levels of severity. Emphasis has been placed on an anatomical and organ-based approach, so that the factors responsible for greater numbers of infections in specific organs and anatomically defined systems can be examined methodically. Although there are several studies comparing csDMARDs and bDMARDs for the development of serious infections, these mainly focus on SIs and so are narrower in scope. Moreover, they lack consensus. This study has the potential to provide a more comprehensive analysis and assist in the acquisition of greater agreement. In the inferential analyses, backward regression was performed. Accordingly, the effect of medications has not only been examined separately, but also, the compounding impacts that combinations of medications may have on each other have been considered. In this study, a large sample has been analysed (28176 patient-visits). This provides considerable statistical power and likely more generalisability regarding the findings. There is also a section on descriptive analysis. This section provides a better understanding of the risk factors and status of the study population and helps to ascertain how far the findings can be generalised. Importantly, this large sample was followed for almost fourteen years (2001 to

2014), which allows longitudinal effects to be more readily captured, thereby strengthening the estimations of SIs calculated in terms of person-years.

3.2. Limitations The main limitations of this study relate to the design of the ARAD questionnaire and the inherent weaknesses associated with the capture and use of self-reported data. Participants may not have known that some illnesses suffered were in fact infections or they may not have remembered to report them when completing a questionnaire some months later for example. Furthermore, it was not possible to validate reported information, since family practitioner confirmation, hospital records and microbiological and other pathology and imaging were unavailable. Reporting in respect to the nature of medications is likely to be reliable, but it is not possible to assess medication compliance. There are also statistical limitations affecting data analysis. In the descriptive analysis, reliance was placed on Chi-Square and P-value calculations. In addition, the reliability of the conclusions in descriptive analysis is compromised by the fact that participants needed to answer a very general question.

4. Conclusion Based on the systematic research, SI is far more common in RA than in the general population. Anti-RA medications have different impacts on this infection, with a huge impact from corticosteroids followed by bDMARDs and csDMARDs. The time of prescribing bDMARDs in the first year or after that, a higher dosage of bDMARDs and combination therapy with bDMARDs all increase the risk of infection.

Compelling evidence has proved that in Australia, RA can increase risk of infection. Although it seems that in the Australian database (ARAD data which is used in the current study) overall, csDMARDs alone during prescription can evoke higher rates of infection than bDMARDs alone; this difference is statistically significant only in self-reports of heart infection, lung infection (P value 0.0156), urinary system infection (P-Value 0.0002), and GIT infection.

Both csDMARDs and bDMARDs are associated with higher risk of infection in RA. All in all, without isolating the first year of taking bDMARDs, it seems that bDMARDs causes less

infection, but more serious infection. The impact of various medications on infection depends on the type of infection and severity of infection.

Serious infection can occur in almost 2.92% of anti RA treatments in Australia, and for females this risk starts in younger ages. Also, in Australia, the majority of patients who develop SIs are taking biologics. It seems that previously taking bDMARDs is a higher risk for patients who are currently taking bDMARDs and those who have never taken this medication. In addition, among different risk factors, which are tested in this review, smoking has a significant connection to the seriousness of infection.

In the next chapter, a comprehensive literature review has been conducted, not only to collect the latest published information in the field but also to investigate the prevalence and status of infection and SI in RA patients and to explore the potential risk factors., including anti-RA medications. In addition, the implications of such infections in clinical practice are also explored and discussed.

References [1] Q. Guo, Y. Wang, D. Xu, J. Nossent, N. J. Pavlos, and J. Xu, “Rheumatoid arthritis: pathological mechanisms and modern pharmacologic therapies,” Bone Res, vol. 6, p. 15, 2018, doi: 10.1038/s41413-018-0016-9.

[2] “Rheumatoid arthritis — Arthritis Australia inflammatory form of arthritis,” Arthritis Australia. https://arthritisaustralia.com.au/types-of-arthritis/rheumatoid-arthritis/ (accessed Jun. 14, 2020).

[3] T. R. A. C. of G. Practitioners, “RACGP - Rheumatoid arthritis.” https://www.racgp.org.au/afp/2010/september/rheumatoid-arthritis/ (accessed Jun. 14, 2020).

[4] J. Dequeker, “Arthritis in Flemish paintings (1400-1700),” Br Med J, vol. 1, no. 6070, pp. 1203–1205, May 1977, doi: 10.1136/bmj.1.6070.1203.

[5] M. Cojocaru, I. M. Cojocaru, I. Silosi, C. D. Vrabie, and R. Tanasescu, “Extra-articular Manifestations in Rheumatoid Arthritis,” Maedica (Buchar), vol. 5, no. 4, pp. 286–291, Dec. 2010, Accessed: May 14, 2019. [Online]. Available: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3152850/.

[6] J. Posalski and M. H. Weisman, “Articular and Periarticular Manifestations of Established Rheumatoid Arthritis,” Rheumatoid Arthritis, pp. 49–61, Jan. 2009, doi: 10.1016/B978-032305475-1.50013-6.

[7] L. Brunier et al., “Prevalence of rheumatoid arthritis in the French West Indies: Results of the EPPPRA study in Martinique,” Joint Bone Spine, vol. 84, no. 4, pp. 455–461, Jul. 2017, doi: 10.1016/j.jbspin.2016.09.003.

[8] R. P. Beasley, H. Retailliau, and L. A. Healey, “Prevalence of rheumatoid arthritis in alaskan eskimos,” Arthritis & Rheumatism, vol. 16, no. 6, pp. 737–742, 1973, doi: 10.1002/art.1780160606.

[9] A. J. Silman and J. E. Pearson, “Epidemiology and genetics of rheumatoid arthritis,” Arthritis Res, vol. 4, no. Suppl 3, pp. S265–S272, 2002, doi: 10.1186/ar578.

[10] M. Bullot and L. Woyzbun, A picture of rheumatoid arthritis in Australia. Canberra: Australian Institute of Health and Welfare, 2009.

[11] J. J. Sacks, Y.-H. Luo, and C. G. Helmick, “Prevalence of specific types of arthritis and other rheumatic conditions in the ambulatory health care system in the United States, 2001-2005,” Arthritis Care Res (Hoboken), vol. 62, no. 4, pp. 460–464, Apr. 2010, doi: 10.1002/acr.20041.

[12] J. Sokolove, “Rheumatoid Arthritis Pathogenesis and Pathophysiology,” 2018, pp. 19– 30.

[13] O. Snir et al., “Antibodies to several citrullinated antigens are enriched in the joints of rheumatoid arthritis patients,” Arthritis Rheum., vol. 62, no. 1, pp. 44–52, Jan. 2010, doi: 10.1002/art.25036.

[14] A. Ioan-Facsinay et al., “Anti-cyclic citrullinated peptide antibodies are a collection of anti-citrullinated protein antibodies and contain overlapping and non-overlapping reactivities,” Ann. Rheum. Dis., vol. 70, no. 1, pp. 188–193, Jan. 2011, doi: 10.1136/ard.2010.131102.

[15] E. A. James et al., “HLA-DR1001 presents ‘altered-self’ peptides derived from joint- associated proteins by accepting citrulline in three of its binding pockets,” Arthritis Rheum., vol. 62, no. 10, pp. 2909–2918, Oct. 2010, doi: 10.1002/art.27594.

[16] G. A. Wells et al., Rationale. Canadian Agency for Drugs and Technologies in Health, 2018.

[17] O. Benjamin, P. Bansal, A. Goyal, and S. L. Lappin, “Disease Modifying Anti- Rheumatic Drugs (DMARD),” in StatPearls, Treasure Island (FL): StatPearls Publishing, 2020.

[18] E. H. S. Choy and G. S. Panayi, “Cytokine Pathways and Joint Inflammation in Rheumatoid Arthritis,” N Engl J Med, vol. 344, no. 12, pp. 907–916, Mar. 2001, doi: 10.1056/NEJM200103223441207.

[19] B. A. Baldo, “Side Effects of Cytokines Approved for Therapy,” Drug Saf, vol. 37, no. 11, pp. 921–943, Nov. 2014, doi: 10.1007/s40264-014-0226-z.

[20] J. B. Galloway et al., “Anti-TNF therapy is associated with an increased risk of serious infections in patients with rheumatoid arthritis especially in the first 6 months of treatment: updated results from the British Society for Rheumatology Biologics Register with special emphasis on risks in the elderly,” Rheumatology (Oxford), vol. 50, no. 1, pp. 124–131, Jan. 2011, doi: 10.1093/rheumatology/keq242.

[21] F. D. Goldman, A. L. Gilman, C. Hollenback, R. M. Kato, B. A. Premack, and D. J. Rawlings, “Hydroxychloroquine inhibits calcium signals in T cells: a new mechanism to explain its immunomodulatory properties,” Blood, vol. 95, no. 11, pp. 3460–3466, Jun. 2000.

[22] H. Lim et al., “Structural Biology of the TNFα Antagonists Used in the Treatment of Rheumatoid Arthritis,” Int J Mol Sci, vol. 19, no. 3, Mar. 2018, doi: 10.3390/ijms19030768.

[23] C. Mabille, Y. Degboe, A. Constantin, T. Barnetche, A. Cantagrel, and A. Ruyssen- Witrand, “Infectious risk associated to orthopaedic surgery for rheumatoid arthritis patients treated by anti-TNFalpha,” Joint Bone Spine, vol. 84, no. 4, pp. 441–445, Jul. 2017, doi: 10.1016/j.jbspin.2016.06.011.

[24] A. J. G. Schottelius, L. L. Moldawer, C. A. Dinarello, K. Asadullah, W. Sterry, and C. K. Edwards, “Biology of tumor necrosis factor-alpha- implications for psoriasis,” Exp Dermatol, vol. 13, no. 4, pp. 193–222, Apr. 2004, doi: 10.1111/j.0906- 6705.2004.00205.x.

[25] D. van der Woude et al., “Gene-environment interaction influences the reactivity of autoantibodies to citrullinated antigens in rheumatoid arthritis,” Nat. Genet., vol. 42, no. 10, pp. 814–816; author reply 816, Oct. 2010, doi: 10.1038/ng1010-814.

[26] H. Mahdi et al., “Specific interaction between genotype, smoking and autoimmunity to citrullinated alpha-enolase in the etiology of rheumatoid arthritis,” Nat. Genet., vol. 41, no. 12, pp. 1319–1324, Dec. 2009, doi: 10.1038/ng.480.

[27] J. E. Hart, F. Laden, R. C. Puett, K. H. Costenbader, and E. W. Karlson, “Exposure to traffic pollution and increased risk of rheumatoid arthritis,” Environ. Health Perspect., vol. 117, no. 7, pp. 1065–1069, Jul. 2009, doi: 10.1289/ehp.0800503.

[28] H. Källberg et al., “Alcohol consumption is associated with decreased risk of rheumatoid arthritis: results from two Scandinavian case-control studies,” Ann. Rheum. Dis., vol. 68, no. 2, pp. 222–227, Feb. 2009, doi: 10.1136/ard.2007.086314.

[29] Ł. Kłodziński and M. Wisłowska, “Comorbidities in rheumatic arthritis,” Reumatologia, vol. 56, no. 4, pp. 228–233, 2018, doi: 10.5114/reum.2018.77974.

[30] H. B. Tenstad, A. C. Nilsson, C. D. Dellgren, H. M. Lindegaard, K. H. Rubin, and S. T. Lillevang, “Use and utility of serologic tests for rheumatoid arthritis in primary care,” p. 7, 2020.

[31] N. A. Accortt et al., “Impact of Sustained Remission on the Risk of Serious Infection in Patients With Rheumatoid Arthritis,” Arthritis Care Res (Hoboken), vol. 70, no. 5, pp. 679–684, May 2018, doi: 10.1002/acr.23426.

[32] K. P. Liang, K. V. Liang, E. L. Matteson, R. L. McClelland, T. J. H. Christianson, and C. Turesson, “Incidence of noncardiac vascular disease in rheumatoid arthritis and relationship to extraarticular disease manifestations,” Arthritis Rheum., vol. 54, no. 2, pp. 642–648, Feb. 2006, doi: 10.1002/art.21628.

[33] S. Li, Y. Yu, Y. Yue, Z. Zhang, and K. Su, “Microbial Infection and Rheumatoid Arthritis,” J Clin Cell Immunol, vol. 4, no. 6, Dec. 2013, doi: 10.4172/2155- 9899.1000174.

[34] K. Thomas and D. Vassilopoulos, “Individual Drugs in Rheumatology and the Risk of Infection,” in The Microbiome in Rheumatic Diseases and Infection, G. Ragab, T. P. Atkinson, and M. L. Stoll, Eds. Cham: Springer International Publishing, 2018, pp. 445– 464.

[35] G. J. Carroll et al., “Undetectable Mannose Binding Lectin and Corticosteroids Increase Serious Infection Risk in Rheumatoid Arthritis,” J Allergy Clin Immunol Pract, vol. 5, no. 6, pp. 1609–1616, Dec. 2017, doi: 10.1016/j.jaip.2017.02.025.

[36] I. C. Olsen, E. Lie, R. Vasilescu, G. Wallenstein, S. Strengholt, and T. K. Kvien, “Assessments of the unmet need in the management of patients with rheumatoid arthritis: analyses from the NOR-DMARD registry,” Rheumatology (Oxford), vol. 58, no. 3, pp. 481–491, Mar. 2019, doi: 10.1093/rheumatology/key338.

[37] A. Richter et al., “Impact of treatment with biologic DMARDs on the risk of sepsis or mortality after serious infection in patients with rheumatoid arthritis,” Ann. Rheum. Dis., vol. 75, no. 9, pp. 1667–1673, Sep. 2016, doi: 10.1136/annrheumdis-2015-207838.

[38] P. Rein and R. B. Mueller, “Treatment with Biologicals in Rheumatoid Arthritis: An Overview,” Rheumatol Ther, vol. 4, no. 2, pp. 247–261, Aug. 2017, doi: 10.1007/s40744-017-0073-3.

[39] K. D. Pile, G. G. Graham, and S. M. Mahler, “Disease-Modifying Antirheumatic Drugs: Overview,” in Compendium of Inflammatory Diseases, M. J. Parnham, Ed. Basel: Springer Basel, 2016, pp. 464–475.

[40] National Health and Medical Research Council (Australia) and Royal Australian College of General Practitioners, Clinical guideline for the diagnosis and management of early rheumatoid arthritis. South Melbourne, Vic.: Royal Australian College of General Practitioners, 2009.

[41] J. A. Singh et al., “2015 American College of Rheumatology Guideline for the Treatment of Rheumatoid Arthritis,” Arthritis & Rheumatology, vol. 68, no. 1, pp. 1–26, 2016, doi: 10.1002/art.39480.

[42] V. P. Bykerk et al., “Canadian Rheumatology Association Recommendations for Pharmacological Management of Rheumatoid Arthritis with Traditional and Biologic Disease-modifying Antirheumatic Drugs,” J Rheumatol, vol. 39, no. 8, pp. 1559–1582, Aug. 2012, doi: 10.3899/jrheum.110207.

[43] M. A. Othman, W. S. W. Ghazali, N. K. Yahya, and K. K. Wong, “Correlation of Demographic and Clinical Characteristics with Rheumatoid Factor Seropositivity in Rheumatoid Arthritis Patients,” Malays J Med Sci, vol. 23, no. 6, pp. 52–59, Nov. 2016, doi: 10.21315/mjms2016.23.6.6.

[44] “MIMS Australia.” https://www.mims.com.au/ (accessed Jul. 11, 2020).

[45] J. A. Zamora-Legoff, M. L. Krause, C. S. Crowson, J. H. Ryu, and E. L. Matteson, “Risk of serious infection in patients with rheumatoid arthritis-associated interstitial lung disease,” Clin. Rheumatol., vol. 35, no. 10, pp. 2585–2589, Oct. 2016, doi: 10.1007/s10067-016-3357-z.

[46] J. Morel et al., “Risk factors of serious infections in patients with rheumatoid arthritis treated with tocilizumab in the French Registry REGATE,” Rheumatology (Oxford), vol. 56, no. 10, pp. 1746–1754, 01 2017, doi: 10.1093/rheumatology/kex238.

[47] J. H. Salmon et al., “Predictive risk factors of serious infections in patients with rheumatoid arthritis treated with abatacept in common practice: results from the Orencia and Rheumatoid Arthritis (ORA) registry,” Ann. Rheum. Dis., vol. 75, no. 6, pp. 1108– 1113, Jun. 2016, doi: 10.1136/annrheumdis-2015-207362.

[48] H. Kawashima et al., “Long-term use of biologic agents does not increase the risk of serious infections in elderly patients with rheumatoid arthritis,” Rheumatol Int, vol. 37, no. 3, pp. 369–376, 2017, doi: 10.1007/s00296-016-3631-z.

[49] S. Subesinghe, A. I. Rutherford, R. Byng-Maddick, K. Leanne Hyrich, and J. Benjamin Galloway, “Recurrent serious infections in patients with rheumatoid arthritis-results from the British Society for Rheumatology Biologics Register,” Rheumatology (Oxford), vol. 57, no. 4, pp. 651–655, 01 2018, doi: 10.1093/rheumatology/kex469.

[50] C. Baradat, Y. Degboé, A. Constantin, A. Cantagrel, and A. Ruyssen-Witrand, “No impact of concomitant methotrexate use on serious adverse event and serious infection risk in patients with rheumatoid arthritis treated with bDMARDs: a systematic literature review and meta-analysis,” RMD Open, vol. 3, no. 1, p. e000352, Feb. 2017, doi: 10.1136/rmdopen-2016-000352.

[51] C. A. F. de Andrade, “Trying to find an answer for an old question: does Rituximab increase the risk of serious infections in patients with rheumatoid arthritis?,” Rheumatology (Oxford), vol. 57, no. 9, pp. 1505–1506, Sep. 2018, doi: 10.1093/rheumatology/kex439.

[52] D. A. Pappas et al., “SAT0196 Repeated rituximab infusions for the therapy of rheumatoid arthritis is not associated with increased rates of serious infections,” Annals of the Rheumatic Diseases, vol. 76, no. Suppl_2, Jun. 2017, doi: 10.1136/annrheumdis- 2017-eular.1752.

[53] J. Henry et al., “Doses of rituximab for retreatment in rheumatoid arthritis: influence on maintenance and risk of serious infection,” Rheumatology (Oxford), vol. 57, no. 3, pp. 538–547, 01 2018, doi: 10.1093/rheumatology/kex446.

[54] S. Jinno, N. Lu, S. R. Jafarzadeh, and M. Dubreuil, “Trends in Hospitalizations for Serious Infections in Patients With Rheumatoid Arthritis in the US Between 1993 and 2013,” Arthritis Care Res (Hoboken), vol. 70, no. 4, pp. 652–658, 2018, doi: 10.1002/acr.23328.

[55] A. Bortoluzzi, G. Sakellariou, G. Carrara, M. Govoni, and C. A. Scirè, “SAT0098 Risk of Hospitalization for Serious Bacterial Infections in Patients with Rheumatoid Arthritis Treated with Biologics. Analysis from The Record Study of The Italian Society for Rheumatology,” Annals of the Rheumatic Diseases, vol. 75, no. Suppl 2, pp. 700–701, Jun. 2016, doi: 10.1136/annrheumdis-2016-eular.4243.

[56] J. A. Singh et al., “Risk of serious infection in biological treatment of patients with rheumatoid arthritis: a systematic review and meta-analysis,” Lancet, vol. 386, no. 9990, pp. 258–265, Jul. 2015, doi: 10.1016/S0140-6736(14)61704-9.

[57] S. Tarp et al., “Risk of serious adverse effects of biological and targeted drugs in patients with rheumatoid arthritis: a systematic review meta-analysis,” Rheumatology (Oxford), vol. 56, no. 3, pp. 417–425, 01 2017, doi: 10.1093/rheumatology/kew442.

CHAPTER 2

Infections in rheumatoid arthritis and strategies for their prevention: A review and discussion of implications for clinical practice

Abstract

Objectives: Serious infections (SIs) in rheumatoid arthritis (RA) are common and may be life- threatening. The goal of this chapter is to present a systematic review of the present literature regarding prevalence and status of infection and SIs in RA and explore potential risk factors including anti RA medications.

Methods: A systematic review was performed that included multiple databases, viz. PubMed, Medline, Scopus, and Google Scholar. Search terms used were ‘Rheumatoid Arthritis AND infection’. Searches were limited to the title of articles, human subjects and non-juvenile arthritis and to those articles published in English.

Results: In total, 3,324 articles were found. After removing duplicates, 825 articles remained for further screening, from which 141 articles were selected. These were further assessed and 110 were then excluded because 31 articles were case reports, 35 focused on young subjects (<16 year and 44 studies focused on non-serious infection. Overall, only 31 studies met our selection criteria.

Conclusion: SIs are far more common in RA than in the general population. Corticosteroids are associated with an appreciable increase in SI risk. Most commonly used and currently favoured synthetic DMARDs confer a small or no risk, biologic DMARDs confer moderate risk in the first year of therapy and then a diminishing risk, thereafter, and higher dose biologic or combination biologic therapy should be avoided since the SI risk is unacceptably high. Undetectable mannose binding lectin (MBL) is a major risk factor for SI in RA, comparable to prednisolone.

Keywords: infections, arthritis, serious infections, risk factors

1. Introduction Rheumatoid Arthritis (RA) is a chronic, systemic autoimmune disorder. It affects over 1% of the world population and confers significant economic burdens, not only on the individual, but also society as a whole. The rheumatoid patient is exposed to many complications including infections, cardiovascular disease and malignancy [1, 2]. Among these, serious infection (SI) (infection that is life-threatening or fatal, requires hospitalization or intravenous antibiotics or results in severe disability) is of special importance, because of the immediate risk of mortality, ongoing morbidity and because of the health economic implications. Serious infections are still the number one cause of death in RA, globally [2]. The most prevalent infections in RA include bronchopulmonary, urogenital, soft tissue and skin, bone/joint sepsis and gastrointestinal infections [1]. Infections in the lung and urogenital system or generalized sepsis in a patient with RA is common and may be fatal, with incident frequencies up to ten times that in the general population [2,3]. Less commonly, infection in RA may affect the central nervous system (CNS), the cardiovascular system or the lymphatic system [4].

It is clear that in RA, there is a marked increase in rates of SI from less than one per 100PYs in the normal population to around 5 per 100PYs in RA overall [5]. Bronchopulmonary, urogenital and skin infections are the most common SIs. The main pathogens are S. pneumoniae, S. aureus, gram-negative bacilli and anaerobes [6]. Some studies have investigated diverse risk factors, such as RA disease pathophysiology, RA medication and immunodeficiency including Mannose Binding Lectin (MBL) deficiency as potential causes for this higher incidence rate [2,7-15] A failure to appreciate this de novo increase in frequency of SIs in RA can give rise to a perception of more frequent SIs in csDMARDs and bDMARDs treated RA.

In this review, we have searched the literature in order to determine and analyse (i) the extent to which RA patients are predisposed/susceptible to developing SIs (ii) the potential risk factors associated with SIs in RA patients, and (iii) whether the rate of SIs is higher in patients who are on medications, such as anti-TNF-α, and DMARDs. The goal was to identify, categorise and evaluate the main causes of SIs in RA. In addition, methods and possible strategies to minimise or prevent infection in RA and, in turn, reduce the rate of hospitalisation and out-of-hospital treatments will be discussed.

2. Methods

2.1. Search strategy and selection criteria A systematic review was performed that included multiple databases, viz. PubMed, Medline, Scopus, and Google Scholar. Search terms used were ‘Rheumatoid Arthritis AND infection’. Searches were limited to the title of articles, human subjects and non-juvenile arthritis and to those articles published in English. The search timeframe was 1996-2015. Articles were only included in this review if they investigated or discussed ‘infection in Rheumatoid arthritis’ specifically focusing on SI in patients over 16 years of age. Eleven cohorts, four reviews, one cross-sectional study, one observational prospective study, five case-control studies, five randomized controlled trials (RCT), three systematic reviews and two meta-analyses were included. Even though diverse methodologies and a relatively long timeframe mitigating against embracement of the modern biologic era were used, the advantages of inclusivity were deemed to outweigh the inconsistencies in methodology. A PRISMA chart has been constructed to show the systematic selection of the articles (Tables 2.1 and 2.2).

3. Results and discussions

3.1. Study selection In total, 3,324 articles identified through PubMed, Medline, Scopus and Google Scholar repository were found. After removing duplicates, 825 articles remained, from which upon further screening, 684 articles were culled due to one or more of the following: (i) population size was very small or the studies were not within the designated time frame; namely 1996 - 2015 (ii) heterogeneous populations, which made it difficult to identify infections pertaining explicitly to RA and (iii) the language in which the articles were published was not English.

Figure 2. 1 Prisma chart showing search results and article selection

Table 2.1. Relationship between medications and infections in RA leading to defective cell-mediated immunity Medications Bacteria Fungal Protozoan Viral Gram-positive: Corticosteroids,  Human herpes virus  Staphylococcus aureus  Candida Albicans  Measles virus - Cyclophosphamide and  Streptococcal spp  Aspergillus Spp  Varicella zoster virus - Azathioprine  Nocardia spp  Adenovirus

 Cytomegalovirus Gram-negative:  Epstein-Barr virus  E.coli  Klebsiella pneumoniae  Pseudomonas aeruginosa  Other Enterobacteriaceae -Corticosteroids  Nocardia Spp -Cyclophosphamide  Listeria monocytogenes  Histoplasma capsule -Other alkylating agents  Coccidioides immiti -Azathioprine  Salmonella Spp -Methotrexate  Mycobacterium Spp  Cryptococcus neoformans -Cyclosporin A  Pneumocystis carina -  Strongyloidiasis stercoralis -Azathioprine  Haemophilus Influenzae -Corticosteroids (high dose)  Streptococcus pneumoniae -Cyclophosphamide

Table 2.2. Summary of studies showing the rate of SIs in patients treated with synthetic and biologic DMARDs

Rate of SIs per 100PYs (synthetic Rate of SIs per 100PYs (biologic Study Group Study Design Reference DMARD) DMARD) Galloway J B, et.al. BSRBR Registry Review (UK) 3.20 (csDMARDs controls) 4.20 (all Bx DMARDs) ^ [63] (2011) Doran, et.al. (2002) RA vs Population Controls 19.23 per 100 PYs* NA [30] German RABBIT Registry Review Listing, et.al. (2005) 2.28 (csDMARDs controls) 6.15 (INX) and 6.42(ETA) [17] (GDR) Lacaille, et.al. (2008) Large RA cohort (n=27,710) 4.5 -5.5 per 100PYs NA [27] Greenberg, et.al. (2010) MTX vs controls (n=7,971) 3.1 – 3.2 per 100 PYs# NA [24] Askling J, et.al. (2006) Swedish Biologics Register - 4.5 (INX, ETA and ADA) [39] Atzeni, et.al. (2012) GISEA Registry (Italy) NA 3.18 (for INX, ADA and ETA) [48] van Dartel SAA, et.al. DREAM Registry (Netherlands) NA 2.91 (over 5 years) [14] (2012) *PYs- Patients years - For the period 1955 – 1994, rates may have declined over time #Mean follow-up (period of observation) was 1.4 years ^ ETA, INX and ADA

Nolla, et.al. (2000) reported that among RA patients, during 1990-1998, the most prevalent bacterial infections were Staphylococcus aureus and Streptococcus pneumonia [16]. Both are Gram-positive cocci. They also showed that skin infection was the principal source of infective disease in RA patients and S. aureus was among the most important pathogens for septic arthritis. S. pneumoniae was also a relevant pathogen in septic arthritis in RA patients, but it was less frequent. The majority of cases of septic arthritis in RA were mono-articular with involvement of the knee, elbow and wrist most often reported (Tables 2.1 and 2.3) [6].

Table 2.3. Serious Infection rates for diverse biologic agents (numbers per 100 PYs) Anti TNF 4.90, [95%CI 4.4-5.4, 57 trials], n = 26492, Cum. Exp. = 29429 years ABT 3.04, 95%CI 2.49-3.72,11 trials, n = 5953, Cum. Exp. = 6070 years RITUX 3.72, 8 trials, n = 2926, Cum. Exp. = 2687 years TCZ 5.45, 13 trials, n = 5547, Cum.Exp. = 4522 yrs.

TOF# 2.93, 14 trials, n = 5671, Cum.Exp. = 12,664 yrs. ETA 4.06 ADA 5.04 GOL 5.31 INX 6.11 CERT 7.59 # denotes long term extension studies, Cum. Exp. denotes cumulative exposure. TNFi = Tumour Necrosis Factor inhibitor, ABT = Abatacept, RITUX = Rituximab, TCZ = Tocilizumab, TOF = Tofacitinib. Data extracted from Strand V et.al, Arthritis Research and Therapy 2015;17:36 [64]

3.3. Risk factor categories Many studies have shown a greater than two-fold increased risk of SI in RA patients [1,2,7-20]. There are several contributing factors involved. Briefly these include:  The pathobiology of the disease itself;  Chronic comorbid conditions, such as diabetes mellitus, heart failure, lung or kidney disease, bronchiectasis and alcoholism;  Age: elderly-onset RA patients are more vulnerable;  Drug dosage, duration of treatment, and side effects: it has been shown that some drugs at high dosage and prolonged treatment therewith confer significant risk, especially in older patients with RA;

 The immunosuppressive nature of at least some of the drugs used to treat RA; these include a range of medications, such as corticosteroids, synthetic DMARDs, and biologic DMARDs;  Genetic factors: these include mannose binding lectin (MBL) deficiency, which has recently been shown to contribute significantly to serious infections in RA and is the commonest form of innate immune deficiency. In addition, hypogammaglobulinemia (common variable immunodeficiency or CVID and selective IgA deficiency or SIgAD), which are much less common but may, nevertheless, occasionally contribute to SIs. Roberts, et.al. (2015) have reported immunoglobulin deficiency after rituximab for lymphoma and rheumatoid arthritis [21,22]; and  Lifestyle factors, such as poor diet, reduced physical activity, smoking, and alcohol consumption.

3.4. The impact of medications (non-biologics) A summary of studies showing the rate of SIs in patients treated with synthetic and biologic DMARDs is shown in Table 2.2. Galloway, et.al. (2011) showed the SI incidence rates to be 42/1000 and 32/1000 patient-years for anti-TNF and csDMARDs respectively. And the risk did not differ significantly between the three agents; adalimumab, etanercept and infliximab. The risk was highest during the first six months of therapy [23]. Greenberg et.al. showed that a major risk factor for infection is the immunosuppressive therapy used. They also showed that newer therapies for RA may lead to increased rates of infection by pathogens, such as Mycobacterium tuberculosis [24]. In another study, to examine the association of methotrexate (MTX) and tumour necrosis factor (TNF) antagonists with the risk of infectious illness, Greenberg et.al. showed that MTX, TNF antagonists and prednisone at doses >10 mg daily were associated with increased risks of infections overall. Low-dose prednisone and TNF antagonists (but not MTX) increased the risk of opportunistic infections [24]. Van Dartel, et.al. (2013) showed the incidence rates for a first serious infection in patients with RA per 100 patient-years were 2.61, 3.86 and 1.66, for adalimumab, infliximab and etanercept, respectively

[14]. The impacts of other medications are discussed below.

3.5. Corticosteroids Corticosteroid (CS) use is a major contributor to SIs in RA. The effects are dose- and duration- dependent [25]. Both high dosage and the duration of CS treatment confer significant risk, especially in older patients with RA. The infection risk has been clearly shown to be dose dependent, but whether there is a minimum safe dose with respect to serious infection risk is unclear. Of considerable concern, a patient who has taken at least 5 mg of prednisolone daily for three months has a 30% chance of hospitalization due to infection [26]. Therefore, in the treatment of RA, in order to minimize the risk of an SI, the lowest possible dose of CS for the shortest possible duration should be prescribed [25]. Increasingly, with the advent of more effective synthetic and biologic DMARDs, the scope to progressively taper and switch from CS to DMARDs alone has increased.

Listing, et.al. showed that there is evidence that glucocorticoids (GCs) increase the risk of serious infections up to 4-fold in a dose- dependent manner. In addition, anti-TNF-α inhibitors increase the serious infection risk up to two-fold. The risk of infection is substantial if patients need higher dosages of GCs in addition to treatment with anti-TNF-α therapy. It was recommended that such combination therapies should avoided, if possible, especially in patients with additional risk factors such as older age or comorbid conditions [20].

3.6 Synthetic DMARDS Whether synthetic DMARDs at recommended doses contribute to infections in RA is uncertain and still a matter of conjecture. Lacaille et.al. (2008) conducted a retrospective, longitudinal study of a population-based RA cohort in British Columbia, Canada (from January 1996 to March 2003). In this study, a total of 27,710 RA patients provided 162,710 person-years of follow-up. The authors showed that 92% of patients had at least one type of mild infection and 18% had a SI. Corticosteroids were shown to be unequivocally implicated in Sis, with an adjusted rate ratio of 1.9 (CI 1.75-2.05) [27]. Importantly, these investigators showed that use of DMARDs without corticosteroids was not associated with an increased risk for SI [adjusted rate ratio of 0.92 (CI 0.85-1.00)]. They concluded that, unlike corticosteroids, synthetic DMARDs, in general, do not elevate the risk of serious infection in RA. It is, however, worth noting that, in their study, the SI rate for RA patients receiving cyclophosphamide (CYC) was 19.8 to 39.4 per 100 patient years of exposure, which is well above the rate seen for SIs in RA

overall (~ 4.4 to 5.5 per 100 Pys in their study), suggesting that some immunosuppressive DMARDs might still be an exception to the rule. CYC of course is now rarely used as a DMARD in uncomplicated RA.

Dixon et.al. (2016) [10] conducted a prospective study of participants in the British Society for Rheumatology Biologics Register (BSRBR). They compared synthetic DMARD-treated patients (n=1,354) with anti-TNF (biologic DMARD)–treated patients (n=7,664). After adjustment for baseline risk, it was concluded that anti-TNF therapy was not associated with increased risk of SI overall, compared with synthetic DMARD treatment. However, they did show that anti-TNF therapy was associated with serious skin and soft tissue infections [10]. The impact of other DMARDs on the development of infections in RA patients has also been investigated. The medication-related findings are set out below:

(i) Cyclosporine (CyA, Neoral), which is a fungal peptide, inhibits interleukin-2 and proliferation of T-cells and promotes apoptosis in macrophages. When used in combination with methotrexate for treatment of severe RA, CyA can increase the rate of urinary tract infection (UTI) [7,28].

(ii) Methotrexate (MTX, Methoblastin)-related infections are varied and appear to be dose- dependent. Because MTX is commonly used in combination with other drugs, it is often difficult to assess the contribution of MTX alone. There have been several studies which have investigated MTX and its role in the development of infection. For example, in a randomized controlled trial (RCT), incorporating 571 RA patients who were treated with a mean MTX dosage of 10.8 mg/week, without concomitant biological DMARDs, Sakai et.al. (2011) showed that MTX did not confer an increased risk for serious infections in RA patients [15]. However, there were limitations to this study, not least the lower mean dosage of MTX than that commonly used in the United States, Australia and Europe. Boerbooms et.al. (1995) in a six- year open prospective study and in a 12-month randomized double blind trial comparing MTX with AZA, showed that the infection rate during MTX treatment was higher in severe RA than in moderate RA. Once again, this highlights the likely contribution of inherent disease activity to SI risk [29].

Doran et.al. [30] reported that the hazard ratio for SIs in RA patients treated with MTX was 0.96 while Greenberg et.al. (2010), who followed a total of 7,971 patients, showed that the rate

of infection per 100 person-years was increased among MTX users. They expanded their studies to TNF antagonists and prednisolone and concluded that both MTX and Prednisolone, at doses more than 10 mg daily, were associated with increased risks for infections overall [24]. Bernatsky et.al. (2007) in a cohort of 23,733 RA patients, showed that methotrexate increases the rate of pneumonia (RR: 1.16, 95% CI: 1.02–1.33) [7] (Table 2.2).

3.7 The impact of medications (biologics) Due to their modes of action and the fact that they target cells involved in the immune system, there is an ongoing concern that these medicines may potentially increase the risk of SI in RA [31-33]. In this study, the orally active tofacitinib, a tyrosine kinase inhibitor (TKI) is included, but other TKIs in development have been excluded. We will now consider the groups of biologic agents in turn.

3.8. TNF-α Inhibitors Biologics such as Adalimumab, Certolizumab, Etanercept, Golimumab and Infliximab inhibit TNF-α and thereby modulate the inflammatory process in RA. However, TNF-α is also important for defence against common and uncommon infections. When TNF function is inhibited, there is increased risk of diverse infections. These include (i) bacterial infections such as Gram-positive and Gram-negative bacteria, Mycobacterium tuberculosis, atypical mycobacterial infection, Listeriosis monocytogenes, (ii) viral infections e.g. cytomegalovirus (CMV), and (iii) fungal infections e.g. Pneumocystis jirovecii, aspergillosis, histoplasmosis, coccidioidomycosis and cryptococcal infections [34,35].

The evidence for re-activation of Mycobacterium tuberculosis infection in RA patients has been discussed in at least two different studies [36]. All TNF inhibitors have a propensity to re- activate tuberculosis. Infliximab appears to confer greater risk than Etanercept [31,34]. There is also a significantly increased rate for Hepatitis B virus reactivation, especially when immunosuppression is diminished or withdrawn. Therefore, a combination of treatments with hepatitis B (HB) antiviral agents in conjunction with TNF inhibitors is suggested in patients with evidence of previous HB infection [35]. In addition, there is a known, albeit small, increase in risk for herpes zoster and a very small risk for leukoencephalopathy (PML) in TNF inhibitor recipients [36,37]. Historically, the greatest risk for PML has been associated with use of Natalizumab in multiple sclerosis, but the risk for TNF blockers and Rituximab in RA is not 48

negligible and will require further study to accurately quantify [31,38] and predict susceptibility. A meta-analysis in 2006 revealed that anti- TNF treatment can also increase the risk of serious pyogenic infections [8]. The German Biologics Registry investigators found the risk of serious pyogenic infection to be two-fold [8,10,17]. In contrast to the above studies, the BSRBR and the Swedish Arthritis Treatment group have reported that a non-significant relative risk ratio exists for severe infections in patients treated with TNF inhibitors [10,39]. These differences may be explained by the longevity of the studies. SIs appear to be much more frequent within the first year of usage / observation. Thus, long term follow-up studies may report lower rates of SI compared to short-term studies.

It is worth noting that van Dartel, et.al. (2012) found that Adalimumab and Infliximab conferred higher, albeit similar risks for serious infection in RA patients, whereas Etanercept conferred lower risk [14]. In addition, Trung, et.al. (2013) in their studies provided a table (Table 2.3) to categorize the risk of infection with different anti-synovitis medications. In that study, it is reported that Etanercept, Infliximab and Golimumab were associated with the highest rates of serious infection. It was shown that Etanercept, Adalimumab, Abatacept and Tocilizumab were associated with opportunistic infections and tuberculosis (TB) [40]. In a systematic review by Greenberg et.al. (2002), it was demonstrated that some of the anti-TNF medicines increased the rates of opportunistic infections while traditional immunosuppressants such as corticosteroids and synthetic DMARDs were major risk factors for serious infection in RA (Table 2.2) [2]. Moreover, Dixon et.al. (2006) in an observational study of a large cohort of RA patients (n=7664) enrolled in the BSRBR, emphasized the important role that TNF has in host defence in the skin and soft tissue [10]. In their study, patients who were treated with anti TNF- α agents, as compared to synthetic DMARDs, developed more serious skin and soft tissue infections. However, importantly, they found that the overall risk of serious infection for anti- TNF medicines compared to synthetic DMARDs was the same in both groups [10].

3.9. Abatacept (ABT), Rituximab, Anakinra, Tofacitinib and Tocilizumab ABT safety has been evaluated in several long-term extension (LTE) studies (duration usually 2-3 years). Within this timeframe, in respect to SIs, ABT performs well with SI rates of 1.6 to 3.6 per 100 PYs of treatment in age unstratified RA recipients [41,42]. Given that up to 60% of these patients were also taking corticosteroids in doses not always clearly defined, the rates are low for the most part and somewhat lower than for most other biologic agents (Table 2.3). The 49

elderly is more vulnerable as is true in respect to SIs in general and especially after there has been an antecedent hospitalization for infection, whereupon rates of 26.5 per 100 PYs apply for ABT and 36.1 for ETA [42,43]. Lahaye, et.al. found that SI rates in Abatacept recipients rose progressively from 1.73 per 100PYs in persons under 50 to 4.65 in persons 50-64, 5.90 in persons 65-74 and 10.38 per 100PYs in persons equal to or greater than 75 years of age [43]. Thus, whilst relatively safe in the young and up to extended middle age, the SI rates rose concerningly for ABT in those over 65 years of age and especially when there has been an antecedent hospitalization for an infection (Table 2.3).

For Rituximab, Tocilizumab and Tofacitinib, the rates of SI are comparable to those reported for all TNF inhibitors. However, it should be noted that the cumulative exposure for most of these agents, like the TNF inhibitors is limited and mostly does not exceed 2 years. Furthermore, not enough additional data is available to evaluate associated SI risk factors in these cohorts. For example, a breakdown for age, corticosteroid dosage and important comorbidities such as diabetes, neutropoenia and lymphopenia is not available sufficiently often to allow these parameters to be taken fully into account in respect to their independent or additive effect on SI risk.

In the case of Infliximab (INX) and tocilizumab (TCZ) there is some data, which suggests that SIs are dose dependent with higher rates seen with higher doses [44]. This has already been referred to in respect to INX. For TCZ, SI rates of 3.4 per 100 patient’s years (100PYs) were observed for comparator groups, 3.5 per 100 PYs for TCZ 4 mg/Kg and 4.9 per 100 PYs for TCZ 8 mg/Kg [45]. In contrast, for Rituximab (RITUX), the SI rates for 500 mg x 2 versus 1000 mg x2 at 24-week intervals were similar at 2.62 and 1.96 SIs per 100PYs [46]. There is a limitation in this systematic review, and it is not clear if discussed doses are adjusted for Body mass index (BMI) or not.

Salliot, et.al. [47] investigated the risk of SIs during treatment of RA with rituximab, abatacept and anakinra. SI frequencies were investigated using meta-analyses of randomized placebo- controlled trials. It is important to remember that this approach inevitably is short term due to the design of the trials. Moreover, sicker patients are often excluded. Nevertheless, no significant increase in the risk for SIs attributable to these biologics was observed. The authors concluded that, based on these randomized placebo-controlled trials, rituximab, abatacept and

anakinra have a relatively good safety profile for SIs. However, an increased risk for SIs was observed for high doses of anakinra (⩾100 mg per day) in patients with comorbidities.

3.10. Risks associated with combination therapies It is now common practice to combine synthetic DMARDs with biologic DMARDs, since efficacy is greater. Some studies have shown that synthetic DMARDs in combination with anti- TNF-α increase the rate of SIs. For example, Atzeni, et.al. (2012) in a case control study examined 2,769 patients with long term RA [48]. Treatment with corticosteroids and other synthetic DMARDs in combination with various anti-TNF agents, viz. Infliximab (INX), Adalimumab (ADA) and Etanercept (ETN), was investigated. The authors found that the risk of SI was significantly different across these medication groups (p<0.0001). In these patients, the following factors were identified as significant infection predictors: (i) The concomitant use of corticosteroids (p<0.046 with hazard ratio (HR) of 1.849) (ii) concomitant DMARD treatment during anti- TNF therapy (p=0.004 with HR of 2.178) (iii) advanced age at the start of anti-TNF treatment (p<0.0001 with HR of 1.03) and (iv) the use of INX or ADA rather than ETN ( p<0.0001with HR 4.291 for INX vs ETA and p=0.023 with HR 1.942 for ADA vs ETA). In this study the authors also found that treatment with anti-TNF was associated with a small, but statistically significant risk of SI (HR of 1.03 and P < 0.0001). In Atzeni et.al’s study, disease duration and the disease severity score were not found to be predictive of serious infection [48].

In a systematic study by Campbell, et.al. (2011), the effect of tocilizumab (TCZ), in combination with MTX, in patients with RA was investigated [49]. The researchers concluded that this combination treatment for RA is associated with a small, but significantly increased risk of adverse effects and infections. Their meta-analysis revealed that tocilizumab 8 mg/kg compared with controls increased the risk of infection. This risk is comparable with other biologic agents, although the risk of serious infection may be less than that for TNF antagonists.

Perhaps more so than any other biologic agent, the capacity of IL-6 antagonists to markedly reduce CRP further compounds the difficulty in recognizing serious infection, since great reliance is usually placed on the CRP concentration when determining the probability of an SI in an unwell rheumatoid patient. Such delays may adversely affect patient outcomes.

3.11. Tuberculosis (TB) and non-tuberculous mycobacterial (NTM) infections In a recent meta-analysis conducted by Winthrop, et.al. (2015), both TB and NTM infections were shown to be increased in patients with RA who have been treated with a range of biologics [36,50] These include Infliximab, Etanercept and Adalimumab, which target TNF-α, as well as Rituximab, which targets CD20 receptors on the surface of B cells. All these agents have been shown to re-activate TB and predispose to NTM infections, albeit at different rates. Infliximab was implicated in TB and NTM infections (11 and 7 cases respectively). In contrast, in this meta-analysis Abatacept was not shown to predispose to TB or NTM infections. It remains important to carefully screen for latent TB, both clinically and otherwise (Mantoux skin testing, Quantiferon GOLD testing) and where necessary, to treat these conditions appropriately before initiating bDMARDs in RA.

3.12. Serological and other laboratory parameters that influence SI risk Diverse cellular and serological abnormalities are known to increase susceptibility to infection. These include neutropoenia, especially in the context of Felty’s syndrome, where high disease activity is often a factor as well, lymphopenia, immunoglobulin deficiencies (innate and acquired) and terminal complement deficiencies, although the frequency of Ig and terminal complement deficiency is low or very low respectively. deficiency is far more common with frequencies in the order of 5-8% in the population in general and 8 - 15% in rheumatoid populations. The prevalence of serious infection in ARAD participants was 2.92 % of all patient visits. The rate in other studies such as the study by Doran et al. in Minnesota US reported 9.6 infections/100 person-years (1). The reasons for this difference are partially due to the different study designs, settings and therapeutic guidelines [2]. Periodontal infection occurs due to almost twenty different bacterial species and occurs about two-fold higher in RA patients. In addition, the prevalence of moderate to severe periodontitis in RA patients is almost 51% which is more than age and gender matched patients with osteoarthritis (26%) [1]. Concurrent diseases in patients with RA include depression (15%), asthma (6.6%), cardiovascular events (6%), cancer (4.5%), and chronic obstructive pulmonary disease 3.5%.

3.13. Mannose Binding Lectin (MBL) and other immune deficiencies Mannose Binding Lectin (MBL) deficiency is implicated in a variety of infections in neonates and children, but less so in otherwise healthy adults [11,51-53] Mannose Binding Lectin (MBL) 52

is a component of the innate immune system. It is a carbohydrate binding protein produced by the liver and is involved in innate immunity [54]. Structurally, this molecule comes in trimer and tetramer forms and binds to the glycan on the pathogen’s cell surface mannose receptor. Generally, immune-compromised patients and patients with chronic diseases or impaired adaptive immune systems including those with Mannose Binding Lectin (MBL) deficiency have increased risks of serious infection [11,55-57] Mannose Binding Lectin (MBL) has also been shown to have roles in manifestations of RA disease and the development of other complications of RA, such as cardiovascular disease [58].

In a recently reported study of risk factors for SIs in RA, both undetectable Mannose Binding Lectin (MBL) and CS use (prednisolone at doses of 5 mg per day or more) were shown to confer a 4-5-fold increased risk for SIs [53]. This takes on greater importance when it is remembered that up to 15% of RA patients have undetectable Mannose Binding Lectin (MBL) and that rates of CS use in RA, although they vary a great deal from centre to centre are still high despite the availability of more efficacious DMARDs (up to 70%) [53]. In fact, apart from severe neutropoenia, such as in Felty’s syndrome for example, no other laboratory marker appears to confer greater SI risk then undetectable Mannose Binding Lectin (MBL). Common variable immunodeficiency (CVID) is estimated to affect up to 1 in 25,000 individuals and can be associated with auto-immune diseases including RA [59-61]. The exact risk associated with CVID or its various disease expressions such as panhypogammaglobulinaemia, selectively reduced immunoglobulins (e.g. IgA deficiency) and IgG subset deficiency in RA is unknown, but given that these deficiencies are much less frequent than undetectable Mannose Binding Lectin (MBL), they are likely to be relatively less important clinically.

Selective IgA deficiency or SIgAD, which is the most common of these immunoglobulin deficiencies, occurs in less than 1 in 100 persons of Arabic descent and in less than 1 in 800 Caucasians in the UK. Although increased rates of severe respiratory tract infections are observed in SIgAD persons, compared to unaffected controls (3-fold increased risk), life- threatening infections were not recorded in this group [62] Elsewhere, risk factors predisposing to the development of hypogammaglobulinemia and infections post-rituximab treatment have been reported [63]. Terminal complement components C5 - C9, otherwise referred to as the membrane attack complex also predispose to recurrent infection, especially with encapsulated organisms, such as Neisseria, but do not appear to associate strongly with autoimmune

diseases and are relatively rare in Caucasians, although not in Afro-Americans and probably not in native Africans.

3.14. Implications for Clinical Practice The treating clinician needs to consider the following when choosing therapeutic agents for patients with RA who are at risk for SIs:

3.14.1. Age

SIs are substantially increased in persons of advanced age -for example in a large USA Medicare beneficiaries’ cohort, the SI rate in those over 65 was 14.2 per 100PYs compared to 4.8 in those less than 65 years of age [43]. A recently reported study by one of the authors indicates that the risk of an SI increases by 19% for every 5 years increase in age and by 41% for every 10-year increase in age [53]. The prescribing clinician should consider, the differing relative risks for SIs when prescribing DMARDs for the old and the very old rheumatoid patient.

3.14.2. Corticosteroid (CS) Use and Dosage

In RA patients, the SI risk is appreciably higher in recipients of CS. Furthermore, this risk is most likely dose dependent. For example, in one study, a daily dose of 10 mg of Prednisolone or more was associated with an odds ratio (OR) for an SI of 4.70, whereas a dose of 1-4.5 mg per day was associated with an OR of 2.57 [9]. Initial use of CS at first presentation may be unavoidable, but scope to wean the dose of CS should be explored, as a matter of priority, once a satisfactory response to synthetic DMARD or biologic therapy has been achieved. The minimum safe dose of CS is unknown. Indeed, in respect to SIs, there may be no safe minimum, but until more definitive data is available, a dose of 3 mg/day may represent a reasonable compromise target for maintenance of lower disease activity and at the same time minimization of SI risk [64,65].

3.14.3. Doses of biologic agents

The dose of any therapeutic agent should be periodically reviewed. For some biologic agents, where there is dose flexibility, lower SI risks have been convincingly demonstrated with lower doses of the biologic agent. For example, for Infliximab (INX), a 3 mg/Kg dose confers less risk than 6 mg/Kg and for Adalimumab (ADA), 40 mg every other week (EOW) confers less 54

risk than 40 mg qw. Similar observations have been made for Tocilizumab (TCZ) where 4 mg/kg was found to confer less SI risk than 8 mg/ kg [46]. Where SI risk is a major concern and disease control will allow, reduced doses of DMARDs in general, including bDMARDs, should be considered or monotherapy with a bDMARDs should be preferred.

3.14.4. Vaccination

Pneumonias and lower respiratory tract infections in general are the most common SIs in all RA patients irrespective of biologic or synthetic DMARD therapy. Pneumococcal vaccination should be advised, unless contra-indicated, and follow-up post- vaccination serology performed to confirm adequate immunity. When it becomes more widely available/accessible, the new subunit zoster vaccine (Shingrix) should be considered, especially in those most at risk due to age.

3.14.5. Comorbidities related and unrelated to RA

Amongst related disorders, consider Felty’s syndrome and other conditions that may give rise to Neutropoenia. Consideration should also be given to innate immune deficiencies such as Mannose Binding Lectin (MBL) deficiency and Hypogammaglobulinemia. Undetectable Mannose Binding Lectin (MBL) concentrations carry a considerable risk for SIs in RA (OR=4.67) comparable to 10 mg of prednisolone daily [3,9]. Since pneumonias are more often fatal in Mannose Binding Lectin (MBL) deficient persons and since 8-15% of RA patients are Mannose Binding Lectin (MBL) deficient (serum concentrations less than 50 ng/mL), there is an even stronger case for pneumococcal vaccination in those with RA with undetectable Mannose Binding Lectin (MBL). The treating clinician should consider determining the Mannose Binding Lectin (MBL) concentration in advance of commencing CS, csDMARDs or bDMARDs therapy, as this information taken together with age and CS usage will inform decision making in respect to the nature and risks of therapy.

4. Conclusion In conclusion, SIs are far more common in RA than in the general population, CS are associated with an appreciable increase in SI risk (5 fold at doses of 10 mg per day or more), most commonly used and currently favoured synthetic DMARDs confer a small or no risk, biologic DMARDs confer moderate risk in the first year of therapy and then a diminishing risk thereafter, and higher dose biologic or combination biologic therapy should be avoided since 55

the serious infection risk is unacceptably high. Combinations of CS and bDMARDs or of csDMARDs and bDMARDs should be used with caution in those with a track record for one or more SIs and perhaps also in the elderly. Undetectable Mannose Binding Lectin (MBL) is a major risk factor for SI in RA, comparable to Prednisolone 10 mg per day or more and measurement thereof will inform SI risk stratification.

[3] T. R. A. C. of G. Practitioners, “RACGP - Rheumatoid arthritis.” https://www.racgp.org.au/afp/2010/september/rheumatoid-arthritis/ (accessed Jun. 14, 2020).

[4] J. Dequeker, “Arthritis in Flemish paintings (1400-1700),” Br Med J, vol. 1, no. 6070, pp. 1203–1205, May 1977, doi: 10.1136/bmj.1.6070.1203.

[9] A. J. Silman and J. E. Pearson, “Epidemiology and genetics of rheumatoid arthritis,” Arthritis Res, vol. 4, no. Suppl 3, pp. S265–S272, 2002, doi: 10.1186/ar578.

[10] M. Bullot and L. Woyzbun, A picture of rheumatoid arthritis in Australia. Canberra: Australian Institute of Health and Welfare, 2009.

[12] J. Sokolove, “Rheumatoid Arthritis Pathogenesis and Pathophysiology,” 2018, pp. 19– 30.

[16] D. van der Woude et al., “Gene-environment interaction influences the reactivity of autoantibodies to citrullinated antigens in rheumatoid arthritis,” Nat. Genet., vol. 42, no. 10, pp. 814–816; author reply 816, Oct. 2010, doi: 10.1038/ng1010-814.

[17] H. Mahdi et al., “Specific interaction between genotype, smoking and autoimmunity to citrullinated alpha-enolase in the etiology of rheumatoid arthritis,” Nat. Genet., vol. 41, no. 12, pp. 1319–1324, Dec. 2009, doi: 10.1038/ng.480.

[18] J. E. Hart, F. Laden, R. C. Puett, K. H. Costenbader, and E. W. Karlson, “Exposure to traffic pollution and increased risk of rheumatoid arthritis,” Environ. Health Perspect., vol. 117, no. 7, pp. 1065–1069, Jul. 2009, doi: 10.1289/ehp.0800503.

[19] H. Källberg et al., “Alcohol consumption is associated with decreased risk of rheumatoid arthritis: results from two Scandinavian case-control studies,” Ann. Rheum. Dis., vol. 68, no. 2, pp. 222–227, Feb. 2009, doi: 10.1136/ard.2007.086314.

[20] Ł. Kłodziński and M. Wisłowska, “Comorbidities in rheumatic arthritis,” Reumatologia, vol. 56, no. 4, pp. 228–233, 2018, doi: 10.5114/reum.2018.77974.

[21] H. B. Tenstad, A. C. Nilsson, C. D. Dellgren, H. M. Lindegaard, K. H. Rubin, and S. T. Lillevang, “Use and utility of serologic tests for rheumatoid arthritis in primary care,” p. 7, 2020.

[22] N. A. Accortt et al., “Impact of Sustained Remission on the Risk of Serious Infection in Patients With Rheumatoid Arthritis,” Arthritis Care Res (Hoboken), vol. 70, no. 5, pp. 679–684, May 2018, doi: 10.1002/acr.23426.

[23] K. P. Liang, K. V. Liang, E. L. Matteson, R. L. McClelland, T. J. H. Christianson, and C. Turesson, “Incidence of noncardiac vascular disease in rheumatoid arthritis and relationship to extraarticular disease manifestations,” Arthritis Rheum., vol. 54, no. 2, pp. 642–648, Feb. 2006, doi: 10.1002/art.21628.

[24] S. Li, Y. Yu, Y. Yue, Z. Zhang, and K. Su, “Microbial Infection and Rheumatoid Arthritis,” J Clin Cell Immunol, vol. 4, no. 6, Dec. 2013, doi: 10.4172/2155- 9899.1000174.

[25] K. Thomas and D. Vassilopoulos, “Individual Drugs in Rheumatology and the Risk of Infection,” in The Microbiome in Rheumatic Diseases and Infection, G. Ragab, T. P. Atkinson, and M. L. Stoll, Eds. Cham: Springer International Publishing, 2018, pp. 445– 464.

[26] G. J. Carroll et al., “Undetectable Mannose Binding Lectin and Corticosteroids Increase Serious Infection Risk in Rheumatoid Arthritis,” J Allergy Clin Immunol Pract, vol. 5, no. 6, pp. 1609–1616, Dec. 2017, doi: 10.1016/j.jaip.2017.02.025.

[27] I. C. Olsen, E. Lie, R. Vasilescu, G. Wallenstein, S. Strengholt, and T. K. Kvien, “Assessments of the unmet need in the management of patients with rheumatoid

arthritis: analyses from the NOR-DMARD registry,” Rheumatology (Oxford), vol. 58, no. 3, pp. 481–491, Mar. 2019, doi: 10.1093/rheumatology/key338.

[28] A. Richter et al., “Impact of treatment with biologic DMARDs on the risk of sepsis or mortality after serious infection in patients with rheumatoid arthritis,” Ann. Rheum. Dis., vol. 75, no. 9, pp. 1667–1673, Sep. 2016, doi: 10.1136/annrheumdis-2015-207838.

[29] M. A. Othman, W. S. W. Ghazali, N. K. Yahya, and K. K. Wong, “Correlation of Demographic and Clinical Characteristics with Rheumatoid Factor Seropositivity in Rheumatoid Arthritis Patients,” Malays J Med Sci, vol. 23, no. 6, pp. 52–59, Nov. 2016, doi: 10.21315/mjms2016.23.6.6.

[30] J. A. Zamora-Legoff, M. L. Krause, C. S. Crowson, J. H. Ryu, and E. L. Matteson, “Risk of serious infection in patients with rheumatoid arthritis-associated interstitial lung disease,” Clin. Rheumatol., vol. 35, no. 10, pp. 2585–2589, Oct. 2016, doi: 10.1007/s10067-016-3357-z.

[31] J. Morel et al., “Risk factors of serious infections in patients with rheumatoid arthritis treated with tocilizumab in the French Registry REGATE,” Rheumatology (Oxford), vol. 56, no. 10, pp. 1746–1754, 01 2017, doi: 10.1093/rheumatology/kex238.

[32] J. H. Salmon et al., “Predictive risk factors of serious infections in patients with rheumatoid arthritis treated with abatacept in common practice: results from the Orencia and Rheumatoid Arthritis (ORA) registry,” Ann. Rheum. Dis., vol. 75, no. 6, pp. 1108– 1113, Jun. 2016, doi: 10.1136/annrheumdis-2015-207362.

[33] H. Kawashima et al., “Long-term use of biologic agents does not increase the risk of serious infections in elderly patients with rheumatoid arthritis,” Rheumatol Int, vol. 37, no. 3, pp. 369–376, 2017, doi: 10.1007/s00296-016-3631-z.

[34] S. Subesinghe, A. I. Rutherford, R. Byng-Maddick, K. Leanne Hyrich, and J. Benjamin Galloway, “Recurrent serious infections in patients with rheumatoid arthritis-results from the British Society for Rheumatology Biologics Register,” Rheumatology (Oxford), vol. 57, no. 4, pp. 651–655, 01 2018, doi: 10.1093/rheumatology/kex469.

[35] C. Baradat, Y. Degboé, A. Constantin, A. Cantagrel, and A. Ruyssen-Witrand, “No impact of concomitant methotrexate use on serious adverse event and serious infection risk in patients with rheumatoid arthritis treated with bDMARDs: a systematic literature review and meta-analysis,” RMD Open, vol. 3, no. 1, p. e000352, Feb. 2017, doi: 10.1136/rmdopen-2016-000352.

[36] C. A. F. de Andrade, “Trying to find an answer for an old question: does Rituximab increase the risk of serious infections in patients with rheumatoid arthritis?,” Rheumatology (Oxford), vol. 57, no. 9, pp. 1505–1506, Sep. 2018, doi: 10.1093/rheumatology/kex439.

[37] D. A. Pappas et al., “SAT0196 Repeated rituximab infusions for the therapy of rheumatoid arthritis is not associated with increased rates of serious infections,” Annals of the Rheumatic Diseases, vol. 76, no. Suppl_2, Jun. 2017, doi: 10.1136/annrheumdis- 2017-eular.1752.

[38] J. Henry et al., “Doses of rituximab for retreatment in rheumatoid arthritis: influence on maintenance and risk of serious infection,” Rheumatology (Oxford), vol. 57, no. 3, pp. 538–547, 01 2018, doi: 10.1093/rheumatology/kex446.

[39] S. Jinno, N. Lu, S. R. Jafarzadeh, and M. Dubreuil, “Trends in Hospitalizations for Serious Infections in Patients With Rheumatoid Arthritis in the US Between 1993 and 2013,” Arthritis Care Res (Hoboken), vol. 70, no. 4, pp. 652–658, 2018, doi: 10.1002/acr.23328.

[40] A. Bortoluzzi, G. Sakellariou, G. Carrara, M. Govoni, and C. A. Scirè, “SAT0098 Risk of Hospitalization for Serious Bacterial Infections in Patients with Rheumatoid Arthritis Treated with Biologics. Analysis from The Record Study of The Italian Society for Rheumatology,” Annals of the Rheumatic Diseases, vol. 75, no. Suppl 2, pp. 700–701, Jun. 2016, doi: 10.1136/annrheumdis-2016-eular.4243.

[41] J. A. Singh et al., “Risk of serious infection in biological treatment of patients with rheumatoid arthritis: a systematic review and meta-analysis,” Lancet, vol. 386, no. 9990, pp. 258–265, Jul. 2015, doi: 10.1016/S0140-6736(14)61704-9.

[42] S. Tarp et al., “Risk of serious adverse effects of biological and targeted drugs in patients with rheumatoid arthritis: a systematic review meta-analysis,” Rheumatology (Oxford), vol. 56, no. 3, pp. 417–425, 01 2017, doi: 10.1093/rheumatology/kew442.

CHAPTER 3

Descriptive analysis of the infection status in rheumatoid

arthritis patients (using ARA data)

Abstract

Objectives: To conduct a descriptive analysis of the type and frequency of self-reported infections in rheumatoid arthritis (RA) based upon reports to the Australian Rheumatology Association Database (ARAD). These include the effects of anti-RA medications in the development of infections across various organs.

Methods: ARAD reports (patients’ responses to questionnaires) from 2001 to 2014 were examined in respect to demographic and treatment categories. Observed differences were subjected to descriptive statistical appraisal.

Results: Based on this analysis, the mean age in RA is 61.47 years and, in the group taking csDMARDs, it is 59.24 years and, in the group taking bDMARDs, it is 62.62 years. Also, patient groups who were taking csDMARDs alone and bDMARDs alone were comparable based on risk factors, such as taking prednisolone, smoking or alcohol consumption. Finally, in comparison to bDMARDs, taking csDMARDs alone was significantly associated with higher rate of infection in a few organs, such as lung, urinary system, and GIT.

Conclusion: Compelling evidence suggests that RA can increase the risk of infection and potentially serious infection and that different medications are potentially associated with this risk. The Australian RA population in ARAD shows that risk factors, such as smoking, can play a role in the development of serious infection. Although it seems that csDMARDs alone is connected to more rates of infection than bDMARDs alone, the difference is only significant in a few types of infections. The findings in this analysis indicate that smoking is a likely contributor to increased infection risk in RA. The Australian RA population in ARAD shows that risk factors, such as smoking, can play a role in the development of serious infection. Although it seems that csDMARDs alone is associated with higher frequencies of infection than bDMARDs alone, the difference is only statistically significant for a few types of infections. Accordingly, these apparent differences require closer scrutiny. Importantly, the findings contrast with those reported in most registries, where bDMARDs use is associated with higher rates of infection or at least serious infection and at least in the first year of treatment. The different definitions

applied for infection in ARAD and the very long follow-up may account for the differences observed.

1. Introduction

Rheumatoid arthritis (RA) as a chronic multisystemic, immuno-inflammatory disease is a common disease affecting millions of people worldwide [1]. In developed countries, RA affects 0ꞏ5–1ꞏ0% of adults, with 5–50 per 100 000 new cases annually[2]. The female to male ratio in this disease is more than three to one. The risk of RA increases with age, perhaps pointing to loss of tolerance as the immune system undergoes age-related loss of antigen discriminatory capacity. Genetic and environmental risk factors also contribute to the risk of RA [3].

There are some well-known risk factors in this disease, including genetic susceptibility, gender, age, smoking, infectious agents, hormonal factors and ethnic factors[4]. Roughly 50% of the risk for RA is attributable to genetic factors. Around 30 genetic loci have been implicated in RA. These suspicious genes have been classified, however, the pathogenesis of their influence in developing RA is still unclear[4]. Smoking is also a main environmental risk factor. Smoking-related tissue necrosis is thought to be of influence in the onset of excessive inflammation and immune response to self-antigens [3]. Age and sex can also play aetiopathogenetic roles.

The incidence, severity, and the outcome of the disease show inconsistencies between diverse ethnical-origin units, which is related to socioeconomic levels, as well as genetic and environmental factors. For example, patients in underdeveloped countries have poorer prognosis. They demonstrate a more severe clinical course due to limited access to medical care and medication, amongst other factors. Studies on RA has revealed that various genetic and environmental factors can influence the disease in diverse ethnical groups[5].

RA signs and symptoms include persistent synovitis and systemic inflammation due to autoantibodies particular to rheumatoid factor (RF) and antibodies to certain peptides. The typical symptoms at onset are symptoms of synovitis (pain, swelling, loss of function, including stiffness, restricted motion, and possible heat and redness in joints, if severe), which are most often in a symmetrical pattern and sometimes accompanied by systemic symptoms, such as lethargy /malaise, weight loss and sometimes fever[6]. Clinical onset of this disease is generally symmetrical involvement of the small joints, pain, morning stiffness, and limitation of movement for more than one hour. RA may also involve

any joint, but most frequently it involves the meta-carpophalangeal (MCP) joints, the proximal interphalangeal (PIP) joints, the wrists, the metatarsophalangeal (MTP) joints and the knee joints[6]. Articular symptoms are usually symmetric and systemic pattern. The large joints which may be involved include the shoulders, elbows, knees and ankles. The small joints include the MCP, PIP, MTP, thumb interphalangeal joint and wrists[7]. The clinical presentation of RA varies, although an insidious onset of pain accompanied by symmetric swelling of the small joints is the most common symptom cluster at the outset [7]. Rheumatoid arthritis also increases the risks of several other comorbidities including cardiac disease, depression, lymphoma and other malignancies [7]. Complications are not limited to the joints and can involve extra-articular tissues including vasculitis and ophthalmic, neurologic, and cutaneous complications[8]. RA is not directly life-threatening, but uncontrolled active rheumatoid arthritis can also lead to joint damage, decreased quality of life, disability, and cardiovascular comorbidities[9].

Complications are not limited to the joints and can involve extra-articular tissues, including the serosal surfaces (pleural and pericardial effusions), bone marrow ( anaemias and cytopaenias) the lungs ( interstitial lung disease ), blood vessels (vasculitis), the eyes (episcleritis and scleritis with blindness due to occasional perforation), neurologic and cutaneous complications[9]. RA is not directly life-threatening but uncontrolled active rheumatoid arthritis can be very debilitating, reduce the quality of life, contribute to substantial disability, and contribute to cardiovascular comorbidities [10]. Importantly, infection is the commonest cause of death in RA; the disease, its treatment and probably co-existent immunodeficiencies all likely contribute to this increased risk [10].

The pathophysiology of RA is yet to be elucidated completely, but it seems that molecular and cellular pathways of inflammation with involvement of both B cells and T cells play important roles. Distinct autoantibodies are always present in the sera of patients[10]. Rheumatoid factor (RF), both IgM rheumatoid factors (IgM-RF) and IgG rheumatoid factors (IgG-RF) are present in different stages of RA pathogenesis. The IgM rheumatoid factors (IgM-RF) are the main RF class found in RA and they can be detected in 60–80% of established cases of RA and 50–60% of RA patients in the early stages of the disease[11]. This implies that RF is probably an outcome of non-specific immune activation[11].

Modern treatment of RA has been substantially improved by the introduction of biologic therapies. For most patients, these drugs represent an effective and safe management strategy; however, serious infections connected with biologic therapies are a major concern for both patients and clinicians. Registry data from the UK and Sweden have shown an augmented risk of serious infection in new anti-TNF starters, especially in the first 6–12 months of treatment [12]. Infections are usually due to the same organisms seen commonly in the general population, and a small number of infections are due to opportunistic infections (OI)[12]. Treatment in RA is usually based on immune suppression through csDMARDs) or (bDMARDs)[13].

1.1. DMARDs

Disease-modifying anti-rheumatic drugs (DMARDs) are drugs which reduce the level of inflammation, slow joint damage and decrease the systemic effects of RA. There are three major groups; these include:  conventional synthetic DMARDs (csDMARDs),  targeted synthetic DMARDs (csDMARDs), and  biological DMARDs (bDMARDs)[14]. csDMARDs alone or Conventional synthetic DMARDs include: 1- Methotrexate (oral or parenteral), 2- Hydroxychloroquine, 3- Sulphasalazine, 4- Leflunomide, 5- Azathioprine, 6- Cyclosporin.

1.2. bDMARDs

These are engineered medications and are designed to regulate the immune response. Hereditarily-engineered proteins initiating from human genes form biologic drugs targeting the specific portions of the immune system that fuel inflammation. csDMARDs alone, such as methotrexate, are less targeted [15].

Biologics are usually used singly or in combination with other non-biologics. What distinguishes biologics, besides how they work and what they target, is their makeup, how they are delivered, and some risks, although all of them probably confer an increased risk for infection. Different groups of biologics include: Tumor necrosis factor inhibitors (TNF- Inhibitors) which block tumor necrosis factor, one of the chemical messengers of inflammation

that drives joint inflammation and destruction. Interleukin-1 (IL-1) blocker ( for example anakinra) which blocks IL-1, the factor with a major role in inflammation, B-cell inhibitor (rituximab), T-cell inhibitors (abatacept), humira/adalimumab, etanercept or brenzys /etanercept, remicade/infliximab, actemra/tocilizumab, simponi/golimumab [15].

Others with different mechanisms of action include actemra/tocilizumab (a monoclonal antibody directed against IL-6 receptor), Interleukin-1 (IL-1) receptor antagonist (for example, anakinra), which blocks IL-1, B cell depletors (rituximab) and T-cell inhibitors (abatacept). A new family of Jak inhibitors, which can be taken orally, is now emerging and is in clinical use with growing uptake. These include tofacitinib and baricitinib.

It should be noted that bDMARDs are not used concomitantly because of concerns regarding still higher rates of serious infection, however, the evidence base underpinning this fear is not strong and newer agents with low infection propensity have not been combined with older agents and studied rigorously in clinical trials. This section will cover the demographic characteristics of RA, SI, different types of infections and their severities. In addition, there will be a descriptive assessment of the association between various modalities of treatment in RA and the severity of different types of infection. Through a comprehensive descriptive analysis, potential associations and other relationships may emerge. Later in this section observed differences and potential relationships will be evaluated statistically and discussed in detail.

1.3. Aims and Objectives

The aim of this study is to increase knowledge about the pattern of RA in the Australian population and to determine the frequency and significance of self-reported infections. Specific objectives in this section include:

• Describe the demographic characteristics of the ARA database and to report the type, severity and frequency of self-reported SIs as well as the relevant potential risk factors for infection. • Describe the different types of infections in RA and their association with the major treatment modalities, csDMARDs alone or bDMARDs alone. • Provide essential tools for other researchers from other parts of the world to perform similar analyses and compare demographic characteristics in different parts of the world.

• Discuss potential associations between different available modalities for treatment and different types of infections and their severity.

2. Methods

2.1. Data Collection

The data were collected from the ARAD, in which a cohort of 3569 RA patients (960 males and 2609 females) who had completed related questionnaires 28176 times (during 2001-2014), were investigated for the development of infections. Among the 3569 patients, 459 patients were eliminated because they had filled out the questionnaire only once. We were left with 3110 patients. Eight duplicates were eliminated, leaving 27709 visits from 3110 patients. All these visits were examined to capture self-reported infections in different organs and the medications that were being taken at the time.

2.2. Statistical Analysis

Amongst the 3110 Rheumatoid Arthritis patients who had taken part in the study and had filled in the questionnaire more than once, the central tendency for age and sex distribution was calculated. In the first step all data were entered in excel. Single, faulty and duplicate reports were eliminated. Then data was divided into two groups of patients who were taking either just csDMARDs alone and patients who were taking just bDMARDs alone. Overall, 1653 visits from 405 patients applied to those taking csDMARDs alone and 323 visits from 80 patients applied to those taking bDMARDs alone. All the patients who were taking both csDMARDs and bDMARDs concurrently were eliminated from the analysis at this stage. Both csDMARDs- alone and bDMARDs-alone participants and the overall RA population were examined closely and compared in respect to sex distribution, age distribution, smoking history, alcohol consumption, and different organ infection. Possible differences were tested with the chi- squared test and the Fisher test, wherever it was applicable.

3. Results and discussions

3.1. Demography of whole RA population

The amount and frequency of smoking [16], alcohol consumption [17], T1DM, T2DM[18] and prednisolone consumption[19] all can play a role in the incidence of infection among RA 70

patients. In addition, a patient’s age and sex can have a different distribution among RA patients and the normal population. Therefore, in the following tables (tables 3.1 to 3.6), these differences are explored and compared.

The mean and median were used as points of estimate and accuracy was measured by Standard Error (Table 3.1). This shows the distribution of age, the number of cigarettes smoked and the duration of smoking, as well as the amount of alcohol consumed.

Table 3.1 Comparison demography of RA (Data collected from ARAD)

Variable Mean SD Median Age in RA 61.48 12.31 63.00 Number of cigarettes smoked among smokers 14.89 13.23 15.00 Duration of smoking among smokers 17.26 13.95 16.00 Alcohol Consumption 1.32 0.47 1.00 units among alcohol consumers

A sample of 1653 visits, pertaining to 405 patients who were receiving csDMARDs alone and 80 patients who were receiving bDMARDs alone, was examined for the impact of several predictor variables: age, gender, alcohol consumption, smoking history and prednisolone intake. The results were compared through chi-squared tests, in the next stage. This is described below.

3.2. Demography of patients taking purely bDMARDs

Only eighty patients were just taking bDMARDs alone during the period of this study (2001 to 2014) without any concurrent csDMARDs. Among this number, 63.75 % were female and 36.25% were male. Smoking as a risk factor for infection was also measured in this population [16].

Table 3.2 Central tendency among patients who received bDMARDs alone Variable Mean SD Median Age 62.62 12.57 62.50 Average number of cigarettes smoked by smokers 25.90 21.67 20.00 Smoking duration (yrs) among smokers 21.53 12.26 22.00

The mean age for the patients receiving bDMARDs alone was 62 years old (Table 3.2). 64 % of the patients who were taking bDMARDs alone were female and 36% were male (Table 3.3).

Table 3.3 Sex distribution among patients who received bDMARDs alone Sex Frequency % Cumulative frequency Cumulative % Male 29 36.25 29 36.25 Female 51 63.75 80 100.00

These differences in the sex distribution are best appreciated in the pie chart (Figure 3.1).

SEX DISTRIBUTION AMONG BDMARDS ALONE

Male Female

Male 36% Female 64%

Note. Sex1= Male; Sex 2= Female Figure 3.1 Sex distribution among patients receiving bDMARDs alone

The majority of patients (almost 82%) who were taking bDMARDs alone were non-smokers. According to centraltendency data, patients, on average, were smokers for 21 years and consumed almost 26 cigarettes per day (Table 3.2, Table 3.4)

Table 3.4 Smoking status amongst patients receiving bDMARDs alone Smoking status amongst patients who received bDMARDs alone Smoking status Frequency % Cumulative Frequency Cumulative % No 66 82.50 66 82.50 Yes 14 17.50 80 100.00 Note. Frequency missing = 325

It is apparent that the majority of patients who received bDMARDs alone were alcohol consumers (Table 3.5).

Table 3.5 Status of alcohol consumption among patients who were taking bDMARDs Alcohol consumption among patients who received bDMARDs alone Alcohol Frequency % Cumulative Cumulative Frequency % Never 32 40.00 32 40.00 Sometimes 39 48.75 71 88.75 Everyday 9 11.25 80 100.00 Note. Frequency missing = 325

A small majority of patients (almost 52%) who were taking bDMARDs alone were taking prednisolone as well (Table 3.6)

Table 3.6 Status of taking methyl prednisolone among patients taking bDMARDs alone Prednisolone status Frequency % Cumulative Frequency Cumulative % Never taken 25 31.25 25 31.25 Also, Currently taking 42 52.50 67 83.75 according to Stopped taking 12 15.00 79 98.75 the following Don’t know 1 1.25 80 100.00 tables, Note. Frequency missing = 325 approximately 7.5% of patients on bDMARDs have current T1DM, whereas 12.5% have current T2DM (Table 3.7, Table 3.8).

Table 3.7 Status of T1DM among patients who were taking bDMARDs alone T1DM Frequency % Cumulative Cumulative Frequency % Never 73 91.25 73 91.25 Current 6 7.50 79 98.75 Past 1 1.25 80 100.00

Frequency missing = 325

T2DM was observed to be more common amongst patients who were receiving bDMARDs alone (Table 3.8).

Table 3.8 Frequency of T2DM amongst patients who were receiving bDMARDs alone T2DM Frequency % Cumulative frequency Cumulative % Never 68 85.00 68 85.00 Current 10 12.50 78 97.50

Past 2 2.50 80 100.00 Note. Frequency missing = 325

3.3. Demography of patients receiving csDMARDs alone

During the period of this study (2001 to 2014), 405 patients were taking csDMARDs alone without any csDMARDs. The mean age was around 59 years (Table 3.9). According to the following table (Table 3.9), among all patients with RA, smokers were, on average, smoking for 12 years about 10 cigarettes per day.

Table 3.9 Mean and central tendency in csDMARDs alone Variable Mean SD Median Age 59.24 12.69 60.00 Smoking status among smokers 10.49 12.12 10.00 Smoking duration (yrs) among smokers 12.40 14.46 9.00

Amongst the 405 patients who were receiving csDMARDs alone, 77.04% were female. In comparison to bDMARDs alone, the proportion of females receiving csDMARDs alone was higher (Table 3.10).

Table 3.10 Sex distribution among patients who received csDMARDs alone Gender Frequency % Cumulative frequency Cumulative % Male 93 22.96 93 22.96 Female 312 77.04 405 100.00

The difference in sex distribution is best appreciated in the following Figure (Figure 3.2). It can be seen that females predominate. .

SEX DISTRIBUTION IN CSDMARDS ALONE

Male Female

Male 23% Female 77%

Figure 3.2 Gender distribution among patients who took csDMARDs alone

The majority of patients receiving csDMARDs alone were non-smokers. According to central tendency data (Table 3.2), the smokers in this group consumed almost 10 cigarette s per day with an average smoking duration of 12 years.

Table 3.11 Smoking status in those receiving csDMARDs alone Smoking status Smoker Frequency % Cumulative frequency Cumulative % Missing 1 0.25 1 0.25 No 367 90.62 368 90.86 Yes 37 9.14 405 100.00

The majority of patients, up to around 90%, were non-smokers, whereas almost 70% were consumers of alcohol (Table 3.9, Table 3.11, Table 3.12).

Table 3.12. Alcohol consumption in those receiving csDMARDs alone Alcohol Frequency % Cumulative frequency Cumulative % Never 123 30.37 123 30.37 Sometimes 209 51.60 332 81.98 Everyday 73 18.02 405 100.00

The significance of differences observed between the csDMARDs-alone and bDMARDs-alone groups were examined by statistical testing. Prednisolone was noted to be used by 46% of patients receiving csDMARDs alone (Table 3.13).

Table 3.13 Prednisolone usage amongst those receiving csDMARDs alone Prednisolone status Frequency % Cumulative frequency Cumulative % Never 110 27.16 110 27.16 Currently 187 46.17 297 73.33 Stopped 107 26.42 404 99.75 Don’t know 1 0.25 405 100.00

T1DM, as another well-known risk for infection, was reported in just 3.71% of all patients who were taking csDMARDs [20] (Table 3.14).

Table 3.14. Insulin Dependent Diabetes Mellitus in CsDMARDs alone Frequency of T1DM in patients taking (csDMARDs) T1DM Frequency % Cumulative Cumulative Frequency % Never 390 96.30 390 96.30 Current 14 3.46 404 99.75 Past 1 0.25 405 100.00

The frequency of T2DM reported was slightly higher than for T1DM, but still less than 10% (Table 3.15).

Table 3.15 Non-insulin dependent diabetes mellitus in those receiving csDMARDs alone

T2DM Frequency % Cumulative Frequency Cumulative % don’t know 1 0.25 1 0.25 Never 379 93.58 380 93.83 Current 20 4.94 400 98.77 Past 5 1.23 405 100.00

3.4. Comparison of patients receiving bDMARDs and patients on csDMARDs alone In order to compare the frequency of qualitative variables, for each variable a separately chi- square table was used (Table 3.17, Table 3.19, Table 3.21, Table 3.23, Table 3.25, Table 3.27, Table 3.29, Table 3.35, Table 3.41, Table 3.45, Table 3.55, Table 3.59). In these tables, the significance of the differences is tested by different test methods, including the Wald chi-square test (pearson Chi-Square), the Continuity-Adjusted Chi-Square test, the Mantel-Haenszel Chi- Square test, the Likelihood Ratio Chi-Square test, the Phi Coefficient, and Cramer’s V. Among these tests, the Wald Chi-Square, which is also known as the Pearson Chi-Square, is the most commonly used test. The null hypothesis is that the frequency of the variable is similar for recipients of biologic DMARDs alone and recipients of conventional synthetic DMARDs alone recipients. A criticism of this test is that the Wald Chi-square fixes the row and column margin totals which, in effect, makes an assumption about the distribution of the variables in the population being studied (Table 3.17)[21].

The second test in this table (Table 3.17) is the Continuity-Adjusted Chi-Square test statistic. This test consists of the Pearson Chi-Square modified with an adjustment for continuity and is dependent on the sample size. The Mantel-Haenszel Chi-Square test is usually related to the Pearson Chi-Square test. In the 2x2 case, as the sample size gets larger, the Mantel-Haenszel and Wald Chi Square statistics tests converge[22]. In the case of 2xC or Rx2 tables, if the variable with more than two categories is ordinal, the Mantel-Haenszel Chi-square is a test for trend while the Pearson Chi-square remains a general test for association[22]. This test is currently calculated and reported in SAS, but it was not evaluated further in this study. The Likelihood Ratio Chi-Square is asymptotically equivalent to the Pearson Chi-Square and

Mantel-Haenszel Chi-Square but not usually used when analysing 2x2 tables. It is used in logistic regression and log linear modelling which involves contingency tables[22]. [22]. Cramer’s V is derived from the chi-square and in the 2 x2 table, which is identical to the Phi coefficient. The contingency coefficient, the Phi coefficient, and Cramer’s V are well-suited for nominal variables in which the order of the levels is meaningless[23].

3.4.1. Prednisolone comparison

Prednisolone is one of the anti-rheumatic medications known to play a role in infection (Chapter 2, section 3.12.2.). It is important to check if prednisolone intake differs between patients who are taking csDMARDs and patients who are taking bDMARDs (Table 3.16).

Table 3.16 Comparison of prednisolone consumption among patients receiving csDMARDs alone and bDMARDs alone Group Response Status Never Currently Stopped Not Total Taking taking taking Known Frequency 110 187 107 1 405 csDMARDs % 22.68 38.56 22.06 0.21 83.51 Row % 27.16 46.17 26.42 0.25 Column % 81.48 81.66 89.92 50.00 Frequency 25 42 12 1 80 bDMARDs % 5.15 8.66 2.47 0.21 16.49 Row % 31.25 52.50 15 1.25 Column % 18.52 18.34 10.08 50.00 Total Frequency 135 229 119 2 485 % 27.84 47.22 24.54 0.41 100

Based on the p-value of the Chi-square in the ARAD sample, there is not a significant difference between the csDMARDs and bDMARDs groups in taking prednisolone. As the number of samples in 25 % of calculating cells was less than 5, other tests (chi- square, likelihood ratio, Mantel- Haenszel, phi Coefficient, Contingency coefficient, Cramer’s V) were also checked, and all of these tests align with the original finding (Table 3.17) [19].

Table 3.17 Chi-squared test for difference in frequency of prednisolone usage between csDMARDs-alone recipients and bDMARDs-alone recipients, sample size=485 Statistic DF Value Probability Chi-Square 3.00 6.15 0.10 Likelihood Ratio Chi-Square 3.00 6.15 0.10 Mantel-Haenszel Chi-Square 1.00 2.25 0.13 Phi Coefficient 0.11 - Contingency Coefficient 0.11 - Cramer's V 0.11 -

3.4.2. Alcohol comparison

There are also other factors which may play a role with respect to infection susceptibility and also impact on general health. Alcohol consumption and cigarette smoking are two such factors. Alcohol consumption in recipients of csDMARDs alone and bDMARDs alone was examined. The results are shown in Table 3.18 and Table 3.19.

Table 3.18 Comparison of alcohol consumption among patients receiving csDMARDs alone and bDMARDs alone

Group Response

Status Never Sometimes Everyday Total taking CsDMARDs Frequency 123 209 73 405 % 25.36 43.09 15.05 83.51 Row % 30.37 51.60 18.02 Column % 79.35 84.27 89.02 bDMARDs Frequency 32 39 9 80 % 6.60 8.04 1.86 16.49 Row % 40.00 48.75 11.25 Column % 20.65 15.73 10.98 Total Frequency 155 248 82 485 % 31.96 51.13 16.91 100

Notes. Sample size is a combination of 405 patients who were taking csDMARDs alone and 80 patients who were bDMARDs alone. Based on the chi-square test, the differences in alcohol consumption between csDMARDs alone and bDMARDs alone were not statistically significant (Table 3.19). Table 3.19 Chi-squared for differences in frequency of alcohol use between recipients of csDMARDs alone and recipients of bDMARDs alone, sample size=485.

Statistic DF Value Probability Chi-Square 2.00 3.86 0.15 Likelihood Ratio Chi-Square 2.00 3.94 0.14 Mantel-Haenszel Chi-Square 1.00 3.85 0.05 Phi Coefficient 0.09 Contingency Coefficient 0.09 Cramer's V 0.09

3.4.3. Smoking comparison

Smoking is another potential risk factor for infection and deterioration in patient health status. Statistically, it seems that more people in the csDMARDs group intend to smoke (Table 3.20a). However, this is not statistically significant different from the other group (Table 3.21).

Table 3.20a Comparison smoking status among patients receiving csDMARDs alone and bDMARDs alone Group Response Status No Yes Missing Total Frequency 367 37 1 405 csDMARDs % 75.67 7.63 0.21 83.51 Row % 90.62 9.14 0.25 Column % 84.76 72.55 100.00 Frequency 66 14 0 80 bDMARDs % 13.61 2.89 0.00 16.49 Row % 82.50 17.50 0.00 Column % 15.24 27.45 0.00 Total Frequency 433 51 1 485 % 89.28 10.52 0.21 100.00

Different studies reveal the strong connection between smoking and RA. For example, Criswell et al. (2002), in a cohort study, showed that those who stopped smoking could have a reduced risk of RA, particularly among postmenopausal women [24]. In a case study by Padyukov et al. (2004), it was shown that the risk of RA with SE of HLA-DR is strongly influenced by the presence of an environmental factor (e.g., smoking) in the population at risk [16]. Costenbader et al. (2006) showed in a cohort study that past and current smoking were related to the development of RA, in particular seropositive RA [25]. In a meta-analysis by Sugiyama et al. (2010), it was shown that smoking is a risk for RA, especially seropositive RA in men. For women, the risk for smokers is about 1.3 times greater than for non- smokers [26]. Di Giuseppe et al. (2014) showed that lifelong cigarette smoking was positively associated with the risk of RA, even among smokers with a low lifelong exposure [27].

Furthermore, other studies show a connection between the effectiveness of smoking cession and better responsiveness of bio-treatment [28]. Sustained smoking cessation within four years of RA diagnosis is connected to a reduction in mortality risk, this rate is same as non- smokers. However, smoking more than 5 years after RA diagnosis increased mortality well above the risk of non-RA patients [29]. According to the ARAD, the rate of smoking between 2001 and 2014 was 10.5% (328/3111). This was almost 8.9% of all patient visits (2484/27712). Table 3.20b shows the rate of smoking in the general population in Australia during the same time.

Table 3.20b Comparing rate of smokers during the years 2001 to 2013, Australia [30] Year %Total smokers 2001 22 2004 20 2007 19 2010 18 2013 15

It seems that the rate of smokers in RA is less than the rate of smokers among the general population in Australia during those years. Two major possibilities for this discrepancy include

the facts that (1) ARAD is a subjective report and data are not reliable, and (2)there are several risk factors for causing RA, and risk factors other than smoking play a more significant role in Australia, especially since the rate of RA disease in Australia is higher than in many other countries [31].

Based on the information in the smoking and alcohol consumption tables, the apparent differences in smoking history and alcohol consumption status are not confirmed, statistically. Accordingly, they are unlikely to account for any differences in infections observed between the csDMARDs and bDMARD groups. There is marginal evidence for differences with respect to smoking between the csDMARDs-alone and bDMARDs-alone groups (Table 3.21). In this test, almost 33% of cells had an expected count of less than five, which means that further statistical testing is required. For this purpose, the likelihood ratio test was performed, and this test confirmed the results[32].

Table 3.21 Chi-squared for differences in frequency of smoking between csDMARDs-alone and bDMARDs-alone groups, sample size=485 Statistical tests DF Value Probability Chi-Square 2.00 5.14 0.08 Likelihood Ratio Chi-Square 2.00 4.72 0.09 Mantel-Haenszel Chi-Square 1.00 4.07 0.04 Phi Coefficient 0.10 Contingency Coefficient 0.10 Cramer's V 0.10

3.4.4. Sex distribution comparison

Sex differences can also play a major role in many conditions. Therefore, it is important to determine if there is a sex difference between patients taking csDMARDs and those taking bDMARDs (Table 3.22).

Table 3.22 Comparison sex distribution among patients receiving csDMARDs alone and bDMARDs alone (sample Size = 485) Group Response Status Male Female Total csDMARDs Frequency 93 312 405 % 19.18 64.33 83.51 Row % 22.96 77.04 Column % 76.23 85.95 bDMARDs Frequency 29 51 80 % 5.98 10.52 16.49 Row % 36.25 63.75 Column % 23.77 14.05 Total Frequency 122 363 485 % 25.15 74.85 100.00

Based on the information in Table 3.23, the Chi-Square test result for sex difference is highly significant, at the level of 0.05. This means that the sex distribution amongst recipients of csDMARDs and bDMARDs is different. Accordingly, gender may confound inferences made in relation to these two groups. In order to confirm this difference and confirm that the small size of the population is not responsible, a Fisher's test was performed (Table 3.24)[18].

Table 3.23 Chi-squared for differences in frequency of sex distribution among recipients of csDMARDs alone and recipients of bDMARDs alone Statistic DF Value Probability Chi-Square 1.00 6.26 0.01 Likelihood Ratio Chi-Square 1.00 5.88 0.02 Continuity Adj. Chi-Square 1.00 5.58 0.02 Mantel-Haenszel Chi-Square 1.00 6.25 0.01 Phi Coefficient -0.11 Contingency Coefficient 0.11 Cramer's V -0.11

Based on the Fisher’s test, it was confirmed that the two populations have a different sex distribution (Table 3.24). Table 3.24 Fisher test for differences in frequency of sex distribution between recipients of csDMARDs alone and recipients of bDMARDs alone

Fisher's Exact Test Cell (1,1) Frequency (F) 93.00 Left-sided Pr <= F 0.01 Right-sided Pr >= F 0.99

Table Probability (P) 0.01 Two-sided Pr <= P 0.02

Based on the statistical tests, the sex distribution is different between csDMARDs and bDMARDs. In the following pie chart, all sex distributions are presented to make this comparison easier (Figure 3.3). According to this chart, in both csDMARDs and bDMARDs, the population of the female sex is greater than the male sex and this difference is greater amongst those taking csDMARDs (Figure 3.3).

Sex differences in csDMARDs alone and bDMARDs alone Biologic Male 6% csDMARDs Male 19% Biologic Male csDMARDs Male

bDMARDS Female bDMARDS Female csDMARDs Female 11% 64% csDMARDs Female

Figure 3.3 Sex distribution among csDMARDs alone and bDMARDs alone

3.4.5. T2DM comparison

Table 3.25 summarises the differences in the frequency of T2DM in recipients of csDMARDs and bDMARDs.

Table 3.25 Comparison of T2DM among patients receiving csDMARDs alone and bDMARDs alone Group Response Status No known Never Current Past Total CsDMARDs Frequency 1 379 20 5 405 % 0.21 78.14 4.12 1.03 83.51 Row % 0.25 93.58 4.94 1.23 Column % 100 84.79 66.67 71.43 bDMARDs Frequency 0 68 10 2 80 % 0.00 14.02 2.06 0.41 16.49 Row % 0.00 85.00 12.50 2.5 Column % 0.00 15.21 33.33 28.57 Total Frequency 1 447 30 7 485 % 0.21 92.16 6.19 1.44 100.00

There is marginal evidence that the frequency of T2DM in ARAD participants is different between patients taking csDMARDs alone and those taking bDMARDs alone. The difference was not significant at the 0.05 level of significance (Table 3.26).

Table 3.26 Chi-squared for differences in frequency of T2DM between csDMARDs alone and bDMARDs alone, sample size= 485 Statistic DF Value Probability Chi-Square 3.00 7.65 0.05 Likelihood Ratio Chi-Square 3.00 6.60 0.09 Mantel-Haenszel Chi-Square 1.00 6.26 0.01 Phi Coefficient 0.13 Contingency Coefficient 0.12 Cramer's V 0.13

3.4.6. T1DM comparison

Table 3.27 summarises the differences in frequency of T2DM in recipients of csDMARDs alone and bDMARDs alone.

Table 3.27 Comparison T1DM among patients on csDMARDs alone and bDMARDs alone Group Response Status No known Never Current Past CsDMARDs Frequency 390 14 1 405 % 80.41 2.89 0.21 83.51 Row % 96.30 3.46 0.25 Column % 84.23 70.00 50.00 bDMARDs Frequency 73 6 1 80 % 15.05 1.24 0.21 16.49 Row percentage 91.25 7.50 1.25 Column % 15.77 30.00 50.00 Total Frequency 463 20 2 485 % 95.46 4.12 0.41 100.00

Based on the chi-square test, there is no difference in the frequency of T1DM between patients who are taking csDMARDs alone and those who are taking bDMARDs alone. However, the population size is low and there is a possibility that using just Chi-squared reduces the accuracy of this test. Therefore, other tests (chi- square, likelihood ratio, Mantel- Haenszel, phi Coefficient, Contingency coefficient, Cramer’s V) have also been assessed (Table 3.28) [18].

Table 3.28 Chi-squared for differences in frequency of T1DM between csDMARDs alone and bDMARDs alone, Sample size=485 Statistic DF Value Probability Chi-Square 2.00 4.46 0.11 Likelihood Ratio Chi-Square 2.00 3.61 0.16 Mantel-Haenszel Chi-Square 1.00 4.41 0.04 Phi Coefficient 0.10 Contingency Coefficient 0.10 Cramer's V 0.10

In the following, we study each organ infection in csDMARDs versus bDMARDs in more detail. In different sections, we discuss different levels of infection, including mild, moderate and severe. Based on the ARAD questionnaire (Appendix M), mild infection is defined as an infection which does not change activities and the patient did not see a doctor and did not require prescription medicine for treatment. Moderate infection is defined as an infection which changes activities occasionally and the patient needed a prescription medication for the symptoms. Severe infection is an infection which can cause a major change in activities and the patient needed to see a doctor and received prescription medication, however, the medication only provided partial relief.

3.4.7. Skin and nail infections comparison

Skin and nail infections are amongst the commonest infections in RA. Skin and nail infection was reported for three different levels of severity. The relationship between the frequency of these levels and type of medication are reviewed in Tables 3.29 and 3.30

Table 3.29 Table of frequency of skin and nail infections in recipients of csDMARDs alone recipients. Skin and nail infection in csDMARDs alone Severity Frequency % Cumulative Cumulative Frequency % Mild 76 53.15 76 53.15 Moderate 56 39.16 132 92.31 Severe 11 7.69 143 100.00 Note. Frequency missing = 1510

Severe skin and nail infection occurred more frequently in patients who were taking bDMARDs alone compared to csDMARDs alone (Tables 3.29 and 3.30). In contrast, other types of infections were either similar or more frequently observed in csDMARDs recipients (Tables 3.29-3.30).

Table 3.30 Table of frequency of skin and nail infections in bDMARDs alone Skin and nail infection in Biologics Severity Frequency % Cumulative Cumulative Frequency % Mild 16 53.33 16 53.33 Moderate 11 36.67 27 90.00 Severe 3 10.00 30 100.00 Note. Frequency missing = 1623

Almost 44 % of patients who were taking csDMARDs alone have reported mild levels of skin/nail infection. This is about 53% of all patients who were taking csDMARDs alone and 82% of all patients who reported mild skin infection (Table 3.31). Almost 9% of patients who were taking bDMARDs alone reported mild infection; this is about 53% of all patients who were taking bDMARDs alone and almost 17% of all patients who reported mild infection (Table 3.31). Almost 6 % of patients who were taking CsDMARDs alone have reported severe level of skin infection. This is about 7.69% of all patients who were taking csDMARDs alone and 78.5% of all patients who reported severe skin infection (Table 3.31). Almost 2% of patients who were taking bDMARDs alone reported severe skin infection, this is about 10% of all patients who were on bDMARDs alone and almost 21% of all patients who reported severe infection (Table 3.31).

Table 3.31 Differences in frequency of skin infections in csDMARDs and bDMARDs alone Response Group Status Mild Moderate Severe Total Frequency 76 56 11 143 % 43.93 32.37 6.36 82.66 csDMARDs Row percentage 53.15 39.16 7.69 Column % 82.61 83.58 78.57 Frequency 16 11 3 30 % 9.25 6.36 1.73 17.34 bDMARDs Row % 53.33 36.67 10.00 Column % 17.39 16.42 21.43 Total Frequency 92 67 14 173 % 53.18 38.73 8.09 100.00

Based on the table 3.32, the point of estimate was used on the Chi-square and calculating p- values. This shows that the null hypothesis cannot be rejected at the 0.05 level of significance. Therefore, the frequency of self-reported infection is similar in both groups (Table 3.32). 88

Table 3.32 Table and figures showing differences in frequency of skin/nail infections in recipients of csDMARDs alone and bDMARDs alone, Sample size=173

Statistic DF Value Probability Chi-Square 2.00 0.20 0.90 Likelihood Ratio Chi-Square 2.00 0.19 0.91 Mantel-Haenszel Chi-Square 1.00 0.03 0.87 Phi Coefficient 0.03 Contingency Coefficient 0.03 Cramer’s V 0.03

3.4.8. Eyes, Ears, nose, Throat (EENT) Infections – a comparison EENT infections are among the most common infections in RA. EENT infection also was reported for three different levels of severity. The data suggest that csDMARDs can increase the frequency of mild and moderate EENT infection, while bDMARDs appeared to increase the frequency of severe EENT infection (Table 3.33, Table 3.34).

Table 3.33 Frequency of eye, ear, nose & throat infections in patients taking csDMARDs alone Ear Nose Throat infection in csDMARDs alone Eent Infection Frequency % Cumulative Cumulative Frequency % Mild 83 44.39 83 44.39 Moderate 82 43.85 165 88.24 Severe 22 11.76 187 100.00 Note. Frequency missing = 1466

According to Table 3.33 and Table 3.34, severe EENT infection occurs more often in patients who are taking bDMARDs alone compared to those who take csDMARDs alone. The frequency of other types of infections was similar in both groups (Table 3.33-3.34)

Table 3.34 Frequency of Ear Nose Throat infections in in patients taking bDMARDs alone Ear Nose Throat infection in recipients of bDMARDs EENT Frequency % Cumulative Cumulative infection Frequency % Mild 10 41.67 10 41.67 Moderate 10 41.67 20 83.33 Severe 4 16.67 24 100.00 Note. Frequency missing = 1629

Almost 10% of patients who were taking csDMARDs alone reported severe EENT infection, this is about 12% of all patients who were taking csDMARDs alone (Table 3.35). In contrast, 16 % of patients who were taking bDMARDs reported severe EENT infection (Table 3.35). The figures in Table 3.35 are just descriptive, and don’t allow the strength of the association to be confirmed.

Table 3.35 Table of differences in frequency of ear, nose, and throat infections in csDMARDs alone and bDMARDs alone Response Group Status Mild Moderate Severe Total Frequency 83 82 22 187 % 39.34 38.86 10.43 88.63 csDMARDs Row % 44.39 43.85 11.76 Column% 89.25 89.13 84.62 Frequency 10 10 4 24 bDMARDs % 4.74 4.74 1.90 11.37 Row% 41.67 41.67 16.67 Column % 10.75 10.87 15.38 Total Frequency 93 92 26 211 % 44.08 43.60 12.32 100.00

Calculating Chi- Square and P-Value for differences in EENT infection reveals that frequency of self-reported infection is similar in both bDMARDs alone and csDMARDs alone. The large value of the chi-square statistic, 0.4737, and the p-value of 0.7891 indicate that the null

hypothesis cannot be rejected at the 0.05 level of significance. Therefore, we conclude that frequency of eye, ear, nose and throat (EENT) infection is similar in both groups (Table 3.36). A criticism of this test is that it fixes the row and column margin totals, which in effect makes an assumption about the distribution of the variables in the population being studied.

Table 3.36 Chi-squared for differences in frequency of EENT infections between csDMARDs alone and bDMARDs alone, sample size=211 Statistic DF Value Probability Chi-Square 2.00 0.47 0.79 Likelihood Ratio Chi-Square 2.00 0.44 0.80 Mantel-Haenszel Chi-Square 1.00 0.27 0.61 Phi Coefficient 0.05 Contingency Coefficient 0.05 Cramer's V 0.05

3.4.9. Heart infections comparison

Heart infection also was reported for three different levels of severity. Based on estimation of the frequency of heart infection, the frequency of heart infection among patients taking bDMARDs alone and csDMARDs alone weas different, with a significantly lower frequency of infection among patients on bDMARDs (Table 3.37). Amongst the large number of reports, 1646 participants did not report any heart infection.

Table 3.37 Table of frequency of Heart infections in csDMARDs alone Heart infection in csDMARDs alone Heart infection Frequency % Cumulative Frequency Cumulative % Moderate 3 42.86 3 42.86 Severe 4 57.14 7 100.00 Note. Frequency Missing = 1646

Amongst bDMARD-alone recipients, only one case of mild heart infection was reported, whereas in csDMARDs-alone recipients, a few patients reported moderate or severe infections (Table 3.38). Patients with a moderate level of infection were 58,66 and 66 years old, while those with a severe level of infection were 59, 60, 65 and 69 years old.

Table 3.38 Table of frequency of Heart infections in bDMARDs alone Heart infection in Biologics Heart Frequency % Cumulative Cumulative infection Frequency % Mild 1 100.00 1 100.00 Note. Frequency missing = 1652

As the difference between csDMARDs and bDMARDs in heart infection appeared to be possibly significant, the differences are demonstrated in a cylindrical graph (Figure 3.4). As it is shown in the graph, the number of patients who were taking biologics and reported heart infection was close to zero or negligible (Figure 3.4). The only patient in this group was 67 years old.

Heart Infection

100

1 Hundreds 0.1

Number of patients in 0.01 Mild Moderate Severe No infection

csDMARDs BDMARDs

Figure 3.4 Comparison of the rate of heart infection in recipients of csDMARDs alone and bDMARDs alone (with logarithm base).

3.4.10. Lung infections comparison

Lung infection was also reported for three different levels of severity. Lung infection is also an important infection in rheumatoid arthritis and, based on Table 3.39, almost 61.29 % of the participants in the csDMARDs group reported moderate lung infection (Table 3.39).

Table 3.39 Table of frequency of Lung infections in csDMARDs alone Lung infection in csDMARDs alone Lung infection Frequency % Cumulative Cumulative Frequency % Mild 25 16.13 25 16.13 Moderate 95 61.29 120 77.42 Severe 35 22.58 155 100.00 Note. Frequency missing = 1498

According to Table 3.39 and Table 3.40, severe lung infection occurs almost twice as frequently in patients who are taking csDMARDs alone compared to those who were taking bDMARDs alone.

Table 3.40 Table of frequency of Lung infections in recipients of bDMARDs alone Lung infection in bDMARDs recipients Lung infection Frequency % Cumulative Cumulative Frequency % Mild 11 40.74 11 40.74 Moderate 10 37.04 21 77.78 Severe 6 22.22 27 100.00 Note. Frequency missing = 1626

Almost 19 % of patients who were taking csDMARDs alone reported severe lung infection. This is about 22 % of all patients who were taking csDMARDs alone and almost 85% of all patients who reported lung infection (Table 3.41). Almost 3% of patients who were taking bDMARDs alone reported severe lung infection, this is about 22% of all patients who were on bDMARDs alone and almost 15% of all patients who reported severe lung infection (Table 3.41). These figures are just descriptive, and any statistical differences need to be confirmed with Chi-squared tests (Table 3.41).

Table 3.41 Table and Figure of differences in frequency of Lung infections in recipients of csDMARDs alone and bDMARDs alone Group by response Group Mild Moderate Severe Total Frequency 25 95 35 15 Lung infection in recipients of % 13.74 52.20 19.23 85.16 csDMARDs alone Row % 16.13 61.29 22.58 Column % 69.44 90.48 85.37 Frequency 11 10 6 27 Lung infection in recipients of % 6.04 5.49 3.30 14.84 bDMARDs alone Row % 40.74 37.04 22.22 Column % 30.56 9.52 14.63 Total 36 105 41 182 19.78 57.69 22.53 100

The difference in lung infection between those taking csDMARDs alone and bDMARDs alone is statistically significant (p-value 0.01), at the level of 0.05 (Table 3.42). The null hypothesis is that the frequency of self-reported infection is similar in both users of bDMARDs alone and csDMARDs alone. The large value of the chi-square statistic, 9.3875, and the small amount of p-value of 0.01 indicate that the null hypothesis can be rejected at the 0.05 level of significance. Therefore, it can be concluded that the frequency of lung infection is different in both csDMARDs alone and bDMARDs alone. In other words, the frequency of lung infection is significantly higher among patients who are taking csDMARDs alone (Figure 3.5).

Table 3.42 Chi-squared for differences in frequency of lung infections between csDMARDs alone and bDMARDs alone, sample size= 182

Statistic DF Value Prob Chi-Square 2.00 9.39 0.01 Likelihood Ratio Chi-Square 2.00 8.32 0.02 Mantel-Haenszel Chi-Square 1.00 3.38 0.07

Phi Coefficient 0.23 Contingency Coefficient 0.22 Cramer's V 0.23

The difference in lung infection in between csDMARDs and bDMARDs is demonstrated in the following chart (Figure 3.5).

Lung Infection 100

0.1 Number of patients in Hundreds

0.01 Mild Moderate Severe No infection

CsDMARDs BDMARDs

Figure 3.5 Comparison of the frequency of lung infection for recipients of csDMARDs alone and bDMARDs alone (with logarithm base).

3.4.11. Gasterointestinal tract (GIT) infections GIT infection also was reported for three different levels of severity. Based on the frequency table below, it can be seen that the frequency of gastrointestinal tract (GIT) infection among patients who were taking biologic DMARDs was much lower than that for recipients of csDMARDs alone (Tables 3.43- 3.44).

Table 3.43 Table of frequency of GIT infections in recipients of csDMARDs alone Gastero Intestinal Tract (GIT) infection in csDMARDs alone GIT Infection Frequency % Cumulative Cumulative Frequency % Mild 7 29.17 7 29.17 Moderate 8 33.33 15 62.50 Severe 9 37.50 24 100.00 Note. Frequency missing = 1629

In recipients of bDMARDs, the only GIT infection reported was of moderate severity, while in recipients of csDMARDs, there were numerous reports of GIT infections in the mild, moderate and severe categories (Tables 3.43- 3.44).

Table 3.44 Table of frequency of GIT infections in in recipients of bDMARDs alone Gastro-intestinal Tract (GIT) infection in recipients of bDMARDs alone Biologics Frequency % Cumulative Cumulative GIT Frequency % infection

Moderate 1 100.00 1 100.00 Note. Frequency missing = 1652

As GIT infection was different between the csDMARDs and bDMARDs groups, we compared this type of infection in these two groups (see Figure 3.6). Based on this figure, csDMARDs is the major contributing factor for this type of infection and, with a minor difference between groups, most of the patients reported a severe type of infection.

GIT infection 100

1 Number of patients in Hundreds

0.1

0.01 Mild Moderate Severe No infection

csDMARDs BDMARDs

Figure 3.6 Comparison of the frequency of GIT infection in patients taking csDMARDs alone and bDMARDs alone

3.4.12. Urinary tract infections (UTI)

Urinary tract infection was reported at three different levels of severity. Based on Table 3.45, moderate infection is the most common type of UTI in recipients of csDMARDs, followed by mild and severe infections (Table 3.45).

Table 3.45 Table of frequency of UTI in recipients of csDMARDs alone Urinary System infection in recipients of csDMARDs alone Kidney and Urinary Frequency % Cumulative Cumulative infection Frequency % Mild 12 15.19 12 15.19 Moderate 56 70.89 68 86.08 Severe 11 13.92 79 100.00 Note. Frequency missing = 1574

Almost 6% of patients who were taking bDMARDs alone reported severe urinary system infection. This is about 25% of all patients who were receiving bDMARDs alone and almost 35% of all patients who reported severe urinary tract infection. These is descriptive information and will be tested in the following section (Table 3.46).

Table 3.46 Table and figure showing differences in frequency of urinary tract infections in recipients of csDMARDs alone and bDMARDs alone Urinary tract infection recipients of bDMARDs Kidney and Urinary Frequency % Cumulative Cumulative infection Frequency % Mild 12 50.00 12 50.00 Moderate 6 25.00 18 75.00 Severe 6 25.00 24 100.00 Note. Frequency missing = 1629

In order to compare UTIs in recipients of csDMARDs alone and bDMARDs alone, the tables were combined. (Table 3.47).

Table 3.47 Table and Figure of differences in frequency of urinary tract infections in csDMARDs alone and bDMARDs alone recipients Group Response Status Mild Moderate Severe Total Frequency 12 56 11 79 csDMARDs % 11.65 54.37 10.68 76.70 Row % 15.19 70.89 13.92 Column % 50.00 90.32 64.71 Frequency 12 6 6 24 bDMARDs % 11.65 5.83 5.83 23.30 Row % 50.00 25.00 25.00 Column % 50.00 9.68 35.29 Total Frequency 24 62 17 103 % 23.30 60.19 16.50 100.00

Differences between UTIs in recipients of csDMARDs alone and bDMARDs alone were tested by Chi-square and p-value. The null hypothesis is that the frequency of Urinary tract infection (UTI) differs between recipients of bDMARDs alone and csDMARDs alone. Examined for the three categories, notably mild, moderate, and severe. The large value of the chi-square statistic, 17.3798, and the low p-value of 0.0002 indicate that the null hypothesis should be rejected at the 0.05 level of significance. Therefore, it was concluded that the frequency of UTI is different for these two groups, and that the observed difference is highly statistically significant. In other words, the frequency of moderate and severe UTI is significantly higher among recipients of csDMARDs alone. The associations observed for moderate and severe UTIs in recipients of csDMARDs alone were not apparent for mild UTIs.

Table 3.48 Chi-squared for differences in frequency of urinary tract infections between csDMARDs alone and bDMARDs alone, sample size=103 Statistic DF Value Probability Chi-Square 2.00 17.38 0.00 Likelihood Ratio Chi-Square 2.00 17.07 0.00 Mantel-Haenszel Chi-Square 1.00 2.61 0.11 Phi Coefficient 0.41 Contingency Coefficient 0.38 Cramer's V 0.41

Almost 10% of patients who were taking csDMARDs alone reported severe urinary tract infections, this is about 14% of all patients who were receiving csDMARDs alone and almost 64% of all patients who reported urinary tract infection (Figure 3.7)

Urinary System infection

100

Hundreds 0.1 Number of patients in 0.01 csDMARDs BDMARDs mild moderate severe No infection Figure 3.7 Frequency of urinary tract infections in recipients of csDMARDs alone and bDMARDs alone

UTI is more prevalent among patients on csDMARDs. UTI is also more prevalent among the female sex. As there is a significant difference between female and male distribution in between csDMARDs and bDMARDs, the current difference in UTI can be partly and completely due to this difference in the sex distribution[33].

3.4.13. Musculoskeletal infections (MSK)

MSK infection also was reported for three different levels of severity. The frequencies for MSK infection in recipients of csDMARDs alone and bDMARDs alone were very similar. Moderate MSK infection was more frequent in recipients of csDMARDs alone. (Tables 3.49-3.50).

Table 3.49 Table of frequency of musculoskeletal system infection in recipient of csDMARDs alone Musculoskeletal system infection in csDMARDs alone Bone Joint and Frequency % Cumulative Cumulative Muscle infection Frequency % Mild 6 19.35 6 19.35 Moderate 16 51.61 22 70.97 Severe 9 29.03 31 100.00 Note. Frequency missing = 1622

The figures in Tables 3.49 to 3.51 are just descriptive and need to be tested further by application of a Chi-squared test.

Table 3.50 Table of frequency of musculoskeletal system infection in bDMARDs alone Musculoskeletal system infection in recipient of bDMARDs Bone Joint and Frequency % Cumulative Cumulative Muscle infection Frequency % Mild 4 33.33 4 33.33 Moderate 4 33.33 8 66.67 Severe 4 33.33 12 100.00 Note. Frequency missing = 1641

Almost 21 % of all patients who were receiving csDMARDs alone or bDMARDs alone reported severe MSK infection; this is about 29 % of all patients who were on csDMARDs alone and almost 69% of all patients who reported severe MSK infection. Almost 9% of patients who were receiving bDMARDs alone reported severe MSK infection, this is about 33% of all patients who were on bDMARDs alone and almost 30% of all patients who reported severe MSK infection (Table 3.51)

100

Table 3.51 Table and figure of differences in frequency of muscular skeletal system infections in recipients of csDMARDs alone and bDMARDs alone Group Response Status Mild Moderate Severe Total Frequency 6 16 9 31 % 13.95 37.21 20.93 72.09 csDMARDs Row % 19.35 51.61 29.03 Column % 60.00 80.00 69.23 Frequency 4 4 4 12 % 9.30 9.30 9.30 27.91 bDMARDs Row% 33.33 33.33 33.33 Column % 40.00 20.00 30.77 Total Frequency 10 20 13 43 % 23.26 46.51 30.23 100.00

The null hypothesis is that the frequency of musculoskeletal (MSK) infection differs between recipients of bDMARDs alone and csDMARDs alone. MSK infection was categorized into mild, moderate and severe. The low value of the chi-square statistic, 1.4013, and the p-value of 0.4963 indicate that the null hypothesis should be rejected at the 0.05 level of significance. Therefore, it can be concluded that frequency of MSK infection is similar in both groups (Table 3.52).

Table 3.52 Chi-squared for differences in frequency of Musculoskeletal system infections between recipients of csDMARDs alone and bDMARDs alone, sample size=4

Statistic DF Value Prob Chi-Square 2.00 1.40 0.50 Likelihood Ratio Chi-Square 2.00 1.39 0.50 Mantel-Haenszel Chi-Square 1.00 0.15 0.70 Phi Coefficient 0.18 Contingency Coefficient 0.18 Cramer's V 0.18 WARNING: 33% of the cells have expected counts less than 5. Chi-Square may not be a valid test.

101

3.4.14. Artificial joint infections

Artificial joint infection also was reported for three different levels of severity. Almost 80% of the artificial joint infections in recipients of csDMARDs were classified as severe infection (Table 3.53).

Table 3.53 Table of frequency of Artificial Joint infection in csDMARDs alone Artificial joint infection in csDMARDs alone Artificial Joint Frequency % Cumulative Cumulative Infection Frequency % Moderate 1 20.00 1 20.00 Severe 4 80.00 5 100.00 Note. Frequency missing = 1648

The only artificial joint infection which was reported in recipients of bDMARDs was classified in the self-report as mild infection (Table 3.54).

Table 3.54 Table of frequency of artificial joint infection in bDMARDs alone Artificial joint infection in Biologics Artificial Joint Frequency % Cumulative Cumulative Infection Frequency % Mild 1 100.00 1 100.00 Note. Frequency missing = 1652

Based on the above frequency table (Table 3.54) artificial joint infections in recipients of bDMARDs alone were numerically less frequent than those in recipients of csDMARDs alone. Using estimation methods and testing methods is not appropriate here because the only infection among patients who were taking bDMARDs alone was categorised as mild, whereas those self-reported by recipients of csDMARDs alone were categorized as moderate or severe (Table 3.54). Furthermore, the numbers are too small to allow meaningful comparison. However, it should be indicated that artificial joint infection is almost always an emergency case and needs hospital admission. Therefore, there should not be tolerance of mild or moderate infection. In other words, the subjective data here do not concur with reality but, still, our conclusion remains that csDMARDs cause more artificial joint infection than bDMARDs.

102

3.4.15. Nervous system infections

Nervous system infection also was reported for three different levels of severity. Based on Table 3.55 Nervous system infection is not common in RA and the only episode of this infection caused mild symptoms and occurred in one patient who was taking csDMARDs alone (Table 3.55).

Table 3.55 Table of frequency of Nervous System infection in csDMARDs alone Nervous system infection in csDMARDs alone Infection Neuro Frequency % Cumulative Cumulative Frequency % Mild 1 100.00 1 100.00 Note. Frequency missing = 1652

There is no report of nervous system infection in patients who were taking bDMARDs alone (Table 3.56).

Table 3.56 Table of frequency of Nervous system infection in recipients of bDMARDs alone Nervous system infection in Biologics Biologic Infection Frequency % Cumulative Cumulative Neuro Frequency % 0 0 0 0 Frequency Missing = 1653

3.4.16. Tuberculosis (TB) infection

Tuberculosis (TB) also was reported at three different levels of severity. However, there were just two reports of moderate to severe level tuberculous infection. Based on the frequency table tuberculous infection was very uncommon in RA (Table 3.57-3.58).

103

Table 3.57 Table of frequency of tuberculous infection in recipients of csDMARDs alone TB infection in csDMARDs alone TB Infection Frequency % Cumulative Cumulative Frequency % Moderate 2 100.00 2 100.00 Note. Frequency missing = 1651

Only two episodes of tuberculous infection with moderate symptoms were reported and these infections were restricted to recipients of csDMARDs alone (Tables 3.57-3.58).

Table 3.58 Table of frequency of tuberculous infection in recipients of bDMARDs alone TB infection in patients taking Biologics TB Infection Frequency % Cumulative Cumulative Frequency % 0 0 0 0 Note. Frequency missing = 1653

3.3.17. Blood infections

Blood infection also was reported at three different levels of severity. The majority of blood infections in recipients of csDMARDs were of moderate severity (54.55%) (Table 3.59).

Table 3.59 Table of frequency of blood infection in csDMARDs alone Blood infection in recipients csDMARDs alone Severity Frequency % Cumulative Cumulative Frequency % Mild 1 9.09 1 9.09 Moderate 6 54.55 7 63.64 Severe 4 36.36 11 100.00 Note. Frequency missing = 1642

In contrast, the majority of blood infection reports in recipients of bDMARDs were reported to be mild (50%) (Table 3.60).

104

Table 3.60 Table of frequency of blood infection in recipients of bDMARDs alone Blood infection in recipients of bDMARDs alone Severity Frequency % Cumulative Cumulative Frequency % Mild 2 50.00 2 50.00 Moderate 1 25.00 3 75.00 Severe 1 25.00 4 100.00 Note. Frequency missing = 1649

Almost 27 % of patients who were taking csDMARDs alone reported severe blood infection, this is about 36 % of all patients who were taking csDMARDs alone and almost 80% of all patients who reported severe blood infection. However, it should be indicated that infection is almost always emergency and needs hospital admission. Therefore, there is no mild or moderate infections. In other words, subjective data here does not coordinate with reality, but still our conclusion stays similar and states that csDMARDs cause more blood infection than bDMARDs. Almost 7% of patients who were taking bDMARDs alone reported severe blood infection. This was about 25% of all patients who were taking bDMARDs alone and almost 20% of all patients who reported severe blood infection (Tables 3.61).

Table 3.61 Table and figure for differences in frequency of blood infections in recipients of csDMARDs alone and bDMARDs alone Group Response Status Mild Moderate Severe Total Frequency 1 6 4 11 % 6.67 40.00 26.67 73.33 csDMARDs Row % 9.09 54.55 36.36 - Column % 33.33 85.71 80.00 - Frequency 2 1 1 4 % 13.33 6.67 6.67 26.67 bDMARDs Row % 50.00 25.00 25.00 Column % 66.67 14.29 20.00 Total Frequency 3 7 5 15 % 20.00 46.67 33.33 100.00

105

The figures from tables 3.59 to 3.61 are just descriptive, and they do not reveal anything about the association. In order to estimate the association, we need to do Chi-squared tests. In order to undertake the Chi-squared test, the null hypothesis is that the frequency of blood infection is greater in the recipients of csDMARDs alone. Blood infection reports allowed categorisation into three groups, notably mild, moderate, and severe. The large value of the chi-square statistic, 3.1169, and the moderately high p-value of 0.2105 indicate that the null hypothesis cannot be rejected at the 0.05 level of significance. Therefore, we conclude that the frequency of blood infection is similar in the two groups. In this test, 83% of cells were less than 5, so other tests (chi- square, likelihood ratio, Mantel- Haenszel, phi Coefficient, Contingency coefficient, Cramer’s V) need to be applied to check the veracity of the findings (Table 3.62).

Table 3.62 Chi-squared for differences in frequency of Blood infections between recipients of csDMARDs alone and bDMARDs alone, sample size= 15

Statistics used DF Value Probability (p-value) Chi-Square 2.00 3.12 0.21 Likelihood Ratio Chi-Square 2.00 2.83 0.24 Mantel-Haenszel Chi-Square 1.00 1.45 0.23 Phi Coefficient 0.46 Contingency Coefficient 0.41 Cramer's V 0.46

3.4.18. Viral Infections

Viral infection was also reported at three different levels of severity. The majority of reported viral infections in both the csDMARDs and bDMARD recipients were moderate in severity (Tables 3.63 to 3.64). Table 3.63 Table of frequency of viral infection in recipients of csDMARDs alone Viral infection in csDMARDs alone Viral Frequency % Cumulative Cumulative Infection Frequency % Mild 30 34.88 30 34.88 Moderate 42 48.84 72 83.72 Severe 14 16.28 86 100.00 Note. Frequency missing = 1567

106

Table 3.64 Table of frequency of viral infection in recipients of bDMARDs alone Viral infection in Biologics Viral Frequency % Cumulative Cumulative infection Frequency % Mild 5 35.71 5 35.71 Moderate 7 50.00 12 85.71 Severe 2 14.29 14 100.00 Note. Frequency missing = 1639

Almost 14 % of patients who were taking csDMARDs alone reported severe viral infection. This is about 16% of all patients who were on csDMARDs alone and almost 87% of all patients who reported severe viral infection (Table 3.65). Almost 2% of patients who were taking bDMARDs alone reported severe viral infection. This is about 14% of all patients who were on bDMARDs alone and almost 12.5% of all patients who reported severe viral infection (Table 3.65).

Table 3.65 Frequency of viral infections in recipients of csDMARDs alone and bDMARDs alone Group Response Status Mild Moderate Severe Total Frequency 30 42 14 86 % 30.00 42.00 14.00 86.00 csDMARDs Row % 34.88 48.84 16.28 Column % 85.71 85.71 87.50 Frequency 5 7 2 14 % 5.00 7.00 2.00 14.00 bDMARDs Row % 35.71 50.00 14.29 Column% 14.29 14.29 12.50 Total Frequency 35 49 16 100 % 35.00 49.00 16.00 100.00

107

The null hypothesis is that the frequency of viral infections is similar in recipients of bDMARDs alone and csDMARDs alone. Viral infections were categorised into three groups, notably mild, moderate, and severe. The values of the chi-square statistic, 0.0356, and the very high p-value of 0.9824, indicate that the null hypothesis should be confirmed at the 0.05 level of significance. Therefore, we conclude that the frequency of viral infection is similar for recipients of csDMARDs alone and bDMARDs alone. In this test, 33% of cells contained a number less than 5. By checking the likelihood ratio and performing other tests (chi- square, likelihood ratio, Mantel- Haenszel, phi Coefficient, Contingency coefficient, Cramer’s V) the findings can be confirmed (Table 3.66).

Table 3.66 Chi-squared test for differences in the frequency of viral infections between patients who were receiving csDMARDs alone and bDMARDs alone, sample size = 100

Statistic DF Value Probability Chi-Square 2.00 0.04 0.98 Likelihood Ratio Chi-Square 2.00 0.04 0.98 Mantel-Haenszel Chi-Square 1.00 0.02 0.89 Phi Coefficient 0.02 Contingency Coefficient 0.02 Cramer's V 0.02

3.5. Chapter discussion and conclusion In this section, the demography of the ARAD data has been discussed. Demography analysis of ARAD data can allow for the comparison of differences between the Australian population and other populations around the world. This comparison is, potentially, helpful for other practitioners when they want to generalise the results of studies. This valuable information also provides a good tool to assess the reliability of inferential analysis findings when ARAD reports are examined (53).

Different studies reveal the strong connection between smoking and RA. For example, Criswell et al. (2002), in a cohort study, showed that abstinence from smoking may reduce the risk of RA among postmenopausal women [24]. A case control study by Padyukov et al. (2004) showed the risk of RA positive with the SE of HLA-DR is strongly influenced by the presence of an environmental factor (e.g., smoking) in the population at risk [16]. 108

In 2006, Costenbader et al., in a cohort study, showed that past and current smoking were related to the development of RA, seropositive RA. In this study it was shown that both smoking intensity and duration were directly related to risk of infection [25]. In a meta- analysis by Sugiyama et al. (2010), it was shown that smoking is a risk for RA, especially seropositive RA in men. For women, the risk for smokers is about 1.3 times greater than for non- smokers [26]. Di Giuseppe et al. (2014) showed that life-long cigarette smoking was associated with the risk of RA, even among smokers with a low life-long exposure [27]. Furthermore other studies show the connection between effectiveness of smoking cession and better responsiveness of bio-treatment [28]. Sustained smoking cessation within four years of RA diagnosis is connected to a reduction in mortality risk, this rate is same as non-smokers. However, smoking more than five years after RA diagnosis increased mortality beyond the risk of non-RA patients [29]. According to the ARAD, the rate of smoking between 2001 to 2014 is 10.5% (328/3111). This is almost 8.9% of all patient visits (2484/27712). Table 3.67 shows the rate of smoking in the general population in Australia during the same time.

Table 3.67 comparing rate of smokers during the years 2001 to 2013, Australia [30] Year %Total smokers 2001 22 2004 20 2007 19 2010 18 2013 15

It seems that the rate of smokers in RA is less than rate of smokers among general population in Australia during those years. Two major possibilities for this discrepancy include (1) ARAD is a subjective report and data are not reliable; and (2) there are several risk factors for causing RA and risk factors other than smoking play more significant roles in Australia, especially since the rate of RA disease in Australia is higher than many other countries [31].

With resoect to alcohol consumption, a few studies have demonstrated that consuming a moderate amount of alcohol is associated with a reduction in the signs and symptoms of 109

arthritis in RA [17] [34]. Overall, based on ARAD reports, the amount of alcohol consumed by RA participants is lower compared to that in the general population (1.32 compared to 2.72 standard drinks per day)[35]. According to the British Society of Rheumatology, alcohol abusers are unsuitable for methotrexate therapy [36]. Rheumatologists should inform RA patients receiving methotrexate (MTX) to limit alcohol intake and to consider changing MTX to a safer medication [36]. Based on this advice, most of the patients consuming excessive alcohol should have been shifted from MTX to bDMARDs; in other words we would expect to see a meaningful difference in drinking alcohol between patients on csDMARDs compared to patients on bDMARDs. However, in ARAD this difference between the two groups is not significant. There is a guideline in Australia to assess alcohol intake in patients before prescribing MTX, but few data question the contribution of alcohol to the risk of hepatotoxicity [37] [38].

Based on ARAD, there is a marginal difference in alcohol consumption between bDMARD and csDMARD users (Mantel-Haenszel Chi-Square P value 0.05). There are several possibilities for this discrepancy. Australian prescribers may not be permitted to change medication based on alcohol consumption. According to Australian therapeutic guidelines, prescribers should assess a patient’s alcohol intake before prescribing methotrexate. According to this guideline, if methotrexate is prescribed for an alcohol abuser, closer kidney and liver assessments are required [39]. In addition, Rajakulendran et al. (2008) includes other medications, such as leflunomide, in this alcohol restriction as well [37]. Other potential reasons for this difference include (1)n relatively few patients are heavy alcohol users,(2) subjective data about drinking alcohol is not reliable, and (3) prescribing biologics was not that common during the study period and (4) most of the alcohol abusers remained in the csDMARDs group. However, the last possibility was not significant in ARAD.

Based on the findings presented in Tables 3.37-3.38 and in Figure 3.4, it can be seen that the rate of heart infection is very low among RA patients. The very small numbers reported makes it difficult to compare the frequency of such infections between recipients of csDMARDs alone and bDMARDs alone. There is a trend toward higher rates of self-reported moderate or severe heart infection in csDMARDs users. However, these findings need to be interpreted with considerable caution, since they are self-reported and participants may not have grasped the distinction between infection involving the heart and other diverse heart conditions. 110

Lung infections were reported frequently in recipients of both csDMARDs and bDMARDs. Interestingly, recipients of bDMARDs alone reported lower rates and milder or less severe lung infections (Table 3.39-3.42). Whether this may be due to a protective effect of bDMARDs is unclear, but this is considered unlikely, even though there may be better outcomes in bDMARD recipients for more severe lung infections. An alternative possibility is that certain synthetic DMARDs, such as methotrexate andlLeflunomide may have conferred greater lung infection susceptibility. Participants may not have been able to easily distinguish between viral infections affecting the respiratory tract and lung infections, which may have resulted in unequal variations in assignment to these two categories of infection.

Urinary tract infections were found to be more common and more severe amongst recipients of csDMARDs alone (65% compared to 25%) (Table 3.45-3.47). Here again, certain csDMARDs may have increased the propensity to UTIs to a greater degree than bDMARDs. Prednisolone use and in particular dosage may also be relevant in this regard. Gastrointestinal tract (GIT) infection was relatively uncommon and not unequivocally associated with any particular treatment group. (Table 3.43-3.44).

In summary, the csDMARDs-alone and bDMARDs-alone treatment groups in the ARAD dataset were found to be well matched and, thus, quite comparable. With respect to self- reported infections of differing severity, lung infections (LRTIs) and urinary tract infections (UTIs) were strongly associated with use of csDMARDs, implying either a biologic DMARD protective effect, which is considered improbable, or a greater propensity to these infections due to the use of one or more synthetic DMARDs, such as methotrexate or leflunomide for example, both of which have been implicated in LRTI.

The descriptive analysis of ARAD reports during 2001 to 2014 shows that, when infections of differing severity are compared between csDMARDs and bDMARD recipients, bDMARDs alone are associated with less risk of infections among Australian patients with RA than csDMARDs alone. Overall, the type of infection, differences in the severity of infections and whether the frequencies differ significantly statistically between patients who are taking csDMARDs alone and patients who are taking bDMARDs alone are shown in table 3.68 and figure 4.1.

111

Table 3.68 Comparison of the frequencies of different infections of varying severity between recipients of csDMARDs alone and bDMARDs alone Distribution of the severity of infection Frequency of the % type of infection Type of Severity csDMARDs bDMARDs p-value of (%) infection difference Mild 16.13 40.74 LRTI Moderate 61.29 37.04 0.0156 9.8 Severe 22.58 22.22 Mild 29.17 0 GITI Moderate 33.33 100 NA 2.82 Severe 37.5 0 Mild 15.19 50 UTI Moderate 70.89 25 0.0002 6.33 Severe 13.92 25 Mild 34.88 35.71 Viral Moderate 48.84 50 0.9819 7.52 Severe 16.28 14.29 Mild 53.15 53.33 Skin/nail Moderate 39.16 36.67 0.9072 13 Severe 7.69 10 Mild 44.39 41.67 EENT Moderate 43.85 41.67 0.8032 14.75 Severe 11.76 16.67 Heart Mild 0 0 Moderate 42.86 100 NA 0. 38 Severe 57.14 0 Mild 19.35 33.33 MSK, Moderate 51.61 33.33 0.4982 3.11 Bone, Joint Severe 29.03 33.33 Mild 0 100 Artificial Moderate 20 0 NA 0.61 joint Severe 80 0 Mild 9.09 50 Blood Moderate 54.55 25 0.2426 0.707 Severe 36.36 25 LRTI: Lower respiratory tract infection; GITI: Gastrointestinal tract infection; UTI: Urinary Tract Infection

In the above tables, p-values indicate whether there is a significant difference in the frequency of infections between recipients of csDMARDs alone and bDMARDs alone.

The literature review also shows that both csDMARDs and bDMARDs can increase the risk of serious infection and non-serious infection. However, according to the literature review,

112

the rate of infection by different medicines is slightly different and, overall, bDMARDs are causing more infections. FORWARD is one of the largest National Databank for Rheumatic Diseases across the world. According to this database, TNFIs have almost the highest rate of serious infection 26.9 (95% CI 24.5‐29.6) compared to non TNFIs 23.3 (95% CI 19.0‐28.5), and csDMARDs 22.4 (95% CI 19.2‐26.1) [18].

In the same study, the smoking rate was compared between csDMARDs and bDMARDs (p value 0.738) and ARAD. No significant difference was identified [18].

It has also been demonstrated that, compared to patients with OA, orthopaedic surgery in RA is associated with a higher risk of infection[40]. This risk also increases in patients who are taking bDMARDs or cs DAMARDs[41]. Due to this risk, the American College of Rheumatology advises stopping TNFα inhibitors one week or more prior to surgery[41]. The British Rheumatology Society also, for the same reason, recommends withholding therapy for 3 to 5 times the half-life of the drug[42], and the Canadian Rheumatology Association reduces this period to 2 half-lives of the drug[43].

Among all the different anti-RA medications, csDMARDs (methotrexate, hydroxychloroquine, sulfasalazine, and azathioprine) are safer[40]. Among bDMARDs, anti- TNFα inhibitor therapy significantly increases the risk of surgical site infection and should be stopped for more than two weeks prior to orthopaedic surgery[40]. Infliximab and etanercept from bDMARDs are usually prescribed in longer disease duration and are associated with further risk of acute surgical site infection (SSI)[44]. Withholding medications before and after a procedure depends on the pharmacokinetic properties of the individual medication and the region of the world[45].

Other risk factors associated with an increased risk of infection include steroid doses over 15 mg/day, coronary artery disease, and being underweight [44]. Therefore, it is important to taper prednisone in the peri-operative management strategy [44]. Sometimes, the risk of csDMARDs and bDMARDs, compared to other risk factors, is ignorable [44].

Based on the different studies in the literature, the risk of infection is not always the same among all bDMARDs. For example, TNF inhibitors and methotrexate are both associated 113

with increased incidence of infections, much more than biologics [46]. However, in a cohort study in the USA conducted with 609 patients with RA before the introduction of biologics, the infection rate was reaching almost 19.64/100 patient-years and, after bDMARDs, was reduced to 12.87/100 patient-years in matched controls. In this study, septic arthritis (14.89; 95% CI: 6.12-73) was the most common infection, followed by osteomyelitis (10.63; 95% CI: 3.39-126)[47]. TNFIs also seem to be associated with an almost 2- to 4-fold increased risk of serious bacterial infections and a slight increase in non-serious infection. Still, a combination of TNF inhibitors with methotrexate can increase the risk of serious infection, significantly. [48][49].

Etanercept and infliximab are other samples of biologics. The risk of serious infection in monotherapy with these medications is the same as for methotrexate [50][51]. These serious infections include bacterial infection, fungal infection, bronchitis, cellulitis, herpes zoster infection, pneumonia, peritonitis, pyelonephritis, sepsis, and tuberculosis [50]. In both etanercept and infliximab, if there is a combination therapy with methotrexate, the risk reaches higher than the risk with methotrexate alone (P 0.05 for both infliximab doses)[52]. On the other hand, adalimumab from bDMARDs assumes to cause limited incidence of serious infections. The overall rate of infections in the pooled adalimumab (1.55/patient-year) is similar to methotrexate monotherapy (1.38/patient year)[53].

With the use adalimumab, the incidence of serious infections is almost 2.03/100 patient-years [48]and, from the most common to the least common, infections include pneumonia, urinary tract infections, and septic arthritis. The safety of adalimumab has been approved in other studies. For example, in a study on 10,000 patients with approximately 12,500 patient-years of adalimumab exposure only 5.1/100 patient-years developed serious infection[49].

Anakinra is another sample from bDMARDs. Anakinra has been connected to serious infection in organs, such as lung and skin (5.37/100 patient-years in compare to 1.65/100 patient-years)[54]. However, it seems that most of this connection to infection occurs when a patient is taking a baseline corticosteroid, otherwise the serious infection rate was substantially lower (2.87/100 patient-years compare to 7.13/100 patient-years)[54].

114

Abatacept also has a higher rate of serious infection compared to many other bDMARDs (78). The rate of serious infection in abatacept sometimes is higher than the infection rate in the methotrexate monotherapy group (2.5% versus 0.9%; 95% CI: 0.3-3.6) [55]. Overall, the incidence of both serious and nonserious bacterial infections increases in abatacept and it is better to avoid prescribing abatacept and TNFIs together [55].

Rituximab, in comparison to other bDMARDs, has been associated with less serious infection. However, there are a few reports regarding a 4- to 7-fold increased risk of reactivation of latent tuberculosis when using TNFIs together with infliximab, and this rate is even more than the combination of etanercept and TNFIs[56].

Overall, as a result of the modes of action in medicine, among bDMARDs TNF inhibitors, anakinra, abatacept, and rituximab can change immune response, leaving patients at an increased risk of infection. This risk increases by combining some bDMARDs and TNFIs. For example, when infliximab is added to methotrexate in compare to methotrexate monotherapy, the risk of serious infection increases, significantly [52]. The most common types of infections in bDMARDs are respiratory tract infections (including pneumonia), following by skin and soft-tissue infections and urinary tract infections [52].

There is also a risk of tuberculosis with TNFIs. Some evidence reveals that that this risk is the highest with infliximab and less with anakinra[52]. Rituximab and abatacept seem to have a lower risk of viral serious infection compared to TNFIs. However, in long term studies, this was not approved [57][58]. Rituximab monotherapy seems to be associated with serious infection when it is prescribed for a longer period of time but, overall, the rate is lower than for many other bDMARDs (94). However, decreases in the levels of IgM during prolonged treatment with rituximab is associated with a higher incidence of opportunistic infections, such as non-Hodgkin's lymphoma (NHL) [58][59]. In most studies, corticosteroids (CS) are assessed among csDMARDs. CS use is associated with an increased relative risk (RR) (1.67, 1.47–1.87) of infection[60]. MTX in a Canadian study was associated with slight increase of risk of pneumonia (RR 1.2; 95% CI 1.0–1.3) [61], while another study from US indicated a decreased infection risk in MTX users [49]. In conclusion, the slightly increased risk of infection in MTX is counterbalanced by the effective control of rheumatic disease, leading to improved function. 115

In one study, the incidence of severe infection with LEF can reach up to 3.3% person-year [62][21]. Hydroxychloroquine (HCQ), sulfasalazine (SSZ), and cyclosporine A (CsA) in a US study were not associated with risk of infection[21]. If there is an association, that is very mild unless the patient is suffering from other conditions, such as transplanted organs[63].

In summary, according to the literature and ARAD results, both bDMARDs and csDMARDs will increase the risk of bacterial infection, especially pneumonia. With some exceptions, it seems that, overall, bDMARDs are associated with higher rates of infection compared to csDMARDs. However, the ARAD analysis depends on the severity of infection, this ratio can change or the differences not be regarded significant (Table 3.68). The reason for this discrepancy might be due to geographical differences; for example, in TB infection, TB risks in TB-endemic areas with TNFIs is much higher than other regions[64].

116

References: [1] M. A. Othman, W. S. W. Ghazali, N. K. Yahya, and K. K. Wong, “Correlation of Demographic and Clinical Characteristics with Rheumatoid Factor Seropositivity in Rheumatoid Arthritis Patients,” Malays J Med Sci, vol. 23, no. 6, pp. 52–59, Nov. 2016, doi: 10.21315/mjms2016.23.6.6.

[2] J. J. Sacks, Y.-H. Luo, and C. G. Helmick, “Prevalence of specific types of arthritis and other rheumatic conditions in the ambulatory health care system in the United States, 2001-2005,” Arthritis Care Res (Hoboken), vol. 62, no. 4, pp. 460–464, Apr. 2010, doi: 10.1002/acr.20041.

[3] S. Ehsan, K. Bhatia, N. Rahman, and National Centre for Monitoring Arthritis and Musculoskeletal Conditions (Australia), A snapshot of arthritis in Australia 2010. Canberra: Australian Institute of Health and Welfare, 2010.

[4] L. Klareskog et al., “A new model for an etiology of rheumatoid arthritis: Smoking may trigger HLA–DR (shared epitope)–restricted immune reactions to autoantigens modified by citrullination,” Arthritis Rheum, vol. 54, no. 1, pp. 38–46, Jan. 2006, doi: 10.1002/art.21575.

[5] M. do S. T. M. Almeida, J. V. M. Almeida, and M. B. Bertolo, “Demographic and clinical features of patients with rheumatoid arthritis in Piauí, Brazil – evaluation of 98 patients,” Revista Brasileira de Reumatologia (English Edition), vol. 54, no. 5, pp. 360– 365, Sep. 2014, doi: 10.1016/j.rbre.2014.02.018.

[6] A. Kleyer et al., “Bone loss before the clinical onset of rheumatoid arthritis in subjects with anticitrullinated protein antibodies,” Annals of the Rheumatic Diseases, vol. 73, no. 5, pp. 854–860, May 2014, doi: 10.1136/annrheumdis-2012-202958.

[7] R. E. Costello et al., “O08 Symptoms in first degree relatives of patients with rheumatoid arthritis: evaluation of data from the symptoms in persons at risk of rheumatoid arthritis questionnaire,” Rheumatology (Oxford), vol. 58, no. Supplement_3, Apr. 2019, doi: 10.1093/rheumatology/kez105.007.

117

[8] C. Girard, G. Haroutunian, and P. A. Guerne, “[Rheumatoid arthritis : cardiovascular and pulmonary manifestations].,” Rev Med Suisse, vol. 15, no. 641, pp. 542–548, Mar. 2019.

[9] D. L. Scott, F. Wolfe, and T. W. J. Huizinga, “Rheumatoid Arthritis,” Lancet, vol. 376, no. 9746, pp. 1094–1108, Sep. 2010, doi: 10.1016/S0140-6736(10)60826-4.

[10] S. de Brito Rocha, D. C. Baldo, and L. E. C. Andrade, “Clinical and pathophysiologic relevance of autoantibodies in rheumatoid arthritis,” Adv Rheumatol, vol. 59, no. 1, p. 2, 17 2019, doi: 10.1186/s42358-018-0042-8.

[11] O. Arntz, B. Pieters, R. Thurlings, P. Kraan, and F. Loo, “P079 Is IgM rheumatoid factor present on circulating extracellular vesicles of rheumatoid arthritis patients a potential biomarker for disease severity?,” in Abstracts, Mar. 2019, p. A34.1-A34, doi: 10.1136/annrheumdis-2018-EWRR2019.68.

[12] A. I. Rutherford, E. Patarata, S. Subesinghe, K. L. Hyrich, and J. B. Galloway, “Opportunistic infections in rheumatoid arthritis patients exposed to biologic therapy: results from the British Society for Rheumatology Biologics Register for Rheumatoid Arthritis,” Rheumatology (Oxford), vol. 57, no. 6, pp. 997–1001, Jun. 2018, doi: 10.1093/rheumatology/key023.

[13] M. D. George and J. F. Baker, “Perioperative management of immunosuppression in patients with rheumatoid arthritis,” Curr Opin Rheumatol, vol. 31, no. 3, pp. 300–306, 2019, doi: 10.1097/BOR.0000000000000589.

[14] S. Offermanns and W. Rosenthal, Eds., “Disease-modifying Anti-rheumatic Drugs (DMARDs),” in Encyclopedia of Molecular Pharmacology, Berlin, Heidelberg: Springer Berlin Heidelberg, 2008, pp. 428–428.

[15] D. L. Scott, “Biologics-Based Therapy for the Treatment of Rheumatoid Arthritis,” Clinical Pharmacology & Therapeutics, vol. 91, no. 1, pp. 30–43, 2012, doi: 10.1038/clpt.2011.278.

[16] L. Padyukov, C. Silva, P. Stolt, L. Alfredsson, and L. Klareskog, “A gene–environment interaction between smoking and shared epitope genes in HLA–DR provides a high risk

118

of seropositive rheumatoid arthritis,” Arthritis & Rheumatism, vol. 50, no. 10, pp. 3085– 3092, 2004, doi: 10.1002/art.20553.

[17] J. R. Maxwell, I. R. Gowers, D. J. Moore, and A. G. Wilson, “Alcohol consumption is inversely associated with risk and severity of rheumatoid arthritis,” Rheumatology (Oxford), vol. 49, no. 11, pp. 2140–2146, Nov. 2010, doi: 10.1093/rheumatology/keq202.

[18] K. L. Grøn et al., “Risk of serious infections in patients with rheumatoid arthritis treated in routine care with abatacept, rituximab and tocilizumab in Denmark and Sweden,” Annals of the Rheumatic Diseases, vol. 78, no. 3, pp. 320–327, Mar. 2019, doi: 10.1136/annrheumdis-2018-214326.

[19] P. Coyne, J. Hamilton, C. Heycock, V. Saravanan, E. Coulson, and C. A. Kelly, “Acute lower respiratory tract infections in patients with rheumatoid arthritis.,” The Journal of Rheumatology, p. 6.

[20] M. F. Doran, C. S. Crowson, G. R. Pond, W. M. O’Fallon, and S. E. Gabriel, “Predictors of infection in rheumatoid arthritis,” Arthritis & Rheumatism, vol. 46, no. 9, pp. 2294– 2300, 2002, doi: 10.1002/art.10529.

[21] F. Wolfe, L. Caplan, and K. Michaud, “Treatment for rheumatoid arthritis and the risk of hospitalization for pneumonia: associations with prednisone, disease-modifying antirheumatic drugs, and anti-tumor necrosis factor therapy,” Arthritis Rheum., vol. 54, no. 2, pp. 628–634, Feb. 2006, doi: 10.1002/art.21568.

[22] M. E. Stokes, C. S. Davis, and G. G. Koch, Categorical Data Analysis Using SAS, Third Edition, 3 edition. Cary, NC: SAS Institute, 2012.

[23] R. Prentice and V. Andersen, “Interpreting heritage essentialisms: Familiarity and felt history,” Tourism Management, vol. 28, no. 3, pp. 661–676, Jun. 2007, doi: 10.1016/j.tourman.2006.02.008.

[24] L. A. Criswell et al., “Cigarette smoking and the risk of rheumatoid arthritis among postmenopausal women:: Results from the Iowa Women’s Health Study,” The American

119

Journal of Medicine, vol. 112, no. 6, pp. 465–471, Apr. 2002, doi: 10.1016/S0002- 9343(02)01051-3.

[25] K. H. Costenbader, D. Feskanich, L. A. Mandl, and E. W. Karlson, “Smoking Intensity, Duration, and Cessation, and the Risk of Rheumatoid Arthritis in Women,” The American Journal of Medicine, vol. 119, no. 6, p. 503.e1-503.e9, Jun. 2006, doi: 10.1016/j.amjmed.2005.09.053.

[26] D. Sugiyama et al., “Impact of smoking as a risk factor for developing rheumatoid arthritis: a meta-analysis of observational studies,” Ann Rheum Dis, vol. 69, no. 01, pp. 70–81, Jan. 2010, doi: 10.1136/ard.2008.096487.

[27] D. Di Giuseppe, A. Discacciati, N. Orsini, and A. Wolk, “Cigarette smoking and risk of rheumatoid arthritis: a dose-response meta-analysis,” Arthritis Res Ther, vol. 16, no. 2, p. R61, Mar. 2014, doi: 10.1186/ar4498.

[28] Y. Saeki et al., “SAT0074 Smoking Cessation Significantly Reduces Failure of BIOLOGICS (BIO)-Treatment in Rheumatoid Arthritis (RA): from the ‘Ninja’ Registry Cohort of Japanese Patients,” Annals of the Rheumatic Diseases, vol. 73, no. Suppl 2, pp. 617–617, Jun. 2014, doi: 10.1136/annrheumdis-2014-eular.1441.

[29] J. A. Sparks et al., “Smoking behavior changes in the early rheumatoid arthritis period and risk of mortality during 36 years of prospective follow-up,” Arthritis Care Res (Hoboken), vol. 70, no. 1, pp. 19–29, Jan. 2018, doi: 10.1002/acr.23269.

[30] I. S. Greenhalgh, EM, Bayly, M, & Winstanley, MH. MM and Winstanley, MH [editors], “Prevalence of smoking—adults,” 2019. .

[31] AIHW, “Rheumatoid arthritis, What is rheumatoid arthritis? - Australian Institute of Health and Welfare,” Aug. 30, 2019. .

[32] C. Mabille, Y. Degboe, A. Constantin, T. Barnetche, A. Cantagrel, and A. Ruyssen- Witrand, “Infectious risk associated to orthopaedic surgery for rheumatoid arthritis patients treated by anti-TNFalpha,” Joint Bone Spine, vol. 84, no. 4, pp. 441–445, Jul. 2017, doi: 10.1016/j.jbspin.2016.06.011.

120

[33] C. D. Albracht, T. N. Hreha, and D. A. Hunstad, “Sex effects in pyelonephritis,” Pediatr Nephrol, Feb. 2020, doi: 10.1007/s00467-020-04492-9.

[34] H. Källberg et al., “Alcohol consumption is associated with decreased risk of rheumatoid arthritis: results from two Scandinavian case-control studies,” Ann. Rheum. Dis., vol. 68, no. 2, pp. 222–227, Feb. 2009, doi: 10.1136/ard.2007.086314.

[35] S. Begg, Australian Institute of Health and Welfare, and School of Population Health, Eds., The burden of disease and injury in Australia 2003. Canberra: Australian Institute of Health and Welfare, 2007.

[36] C. R. Holroyd et al., “The British Society for Rheumatology biologic DMARD safety guidelines in inflammatory arthritis,” Rheumatology (Oxford), vol. 58, no. 2, pp. e3–e42, Feb. 2019, doi: 10.1093/rheumatology/key208.

[37] S. Rajakulendran, K. Gadsby, and C. Deighton, “Rheumatoid arthritis, alcohol, leflunomide and methotrexate. Can changes to the BSR guidelines for leflunomide and methotrexate on alcohol consumption be justified?,” Musculoskeletal Care, vol. 6, no. 4, pp. 233–246, 2008, doi: 10.1002/msc.135.

[38] L. Tilling, S. Townsend, and J. David, “Methotrexate and hepatic toxicity in rheumatoid arthritis and psoriatic arthritis,” Clin Drug Investig, vol. 26, no. 2, pp. 55–62, 2006, doi: 10.2165/00044011-200626020-00001.

[39] L. K. Mercer et al., “Risk of lymphoma in patients exposed to antitumour necrosis factor therapy: results from the British Society for Rheumatology Biologics Register for Rheumatoid Arthritis,” Ann Rheum Dis, vol. 76, no. 3, pp. 497–503, Mar. 2017, doi: 10.1136/annrheumdis-2016-209389.

[40] M. D. George and J. F. Baker, “Perioperative management of immunosuppression in patients with rheumatoid arthritis:,” Current Opinion in Rheumatology, vol. 31, no. 3, pp. 300–306, May 2019, doi: 10.1097/BOR.0000000000000589.

121

[42] J. Ledingham and C. Deighton, “Update on the British Society for Rheumatology guidelines for prescribing TNFα blockers in adults with rheumatoid arthritis (update of previous guidelines of April 2001),” Rheumatology (Oxford), vol. 44, no. 2, pp. 157– 163, Feb. 2005, doi: 10.1093/rheumatology/keh464.

[43] V. P. Bykerk et al., “Canadian Rheumatology Association Recommendations for Pharmacological Management of Rheumatoid Arthritis with Traditional and Biologic Disease-modifying Antirheumatic Drugs,” J Rheumatol, vol. 39, no. 8, pp. 1559–1582, Aug. 2012, doi: 10.3899/jrheum.110207.

[44] S. Momohara et al., “Prosthetic joint infection after total hip or knee arthroplasty in rheumatoid arthritis patients treated with nonbiologic and biologic disease-modifying antirheumatic drugs,” Modern Rheumatology, vol. 21, no. 5, pp. 469–475, Oct. 2011, doi: 10.3109/s10165-011-0423-x.

[45] M. Pombo-Suarez and J. J. Gomez-Reino, “Abatacept for the treatment of rheumatoid arthritis,” Expert Review of Clinical Immunology, vol. 15, no. 4, pp. 319–326, Apr. 2019, doi: 10.1080/1744666X.2019.1579642.

[46] J. S. Smolen et al., “EULAR recommendations for the management of rheumatoid arthritis with synthetic and biological disease-modifying antirheumatic drugs: 2016 update,” Ann Rheum Dis, vol. 76, no. 6, pp. 960–977, Jun. 2017, doi: 10.1136/annrheumdis-2016-210715.

[47] K. Yamaoka, “Tofacitinib for the treatment of rheumatoid arthritis: an update,” Expert Rev Clin Immunol, vol. 15, no. 6, pp. 577–588, 2019, doi: 10.1080/1744666X.2019.1607298.

[48] D. Shouval and O. Shibolet, “Immunosuppression and HBV reactivation,” Semin. Liver Dis., vol. 33, no. 2, pp. 167–177, May 2013, doi: 10.1055/s-0033-1345722.

[49] A. L. Smitten et al., “The risk of hospitalized infection in patients with rheumatoid arthritis,” J. Rheumatol., vol. 35, no. 3, pp. 387–393, Mar. 2008.

122

[50] K. Koetz, E. Bryl, K. Spickschen, W. M. O’Fallon, J. J. Goronzy, and C. M. Weyand, “T cell homeostasis in patients with rheumatoid arthritis,” Proc. Natl. Acad. Sci. U.S.A., vol. 97, no. 16, pp. 9203–9208, Aug. 2000, doi: 10.1073/pnas.97.16.9203.

[51] K. L. Winthrop et al., “ESCMID Study Group for Infections in Compromised Hosts (ESGICH) Consensus Document on the safety of targeted and biological therapies: an infectious diseases perspective (Soluble immune effector molecules [II]: agents targeting interleukins, immunoglobulins and complement factors),” Clin. Microbiol. Infect., vol. 24 Suppl 2, pp. S21–S40, Jun. 2018, doi: 10.1016/j.cmi.2018.02.002.

[52] F. D. Nard et al., “Risk of hepatitis B virus reactivation in rheumatoid arthritis patients undergoing biologic treatment: Extending perspective from old to newer drugs,” World J Hepatol, vol. 7, no. 3, pp. 344–361, Mar. 2015, doi: 10.4254/wjh.v7.i3.344.

[53] Z. Zhang, W. Deng, Q. Wu, and L. Sun, “Tuberculosis, hepatitis B and herpes zoster in tofacitinib-treated patients with rheumatoid arthritis,” Immunotherapy, vol. 11, no. 4, pp. 321–333, 2019, doi: 10.2217/imt-2018-0113.

[54] S. S. Jick, E. S. Lieberman, M. U. Rahman, and H. K. Choi, “Glucocorticoid use, other associated factors, and the risk of tuberculosis,” Arthritis Rheum., vol. 55, no. 1, pp. 19– 26, Feb. 2006, doi: 10.1002/art.21705.

[55] C.-C. Lai, M.-T. G. Lee, S.-H. Lee, S.-H. Lee, S.-S. Chang, and C.-C. Lee, “Risk of incident active tuberculosis and use of corticosteroids,” Int. J. Tuberc. Lung Dis., vol. 19, no. 8, pp. 936–942, Aug. 2015, doi: 10.5588/ijtld.15.0031.

[56] T. Koike et al., “Postmarketing surveillance of safety and effectiveness of etanercept in Japanese patients with rheumatoid arthritis,” Mod Rheumatol, vol. 21, no. 4, pp. 343– 351, Aug. 2011, doi: 10.1007/s10165-010-0406-3.

[57] H.-G. Yoo, H. M. Yu, J. B. Jun, H.-S. Jeon, and W.-H. Yoo, “Risk factors of severe infections in patients with rheumatoid arthritis treated with leflunomide,” Mod Rheumatol, vol. 23, no. 4, pp. 709–715, Jul. 2013, doi: 10.1007/s10165-012-0716-8.

123

[58] “Hydroxychloroquine inhibits calcium signals in T cells: a new mechanism to explain its immunomodulatory properties. - Abstract - Europe PMC.” https://europepmc.org/article/med/10828029 (accessed Jun. 18, 2020).

[59] F. M. P. Meier, M. Frerix, W. Hermann, and U. Müller-Ladner, “Current immunotherapy in rheumatoid arthritis,” Immunotherapy, vol. 5, no. 9, pp. 955–974, Sep. 2013, doi: 10.2217/imt.13.94.

[60] W. G. Dixon, S. Suissa, and M. Hudson, “The association between systemic glucocorticoid therapy and the risk of infection in patients with rheumatoid arthritis: systematic review and meta-analyses,” Arthritis Res Ther, vol. 13, no. 4, p. R139, 2011, doi: 10.1186/ar3453.

[61] S. Bernatsky, M. Hudson, and S. Suissa, “Anti-rheumatic drug use and risk of serious infections in rheumatoid arthritis,” Rheumatology (Oxford), vol. 46, no. 7, pp. 1157– 1160, Jul. 2007, doi: 10.1093/rheumatology/kem076.

[62] K. A. Jenks, L. K. Stamp, J. L. O’Donnell, R. L. Savage, and P. T. Chapman, “Leflunomide-associated infections in rheumatoid arthritis,” J. Rheumatol., vol. 34, no. 11, pp. 2201–2203, Nov. 2007.

[63] C. J. Edwards, C. Cooper, D. Fisher, M. Field, T. P. van Staa, and N. K. Arden, “The importance of the disease process and disease-modifying antirheumatic drug treatment in the development of septic arthritis in patients with rheumatoid arthritis,” Arthritis Rheum., vol. 57, no. 7, pp. 1151–1157, Oct. 2007, doi: 10.1002/art.23003.

[64] T.-L. Liao, C.-H. Lin, G.-H. Shen, C.-L. Chang, C.-F. Lin, and D.-Y. Chen, “Risk for Mycobacterial Disease among Patients with Rheumatoid Arthritis, Taiwan, 2001-2011,” Emerging Infect. Dis., vol. 21, no. 8, pp. 1387–1395, Aug. 2015, doi: 10.3201/eid2108.141846.

124

CHAPTER 4

Inferential Analysis of Infection in Rheumatoid Arthritis

125

Abstract Objective: To conduct an inferential analysis of the association between the risk of infection and each Anti-RA medication. The current analysis provides valuable information concerning the relative frequency of self-reported infections in users of diverse anti-rheumatic therapies. Various organs including eye, ear, nose, throat, lungs, urinary tract, heart, gastrointestinal tract, CNS were examined as well as systemic infections of a viral and pyogenic nature (sepsis / septicaemia) are investigated.

Methods Self-reported and unverified data concerning infections was collected from 3110 Australian Rheumatology Association Database (ARAD) participants, who reported sequentially from 2001 to 2014. Through the processes of data cleaning all duplicated answers, single answers and faulty reports were deleted. Overall 27,709 visit reports were available. Data was tested by multinominal logistic regression in SAS software. Mild, moderate and severe infections assigned according to a priori descriptive criteria were categorised in relation to organ involvement / body system affected and examined in relation to current therapy.

Results: The most frequent infections reported by ARAD participants were EENT system infections (eye, ear, nose and throat,14.75%) followed by skin and nail infections (13%), lung infections (9.83%), and viral infections of any type (7.52%). Based on the same database, the most commonly used bDMARDs were Etanercept, followed by Adalimumab. Amongst csDMARDs the most commonly used medications were: Methotrexate, Hydroxychloroquine and Sulphasalazine. Among all these medications safest medication in most common infections were as following. Etanercept and Methotrexate the safest for EENT infection, Etanercept and Adalimumab the safest in lung infection, and Leflunomide safest option in skin and nail infections (Table 4.53).

Conclusion: Both csDMARDs and bDMARDs are shown to be associated with higher risk of infection in RA. It seems that prednisolone (with consumption prevalence of 3.33%) followed by cyclosporine (with a consumption prevalence of 0.05%) are the most common medications in most of the moderate to severe infections throughout body. In comparison to

126

csDMARDs, if we consider prevalence of consumption, bDMARDs are rarely causing moderate or severe infections. Some medications are playing a paradoxical role. For example, although taking Adalimumab usually increases the risk of infection in skin and nail infection, in comparison to other medications, it was found to be associated to a reduction in prevalence of artificial joint or GIT infections. Overall, bDMARDs seems to be safer with lower risk of infection as compare to csDMARDs.

127

1. Introduction Rheumatoid arthritis (RA) affects approximately 0.8% of adults and is a cause of significant morbidity and mortality [1]. RA also affects approximately three times more females than males [3]. Disease onset is commonly between 40 and 70 years of age, though it can begin at any age. Understanding the pathogenesis of RA has progressed over the past few decades resulted in the development of more effective anti-RA medications [4].

Conventionally, in RA, non-steroidal anti-inflammatory drugs, glucocorticoids, and disease- modifying antirheumatic drugs (DMARDs) are used to treat the disease [5]. The most widely used DMARD is methotrexate (MTX), which is the basis of most treatment programs for rheumatoid arthritis [6]. MTX has the highest retention rates compared to other available medications [7].

Despite progress in developing more efficacious treatment for RA, the risk of infections in patients receiving biologic or conventional treatments has not been substantially reduced [7]. One theory for this disturbing statistic is that immunosuppressive or disease-modifying treatments are often required in those most vulnerable to infections, such as elderly patients and patients with multiple comorbidities. This naturally increases the risk of infection after treatment with anti-RA medications. Rheumatologists should be aware of the specific patterns of infection risk that treatment with anti-RA medications confer [8]. This is especially important with newer treatment modalities. By understanding the risk of infection in different organs and the severity of those infections, potential risk factors and their connection to other treatments, health practitioners can adjust medications and institute preventative measures accordingly. For example, measures such as appropriate screening for and treatment of chronic hepatitis B virus infection, to ensure optimal vaccination against respiratory pathogens (influenza virus and pneumococcus) and, where appropriate, offer chemoprophylaxis in patients susceptible to Pneumocystis jirovecii pneumonia [9]. Patients who have had a splenectomy or in whom there is chronic sino-pulmonary infection, including bronchiectasis can be identified for increased vigilance, vaccination and fast-tracking when infections flare or develop [9].

128

Conventionally, in RA, analgesics, non-steroidal anti-inflammatory drugs, glucocorticoids (GC), and disease-modifying anti-rheumatic drugs (DMARDs) are used to treat the disease. The former two only suppress symptoms, whereas GC and DMARDs suppress symptoms and importantly also modify the course of the disease. The most widely used DMARD is methotrexate (MTX). MTX usually forms the basis of most treatment programs for rheumatoid arthritis.58–60. It is noteworthy that MTX has the highest retention rates compared to other available medications [3].

Previous and ongoing research into therapeutic possibilities for RA has led to the development of potent, biologic medications. Using effective medication should be associated with goal- oriented treatment plans with regular appraisals of disease activity[10]. The treatment goals for RA have shifted from mainly symptomatic relief to minimising or eliminating disease activity and in turn altering the progression of the disease. This can potentially improve long-term outcomes and reduce morbidity rates [10]. Better treatment strategies have significantly moderated the severity of RA in the overall population. This has resulted in lower rates of joint replacement and reduced hospital admissions for RA.

Lower frequencies for vasculitis have also been reported [11]. Better use of csDMARDs treatment has probably contributed to this improvement because the beginning of the decline in these measures of disease was noted prior to the use of biologic DMARDs however bDMARDs may well have reinforced these effects. The use of bDMARDs has further reduced symptoms and has improved functional and work capacity[12]. The pro-inflammatory cytokines, such as IL-1, IL-6, and tumour necrosis factor (TNF), have been shown to play an integral role in RA pathogenesis. Therefore, the development of biologic agents, which target these mediators could be expected to impact disease activity significantly. IL-1 antagonists proved to be disappointing, but TNF and IL-6R blockers and IL-6 monoclonal antibodies have shown much superior efficacy[13].

The approach to the RA treatment has changed during the time and is different among different nations. During 2001 to 2004, National RA treatment in Australia was based on taking simple analgesics (e.g., paracetamol), Omega-3 supplements, patient education, physical therapy and exercise, applying Ice and/or heat, and enhanced primary care referrals (e.g., occupational therapy and physiotherapy[14]. 129

For medicine, usually NSAIDs or COX-2 inhibitors were prescribed in the early stages. If symptoms continued, the patient was referred to a rheumatologist, where csDMARDs plus a low dose corticosteroid was prescribed. If disease signs and symptoms were still not controlled, a rheumatologist could start advanced therapy with DMARDs, leflunomide, cyclosporine or even the biologic agents, anakinra, anti-TNFs and rituximab[14].

At the same time, almost another 22 different RA management guidelines (American, APLAR, Australian, Brazilian, British Columbia, British Society for Rheumatology: established and early, Canadian, EULAR, French, German, Hong Kong, Indian, Latin American, Mexican, England, Scotland, South African, Spanish, Swedish, Treat to target, Turkish) show that several general principles were followed. In all these guidelines, remission or low disease activity is the preferred target. csDMARDs usually started as soon as possible after the diagnosis and disease activity monitoring, regularly. There is an emphasis in all of these guidelines that methotrexate is the best initial treatment, and that this can be usefully enhanced with temporary glucocorticoid treatment. Biologic DMARDs were usually used in persistently active disease in patients who have already received methotrexate/other csDMARDs. As soon as the patient achieved a sustained remission, biologics can be tapered[15].

There are a few minor differences about the value and place for using combinations of csDMARDs. For example, EULAR guidance is uncertain about using csDMARDs, however, according to ACR guidance, using csDMARDs is essential. NICE guidance recommends only starting biologics in patients with disease that has not responded to intensive therapy, using a combination of conventional DMARDs[16].

Another difference in these guidelines is in the treatment of moderately active RA. While, different guidelines have ignored to separate moderately active RA from others, the ACR guidance strongly recommends considering treating moderate RA disease intensively[15].

According to the Australian guidelines for RA treatment in 2020, the guidelines were changed briefly. In the Australian guideline, the initial treatment starts with simple analgaesics (e.g., paracetamol), and omega-3 supplements. In the meantime, patient education (e.g., Arthritis

130

Australia), physical exercise, applying ice and/or heat, and enhanced primary care referrals (e.g., physiotherapy, occupational therapy, podiatry, psychology and others) are important[19]. Using NSAIDs or COX-2 inhibitors should start after assessment of potential side effects in first line treatment. If advanced therapy is required, a rheumatologist can prescribe a combination of csDMARDs (eg. methotrexate, hydroxychloroquine, sulfasalazine) with biological agents[19].

In this chapter, the impact of risk factors and, in particular, of anti-RA medications on the frequency of infections in different organs/systems is examined in detail. In addition, based on patient reports, the severity of infections, the frequency of different types of infections and the association with different anti-RA medications have been investigated.

1.1. Aims The aims of this chapter are to determine the: • frequency of self-reported infections in different organs in RA;

• frequency of prescribed anti-RA medication uses among patients in ARAD;

• impact of different anti- RA medications on self-reported infections; and

• impact of different anti-RA medications on infection severity.

1.2. Hypothesis The aims are based on the following hypothesis: Infections are very common in RA and there may be differential effects of anti-rheumatic drugs on the type and severity of infections that occur in RA.

The following topics will be discussed: 1- Frequency of different types of infection in RA and categorization of these types of infections, 2- Frequency of different prescribed anti RA medications, 3- Impacts of anti-RA medications on different types of infection, and 4- impacts of anti-RA medications on severity of infections.

131

2. Methods

2.1 Data Collection Data were collected from the Australian Rheumatology Association Database (ARAD), in which a cohort of 3569 RA patients (960 males and 2609 females), who had regularly completed questionnaires (28,176 person-reports in total during 2001-2014) and had self- reported in respect to infections, were investigated for the type and severity of infection and how these related to currently used anti-RA medications. Among the 3569 patients, 459 patients were eliminated because they had only completed a questionnaire once. After removing 8 duplications 27,709 reports from 3110 patients remained. All underwent a series of inferential analysis with descriptive tests and logistic regression using SAS software. Only patients who were currently taking anti-RA medications were selected and the number and severity of infections in different organs were analysed.

2.2. Statistical Analysis

All 27,709 visits from 3110 patients were subjected to a series of inferential analyses with descriptive tests and logistic regression in SAS software to extract the required data. Currently used anti-RA medications were selected and the impacts on different levels of severity of infection were analysed.

3. Results and Discussions Based on Table 4.1 and Figure 4.1, the most prevalent infection in RA was EENT infection, followed by skin and nail infection and lung infection. These data are based on patient reports from 2001 to 2014. The guidelines in each region and the changes in prescription habits during time influence these frequencies significantly.

132

Frequency of self‐reported infections

16.00% 14.75% 14.00% 13%

12.00% 9.83% 10.00% 7.52% 8.00% 6.33% 6.00%

4.00% 2.82% 3.11%

2.00% 0.61% 0.71% 0.12% 0.16% 0.38% 0.00%

Figure 4.1 Frequency of self-reported organ infections in RA based on patient visits during 2001-2014

In order to have a better estimation of the meaning of association between a particular type of infection and medications, it is necessary to know the frequency with which medications have been used by ARAD participants.

Based on ARAD, the most common to the least common prescribed medications during 2001 to 2014 among RA patients were: Etanercept, Adalimumab, Methotrexate and Folic acid, Hydroxychloroquine, Sulphasalazine, Rituximab, Abatacept, Prednisolone, Tocilizumab, Infliximab, Leflunomide, Anakinra, Azathioprine, Cyclosporine, IM Gold and Penicillamine (Figure 4.2). In this sample, investigation is based on the patient visit reports. The category of currently taking medication does not include visits of patient who were previously taking or had stopped taking that particular type of medication. Also, the data inputs entirely depend on the patient reports and may include a smaller sample than other studies in these categories.

133

Table 4.1 Frequency of taking each medication based on patient visit data Type of Name of Population Currently Population Never Taken% Medication Medication taking% Azathioprine 0.068% 4.64% (1286/27711) (19/27711)

Cyclosporine 0.05% 4.41% (16/27711) (1224/27711) csDMARDs Leflunomide (Arava 1.18% 1.01% (Leflunomide)) (327/27711) (281/27711)

Methotrexate/Folic 19.19% 58.41% Acid (5318/27711) (16188/27711) Hydroxychloroquine 17.06% 40.19% (4730/27711) (11139/27711) Sulphasalazine 10.63% 40.96% (2947/27711) (11352/27711)

Abatacept 3.66% 92.77% (1016/27711) (25708/27711)

Adalimumab 22.18% 57.71% (6149/27711) (15993/27711) bDMARDs Anakinra 0.14% 96.56% (39/27711) (26758/27711) Certolizumab 0.88% 97.35% (246/27711) (26979/27711) Etanercept 30.42% 47.61% Golimumab 1.72% 96.08% (479/27711) (26625/27711) Infliximab 2.67% 89.86% (742/27711) (24903/27711) Rituximab 4.26% 9.34% (1183/27711) (2589/27711) Tocilizumab 2.72% 94.83% (756/27711) (26281/27711) Prednisolone 3.33% 0.45% (924/27711) (127/27711) Independent IM Gold Injection 0.05% 3.52% Group (14/27711) (978/27711) Penicillamine 0.0036% 0.47% (1/27711) (1303/27711)

134

Frequency of medications in RA

22 19 17

0.003 0.05 0.05 0.06 0.14 1.1 2.6 2.7 3.3 3.6 4.26

Figure 4.2 Frequency of medication use in RA patients

In order to calculate the effect of medication on different organ infections, logistic regression was used first and, if there were differences, pairwise chi-squares were calculated. This method is better than pairwise chi-square at the outset because, with pairwise chi-square, the risk of mistakes in each test with a p value of 0.05 is up to 5%. When this test is performed more often, the potential error risk adds up and, for example, if 10 pairwise Chi-square tests are done, the risk of one wrong answer approaches almost 40%. By doing logistic regression, the characteristics of the overall test can be evaluated first, followed by pairwise tests for each factor. Logistic regression takes into account the duration of medication uses as well, because all occasions upon which the patient has reported are taken into account[20].

3.1. Different organ infections In the following part of this chapter we discuss infection in different organs separately. Multinomial logistic regression was used to compare different severities of each organ infection with the control group (those who did not have this type of infection). The reason for using this model was because the outcome was a non-binary categorical variable. Using pair-wise Chi-square test without applying the model could potentially increase errors

135

because the number of pairwise tests would be far too high. After checking this test, we assess convergence status table for the model. It shows if the test meets the criteria for accuracy and the variables fit the statistical model (Appendices tables A1-A5, B1-B5, C1-C5, D1-D5, E1- E5, F1-F5, G1-G5, H1-H5, I1-I5, J1-J5, K1-K5).

3.2. Eye, Ears, Nose and Throat (EENT) infection - analysis of Anti-RA medicines Amongst 21,506 observations 1050/21506 (4.88%) reported at least mild EENT infection, 1829/21506 (8.5%) reported moderate infection, and 406/21506 (1.88%) reported severe infection, whereas 18221/21506 (84.72 %) reported no EENT infection at all. Overall the results show a significant difference [(chi square (χ2) of 431.3 with a p-value < 0.0001)] between the variable effects on the EENT infection (Appendix Tables A.7) [20].

The convergence status table for the model (Appendix Tables A.5-A.7) shows that the test meets the criteria for accuracy and the variables fit the statistical model. Overall the test shows a significant difference (likelihood ratio Chi-square of 431.2272 with a p-value of less than 0.0001) for variable effects in relation to EENT infection. This means that one or more of the medications under study are really associated with EENT infection [20] (Appendix table A.7).

The Wald Chi-square for the overall test is also highly significant (0.0001), with a Chi-square of almost 415 among 144 degrees of freedom. In other words, the impact on the EENT infection is not the same in different groups. This Chi-square p-value is almost equivalent to the p-value in the overall Pearson test. Indeed, the logistic regression result is much the same as the frequency table (Table 3.35) result because it is a large sample. As the model used is a logistic regression and not a linear regression model, Chi-square is used to test comparisons.

The Score test (Lagrange multiplier test) requires estimating only a single model. The test statistic is calculated based on the slope of the likelihood function at the observed values of the variables in the model. Usually the score tests are compared when we add parameters and it gives us an estimation of how far the accuracy of the test improves by adding new parameters or deleting existing parameters [21] (Appendix table A.7). During the backward stepwise model in the next part of the model, the effects of the medications are dropped one by one to see how much change occurs in the chi-square and to obtain an estimation of the amount of impact of that medication on increasing EENT infection [22] (Appendix table A8-A.31).

136

3.2.1. Wald Chi-square, Likelihood ratio test and Score test to test significance of differences As the size of the study population in this study is high enough, any of these three tests can be used, but if the size of the sample is small, then all three tests need to be used to increase the reliability of the conclusions. Among these tests, the likelihood ratio test is the most reliable test, because it stays unchanged even if the data is reparametrized [22] (Appendix table A.7).

3.2.2. Effects of medications on Eye Ear Nose and Throat (EENT) infection As the medication effects in this model are all qualitative, the degree of effect (impact) on the particular infection can be easily worked out by comparing these categorical variables a backward procedure in multinominal logistic regression was used. According to the summary table for the backward procedure (Table 4.2 and Appendix Table A.32), the least significant effect is from Azathioprine followed by Certolizumab, Penicillamine, IM Gold Injection, Rituximab, and Golimumab. However, the effect of all of these medications was minimal. Accordingly, they were dropped from the model (Table 4.2 and Appendix Table A.32).

Table 4.2 Summary of backward elimination of anti-RA medications and risk of EENT infection Summary of Backward Elimination Step Effect DF Number Wald Pr > ChiSq Variable Removed In Chi-Square Label 1.00 Azathioprine 9.00 17.00 4.99 0.84 Azathioprine 2.00 Certolizumab 9.00 16.00 5.45 0.79 Certolizumab 3.00 Penicillamine 9.00 15.00 7.20 0.62 Penicillamine 4.00 IM Gold injection 9.00 14.00 9.19 0.42 IM Gold injection 5.00 Rituximab 9.00 13.00 13.65 0.14 Rituximab 6.00 Golimumab 6.00 12.00 11.22 0.08 Golimumab

According to Table 4.3, the following medications have significant association with either producing or reducing the risk of EENT infection in RA. These medications include Etanercept, Adalimumab, Anakinra, Infliximab, Abatacept, Tocilizumab, F Methotrexate (plus Folic acid),

137

Hydroxychloroquine, Sulphasalazine, Leflunomide, Cyclosporine, and Prednisolone (Table 4.3 and Appendix Table A.33).

Table 4.3 Medications implicated in the promotion of EENT infection Type 3 Analysis of Effects Effect DF Chi-Square Pr > ChiSq Abatacept 9.00 18.02 0.04 Adalimumab 9.00 22.41 0.01 Anakinra 9.00 18.27 0.03 Cyclosporine 9.00 47.34 <.0001 Etanercept 9.00 52.14 <0.0001 Folic acid plus 3.00 9.42 0.02 Methotrexate Hydroxychloroquine 9.00 23.37 0.01 Infliximab 9.00 31.02 0.0003 Leflunomide 9.00 17.53 0.04 Prednisolone 9.00 29.48 0.0005 Sulphasalazine 9.00 26.74 0.0015 Tocilizumab 9.00 18.10 0.03

In Table 4.4, the effect of each medication was examined in turn:

Etanercept (ETA): The current use of Etanercept (Etanercept) seems to marginally increases the overall risk of mild EENT infection up to 18 times (CI: 0.989 to 1.43, P value 0.06), but somewhat paradoxically, amongst all biologics used, the use of Etanercept was associated with a significant reduction in the chance of severe EENT infection (Table 4.4-4.5 and Appendix Tables A.34- A.35).

Adalimumab (ADA): The current use of Adalimumab is associated with an increase (P Value: 0.0021) in the risk of mild and moderate EENT infection. The amount of this increase is in turn almost 33 times greater for mild (CI: 1.110 to 1.605, P value 0.0021) and 20 times greater for moderate (CI: 1.041 to 1.390, P value 0.0122) EENT infections (Table 4.4-4.5 and Appendix Tables A.34- A.35).

Infliximab (INX): The current use of Infliximab is associated with a higher risk of mild (P value of 0.0002) and moderate EENT infection (P value of 0.0007). For mild EENT infection,

138

the amount of this increase is up to almost 90 times (CI: 1.352 to 2.682) greater compared to participants who have never taken Infliximab, whereas for moderate EENT infection it was 60 times greater (CI: 1.220 to 2.109) (Table 4.4-4.5 and Appendix Tables A.34- A.35).

Abatacept (ABT): The current use of Abatacept is associated with an increase in the risk of mild EENT infection (P value: 0.0335). The amount of this increase is 40 times greater than in patients who have never taken this medication (CI: 1.027 to 1.908) (Table 4.4-4.5 and Appendix Tables A.34- A.35). The risk for moderate and severe EENT infections was not significant.

Tocilizumab (TOC): The current use of Tocilizumab is associated with an increase in the risk of mild (P value: 0.0036) and moderate (P value: 0.0143) EENT infection. The amount of this increase is almost 64 times greater in mild EENT infection (CI: 1.175 to 2.283) and 39 times greater (CI: 1.068 to 1.812) for moderate infection, compared to patients who have never taken this medication. (Table 4.4-4.5 and Appendix Tables A.34- A.35).

Methotrexate (plus Folic acid): The use of Methotrexate (plus Folic acid) increased the risk of infection more than 16 times, but among all csDMARDs users, this medication is associated with a reduction in the risk of moderate (P value: 0.0049, CI: 0.752 to 0.950) EENT infection (Table 4.4-4.5 and Appendix Tables A.34- A.35).

Hydroxychloroquine (HCQ) and Sulphasalazine (SAS): Modest increases in rates of EENT infection were observed for both agents, but these increases were not statistically significant. (Table 4.4-4.5 and Appendix Tables A.34- A.35).

Leflunomide (LEF): The current use of Leflunomide is associated with an increase in risk of mild EENT infection (P value: 0.017). The amount of this increase is 31 times greater than in patients who have never taken this medication (CI:1.065 to 1.613) (Table 4.4-4.5 and Appendix Tables A.34- A.35). Changes in the rates of moderate and severe EENT infection for LEF were not significant.

Cyclosporine A (CYA): The current use of cyclosporine A is associated with an increase in the risk of moderate (P value: 0.0001) and severe (P value: 0.0202) EENT infection. The amount of this increase is almost 180 times for moderate infection (CI: 1.833 to 4.403) and 170 times (CI: 1.173 to 6.577) for severe infections compared to patients who have never taken this medication (Table 4.4-4.5 and Appendix Tables A.34- A.35).

Prednisolone: The current use of prednisolone is associated with a significant increase in the risk of severe EENT infection (P value: 0.0329). The amount of this increase is almost 48 times

139

greater than that in patients who have never taken this medication (CI: 1.032 to 2.118) (Table 4.4-4.5 and Appendix Tables A.34- A.35).

In summary, multiple DMARDs were found to be associated significantly with mild or moderate EENT infections, whereas only cyclosporin and prednisolone were found to be associated with severe EENT infection.

Conclusion: The findings demonstrate differential risk for EENT infections for users of both csDMARDs and bDMARDs. Cyclosporine A and prednisolone confer high risk for example in comparison to HCQ, SAS and to a lesser extent MTX/FA and LEF. Amongst bDMARDs users, TNF inhibitors and Tocilizumab confer high risk compared to Abatacept. Whether the differences between TNF inhibitors are clinically important is doubtful.

Methotrexate (plus Folic acid) confer lower risk for EENT infection, whereas cyclosporine, prednisolone and infliximab are associated with the highest rates for EENT infections (Table 4.4-4.5 and Appendix Tables A.34- A.35).

140

Table 4.4 Analysis of maximum likelihood estimate in EENT infection Analysis of Maximum Likelihood Estimates Parameter Medication EENT DF Estimate Standard Wald Pr > ChiSq Status Infection Error Chi- Square Adalimumab currently taking Mild 1.00 0.29 0.09 9.42 0.0021

currently taking Moderate 1.00 0.19 0.07 6.28 0.0122 currently taking Severe 1.00 -0.08 0.15 0.29 0.5873 Cyclosporin currently taking Mild 1.00 0.53 0.33 2.46 0.1168 currently taking Moderate 1.00 1.04 0.22 21.80 <.0001 currently taking Severe 1.00 1.02 0.44 5.39edrt 0.0202 Etanercept Stopped taking Severe 1.00 -0.40 0.15 7.47 0.01 Don’t know mild 1.00 1.30 0.54 5.73 0.02 Don’t know Moderate 1.00 1.92 0.34 31.40 <.0001 currently taking mild 1.00 0.17 0.09 3.38 0.07 currently taking Severe 1.00 -0.34 0.14 5.47 0.02 Infliximab Stopped taking Moderate 1.00 -0.21 0.11 3.50 0.06 currently taking mild 1.00 0.64 0.17 13.59 0.0002 currently taking Moderate 1.00 0.47 0.14 11.46 0.0007 Folic acid currently taking Moderate 1.00 -0.17 0.06 7.92 0.0049 plus Methotrexate Cyclosporine Stopped taking Moderate 1.00 0.20 0.07 8.71 0.0032 Stopped taking Severe 1.00 0.47 0.13 12.38 0.0004 currently taking Moderate 1.00 1.04 0.22 21.80 <.0001 currently taking Severe 1.00 1.02 0.44 5.40 0.02

Arava currently taking Mild 1.00 0.27 0.10 6.50 0.01 (Leflunomide) currently taking Moderate 1.00 0.15 0.08 3.02 0.08 currently taking Severe 1.00 0.0064 0.17 0.0014 0.97

Prednisolone Stopped taking mild 1.00 0.33 0.11 9.34 0.0022 Stopped taking Moderate 1.00 0.26 0.08 9.27 0.0023 Stopped taking Severe 1.00 0.50 0.18 7.37 0.01 currently taking Severe 1.00 0.39 0.18 4.55 0.03

141

Table 4.5 Estimation of Odd’s ratios in EENT infection Odds Ratio Estimates Effect EENT Point 95% Wald Infection Estimate Confidence Limits Adalimumab- currently taking vs never Mild 1.335 1.110 1.605 taken Moderate 1.203 1.041 1.390 Severe 0.923 0.692 1.232 Etanercept - currently taking vs never Mild 1.19 0.99 1.43 taken Moderate 1.09 0.95 1.26 Severe 0.71 0.54 0.95 Infliximab currently taking vs never Mild 1.90 1.35 2.68 taken Moderate 1.60 1.22 2.11 Severe 0.69 0.33 1.43 Cyclosporine - -currently taking vs Mild 1.70 0.88 3.29 never taken Moderate 2.84 1.83 4.40 Severe 2.78 1.17 6.58 Prednisolone currently taking vs never Mild 1.18 0.96 1.46 taken Moderate 1.14 0.97 1.34 Severe 1.48 1.03 2.12

3.3. Chest or lung infection - analysis of anti-RA medicines Amongst 21506 observations, 371/21506 (1.72 %) were self-reported mild infections, 1379/21506 (6.41%) were self-reported moderate infections and 624/21506 (2.9%) were self- reported severe infections. In contrast, for 19132/21506 (88.96 %) patient visits, no infections were reported. In this model, categories of reported chest or lung infection were compared to participants who reported no chest or lung infection. A multinomial logistic regression model was used. The reason for using this model is because the outcome is a non-binary categorical variable. Using pairwise Chi-square test without using the model can increase potential mistakes because the number of comparisons is high (14). The model convergence status table (Appendix Tables B.5-B.7) is used to assess that the test meets the criteria for accuracy and the variables fit the statistical model (14) (Appendix table B.7).

In the model fit statistics table (Appendix table B.7), the likelihood ratio or lr (difference between -2 Log L or Deviance in the model which contains just the intercept and the one which contains both the intercept and covariates) is 383.2851. The P value is highly significant. This shows that a model with covariates is making the test more rigorous and that covariates are

142

actually impacting the cofactors in the lung infection. Other tests such as SC and AIC are also used to recheck this conclusion [22] (Appendix table B.6).

The Wald Chi-square test for overall test is also highly significant (0.0001) with a Chi-square of almost 397 among 162 degrees of freedom. In other words, the impact on lung infection is not the same in different groups. This Chi-square P value is almost equivalent to the p value in the overall Pearson test. Indeed, the logistic regression result is much the same as that for the frequency table (Table 3.41) result because it is a large sample. As the model used is logistic regression and not a linear regression, the Chi-square test permits comparison. [21] (Appendix table B.7).

During the backward stepwise model in the next part of the model, the effects of the medications are dropped one by one to see how much change occurs in the Chi-square and to get an estimation of the amount of impact of that medication on the frequency of lung infection[22] (Appendix table B8-B.31).

3.3.1. Wald Chi-square, likelihood ratio test and score test to test significance of differences As the size of the study population in this study was large enough, any of these three tests can be used. Had the size of the sample been small, then it would have been necessary to use all three tests to increase the reliability of the conclusions. Among these tests, the likelihood ratio test is the most reliable test, because it remains unchanged even if reparameterization is necessary [22] (Appendix table B.7).

3.3.2. Effects of different medications on lung infection

The effects of each variable on the lung infection was investigated directly by studying its coefficient. backward procedure in logistic regression, with three levels of mild, moderate and severe infections was applied because the outcome was categorical. The backward stepwise 143

procedure was used here because it is more accurate than the forward procedure and considers the accumulating effect of all variables and starts with a bigger model. The model-fit statistics show that, as using this large model is still fitted to the data, it can be used accordingly. Also, there is no collinearity and no two variables are identical. This makes it easier to use the backward model.

As the variables are all categorical variables, the effects of each variable on lung infection can be examined by studying its coefficient, directly [21].

For this section, a backward procedure in multinominal logistic regression was preferred. Logistic regression is required, because the outcome is categorical and as lung infection has three categories of severity, viz: mild, moderate and severe and a no infection category as well, a multinominal logistic regression is appropriate. Also, using backward stepwise is preferable here because it is more accurate than a forward procedure and considers the accumulating effect of all variables and starts with a bigger model. As the model fit statistics show that using this large model is still well fitted to the data, it can be used with confidence. Furthermore, there is no collinearity and no two variables are identical. This makes it easier to use the backward model [22].

According to the summary table of results derived from use of the backward procedure, the least significant effect is from Certolizumab followed by Penicillamine, Methotrexate (plus Folic acid), Azathioprine, Rituximab, Infliximab, Tocilizumab, Golimumab, Arava (Leflunomide), and Adalimumab. However, the effect of all these medications was found to be minimal and so they were eliminated from the model (Table 4.6 and Appendix Table B.32).

144

Table 4.6 Summary of backward elimination of anti-RA medications and risk of lung infection

Summary of Backward Elimination

Effect Number Wald Step Removed DF In Chi-Square Pr > ChiSq

1 Certolizumab 9 18 5.2822 0.8090

2 Penicillamine 9 17 8.9767 0.4394

3 Methotrexate and Folic acid 3 16 3.0046 0.3909

4 Azathioprine 9 15 10.4127 0.3181

5 Rituximab 9 14 10.6214 0.3026

6 Infliximab 9 13 12.4421 0.1895

7 Tocilizumab 9 12 14.5987 0.1026

8 Golimumab 6 11 11.7349 0.0682

9 Arava (Leflunomide) 9 10 16.1280 0.0643

10 Adalimumab 9 9 16.1807 0.0632

According to the type 3 analysis of effects, the following medications have significant association with increasing or reducing the propensity for lung infection in RA. These medications include Etanercept, Anakinra, Abatacept, Hydroxychloroquine, Hydroxychloroquine, Sulphasalazine, Cyclosporine, Prednisolone, and IM Gold injections (Table 4.7 and Appendix Table B.33).

145

Table 4.7 Anti-rheumatic medications and propensity for lung infection

Type 3 Analysis of Effects

Wald Effect DF Chi-Square Pr > ChiSq

Etanercept 9 31.4874 0.0002

Anakinra 9 20.0990 0.0173

Abatacept 9 34.9246 <.0001

Methotrexate 9 20.5746 0.0147

Hydroxychloroq 9 24.4648 0.0036 uine

Sulphasalazine 9 20.8255 0.0134

Cyclosporine 9 20.6307 0.0144

Prednisolone 9 67.5034 <.0001

IM Gold 9 19.8810 0.0187

In the analysis of maximum likelihood, the statistical findings in respect to each medication examined are shown in detail, in this section we just discuss significant effect of medications which are currently being taken by the patient:

Abatacept: Currently taking ABT was found to be associated with a significant increase in the propensity for moderate lung infection (P value: <.0001). The size of this increase equates to almost 70 times more in the case of moderate infection compared to patients who don’t take Abatacept at all (CI: 1.36 to 2.161) (Table 4.8-4.59 and Appendix Tables B.34- B.35).

Hydroxychloroquine: Currently taking HCQ is strongly associated with an increase in the rate of severe infection in RA (P value: 0.0001). Taking this medication is associated with 53 times greater risk for severe infection (CI: 1.23 to 1.91). The effect of taking Hydroxychloroquine in increasing moderate level of infection is marginal (P value: 0.049) and can reach to 17 times more risk (Table 4.8-4.59 and Appendix Tables B.34- B.35).

146

Sulphasalazine: Currently taking SAL is marginally associated with an increase in the rate of severe infection in RA (P value: 0.0689). However, the amount of this increase is ignorable (Table 4.8-4.59 and Appendix Tables B.34- B.35).

Cyclosporine A: Current use of CYA was found to be associated with an increased propensity for mild (P value 0.0012) and moderate (P value 0.0064) lung infection. Taking this medication is associated with 243 times greater risk for mild infection and 105 times greater risk for moderate infection (Table 4.8-4.59 and Appendix Tables B.34- B.35).

Prednisolone: Currently taking prednisolone was found to be associated with an increased propensity for all categories of lung infection. Taking prednisolone was associated with a 63 times greater propensity for mild infection and a 33- and 140-times greater propensity for moderate and severe lung infection,, respectively (Table 4.8-4.59 and Appendix Tables B.34- B.35).

Methotrexate: Taking methotrexate was also associated with significant increase in the rate of moderate infection (p‐value < 0. 0.0166). Taking this medication is associated with 4 times greater risk for moderate infection (CI: 1.313 to 15.218).

IM Gold Injection: No impact of IM Gold injections on lung infection was identified (Table 4.8-4.59 and Appendix Tables B.34- B.35).

In summary, multiple DMARDs were found to be associated with mild or moderate lung infection, whereas only use of Prednisolone was found to be associated with severe lung infection.

147

Table 4.8 Analysis of maximum likelihood estimate in lung infection

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfLung DF Estimate Error Chi-Square Pr > ChiSq Etanercept currently taking Mild 1 -0.1632 0.1350 1.4614 0.2267 Etanercept currently taking Moderate 1 -0.0438 0.0677 0.4200 0.5169 Etanercept currently taking Severe 1 -0.1177 0.1018 1.3382 0.2473 Anakinra currently taking Mild 1 1.1898 1.0395 1.3102 0.2524 Anakinra currently taking Moderate 1 1.0875 0.6364 2.9208 0.0874 Anakinra currently taking Severe 1 -11.1525 348.5 0.0010 0.9745 Abatacept currently taking Mild 1 0.3420 0.2178 2.4657 0.1164 Abatacept currently taking Moderate 1 0.5392 0.1180 20.8696 <.0001 Abatacept currently taking Severe 1 -0.0929 0.2079 0.1996 0.6550 Hydroxychloroq currently taking Mild 1 0.2042 0.1458 1.9617 0.1613 uine Hydroxychloroq currently taking Moderate 1 0.1582 0.0804 3.8734 0.0491 uine Hydroxychloroq currently taking Severe 1 0.4302 0.1114 14.9202 0.0001 uine Sulphasalazine currently taking Mild 1 0.3020 0.1660 3.3084 0.0689 Sulphasalazine currently taking Moderate 1 0.0113 0.1022 0.0122 0.9120 Sulphasalazine currently taking Severe 1 0.00650 0.1447 0.0020 0.9641 Cyclosporine currently taking Mild 1 1.2314 0.3793 10.5374 0.0012 Cyclosporine currently taking Moderate 1 0.7209 0.2642 7.4466 0.0064 Cyclosporine currently taking Severe 1 0.2533 0.4633 0.2990 0.5845 Prednisolone currently taking Mild 1 0.4916 0.1786 7.5773 0.0059 Prednisolone currently taking Moderate 1 0.3192 0.0919 12.0656 0.0005 Prednisolone currently taking Severe 1 0.8943 0.1574 32.2591 <.0001 IM Gold currently taking Mild 1 -1.3342 1.0067 1.7565 0.1851 IM Gold currently taking Moderate 1 0.2764 0.2666 1.0746 0.2999 IM Gold currently taking Severe 1 -0.1061 0.4610 0.0529 0.8181

Conclusion: Differential effects on the propensity to lung infections were observed with csDMARDs and bDMARDs. Prednisolone, CYA and HCQ all increased this propensity, whereas SAS and IM Gold did not or there was insufficient data to draw firm conclusions. Amongst bDMARDs, ABT was significantly associated with an increased frequency of moderate lung infections, whereas ETA and Anakinra were somewhat surprisingly associated with possible reduced rates of lung infection. Amongst serious infections in RA, lung infections are the most common. In patients with chronic lung diseases, such as COPD, bronchiectasis and in those with a past history of one or more attacks of pneumonia, for example, this new data could be factored into treatment selection.

148

Table 4.9- Estimation of odds ratios in lung infection

Odds Ratio Estimates Point 95% Wald Effect InfLung Estimate Confidence Limits Etanercept currently taking vs never taking 1 0.849 0.652 1.107 Etanercept currently taking vs never taking 2 0.957 0.838 1.093 Etanercept currently taking vs never taking 3 0.889 0.728 1.085 Anakinra currently taking vs never taking 1 3.286 0.428 25.207 Anakinra currently taking vs never taking 2 2.967 0.852 10.327 Anakinra currently taking vs never taking 3 <0.001 <0.001 >999.99 9 Abatacept currently taking vs never taking 1 1.408 0.919 2.157 Abatacept currently taking vs never taking 2 1.715 1.361 2.161 Abatacept currently taking vs never taking 3 0.911 0.606 1.370 Hydroxychloroquine currently taking vs never 1 1.227 0.922 1.632 taking Hydroxychloroquine currently taking vs never 2 1.171 1.001 1.371 taking Hydroxychloroquine currently taking vs never 3 1.538 1.236 1.913 taking Sulphasalazine currently taking vs never taking 1 1.353 0.977 1.873 Sulphasalazine currently taking vs never taking 2 1.011 0.828 1.236 Sulphasalazine currently taking vs never taking 3 1.007 0.758 1.337 Cyclosporine currently taking vs never taking 1 3.426 1.629 7.206 Cyclosporine currently taking vs never taking 2 2.056 1.225 3.451 Cyclosporine currently taking vs never taking 3 1.288 0.520 3.194 Prednisolone currently taking vs never taking 1 1.635 1.152 2.320 Prednisolone currently taking vs never taking 2 1.376 1.149 1.648 Prednisolone currently taking vs never taking 3 2.446 1.796 3.330 IM Gold currently taking vs never taking 1 0.263 0.037 1.894 IM Gold currently taking vs never taking 2 1.318 0.782 2.223 IM Gold currently taking vs never taking 3 0.899 0.364 2.220

149

3.4. Skin and Nail infection - analysis of Anti-RA medicines Amongst 21506 patient-visit observations 1253/21506 (5.82 %) self-reported mild infection, 1039/21506 (4.83%) self-reported moderate infection and 361/21506 (1.67%) self-reported severe infection. In contrast, for 18853/21506 (87.66 %) patient-visits, no infections were reported. In this model, participants who developed different severities of skin and nail infection were compared to participants who did not develop this type of infection (Appendix tables C1- C3). A multinomial logistic regression model was used to evaluate these reports. The reason for using this model is because the outcome is a non-binary categorical variable. Using pairwise Chi-square test without using the model can increase potential mistakes because the number of comparisons is large[20].

The model convergence status table (Appendix Tables C.5-C.7) shows that the test meets the criteria for accuracy and the variables fit the statistical model. Overall, the test shows that anti- RA medicines have an effect on the risk of skin and nail infection (lr Chi-square of 386.3201 with a P value of less than 0.0001) [20] (Appendix table C.7).

In the model fit statistics table (Appendix table C.7), the likelihood ratio or lr (difference between -2 Log L or Deviance in the model which just contains intercept and the one which contains intercept and covariates) is 386.3201. The P value is highly significant (Appendix table C.7). This shows that a model with covariates is making the test more robust and that covariates are actually impacting cofactors in skin and nail infection. Other tests, such as SC and AIC, are also used to recheck this conclusion [22] (Appendix table C.6).

Wald Chi-square for overall test is also highly significant (0.0015), with a Chi-square of almost 85 among 50 degrees of freedom. In other words, the impact on the skin infection is not the same in different groups. This Chi-square is almost equivalent to the p-value in the overall Pearson test. Indeed, the logistic regression result is much the same as the frequency table (Tables 3.29-3.30) result because it is a large sample. As the model used was a logistic regression and not a linear regression model, the Chi-square test was used for comparison. [21] (Appendix table C.7).

150

During the backward stepwise procedure in the next part of the model, the effects of the medications are eliminated, one by one, to see how much change occurs in the Chi-square and to get an estimation of the amount of impact of that medication in increasing skin and nail infections [22] (Appendix tables C8-C.31).

3.4.1. Effects of different medications on skin and nail infection

For this section, a backward elimination procedure was preferred, utilising multinominal logistic regression. Logistic regression is required because the outcome is categorical and because nail and skin infection have three categories, viz: mild, moderate and severe. Furthermore, there is a no infection category. Accordingly, multinominal logistic regression is a more appropriate model (Appendix Table C.4). A backward stepwise procedure is used here because it is more accurate than a forward procedure and considers the accumulating effect of all variables and also because it starts with a bigger model. As the model fit statistics (Appendix Table C.6) show, using this large model is still well fitted to the data, so it can be used appropriately. Moreover, since there is no collinearity and no two variables are identical, it is easier to use the backward model [22].

According to the summary of the backward procedure, as shown in Table 4.10,, the least significant effect is from Certolizumab, followed by Hydroxychloroquine, IM Gold injection, Abatacept, Tocilizumab, Penicillamine, Golimumab, Anakinra, and Azathioprine. However, the effect of all these medications was minimal and so they were eliminated from the model (Table 4.10 and Appendix Table C.32).

151

Table 4.10 Summary of backward elimination of anti-RA medications and risk of skin and nail infection. Summary of Backward Elimination Step Effect DF Number Wald Pr > ChiSq Removed In Chi-Square 1.00 Certolizumab 9.00 17.00 7.14 0.62 2.00 Hydroxychloroquine 9.00 16.00 7.26 0.61 3.00 IM Gold injection 9.00 15.00 9.70 0.38 4.00 Abatacept 9.00 14.00 11.59 0.24 5.00 Tocilizumab 9.00 13.00 10.33 0.32 6.00 Penicillamine 9.00 12.00 12.72 0.18 7.00 Golimumab 6.00 11.00 9.98 0.13 8.00 Anakinra 9.00 10.00 14.79 0.10 9.00 Azathioprine 9.00 9.00 15.79 0.07 10.00 Cyclosporine 9.00 8.00 15.94 0.07

According to the type 3 analysis of effects, as shown in Table 4.11, the following medications have significant impact on producing or reducing the risk of skin and nail infection in RA. These medications include: Etanercept, Adalimumab, Infliximab, Rituximab, Methotrexate (plus Folic acid), Sulphasalazine, Leflunomide, Prednisolone (Table 4.11 and Appendix Table C.33).

Table 4.11 Medications associated with a propensity to increase skin and nail infections Type 3 Analysis of Effects Effect DF Wald Chi-Square Pr > ChiSq Etanercept 9.00 24.94 0.00 Adalimumab 9.00 21.42 0.01 Infliximab 9.00 29.90 0.00 Rituximab 9.00 24.22 0.00 Methotrexate (plus Folic acid) 3.00 25.61 <.0001 Sulphasalazine 9.00 34.68 <.0001 Leflunomide 9.00 26.58 0.00 Prednisolone 9.00 38.01 <.0001

152

In the analysis of maximum likelihood, the effect of each medication was examined in more detail:

Etanercept: The use of ETA was associated with a reduced frequency of skin and nail infections. Moderate infections were observed to occur more than 19-fold less often (CI: 0.679 to 0.971) and severe infections 30 times less often (CI: 0.522 to 0.956). Among all biologics taken Etanercept was associated with a reduction in moderate and severe skin and nail infection (Table 4.12-4.13 and Appendix Tables C.34- C.35).

Adalimumab: Currently taking Adalimumab was associated with a slight increase in the frequency of mild infection (P Value: 0.0076). The amount of this increase is almost 24 times greater than that for patients who have never taken Adalimumab (CI: 1.061 to 1.468) (Table 4.12-4.13 and Appendix Tables C.34- C.35).

Infliximab: Currently taking Infliximab was associated with an increased frequency for severe skin and nail infection. (P value of 0.0404). The amount of this increase is up to almost 72 times more than that for participants who have never taken Infliximab. (CI: 1.024 to 2.911) (Table 4.12-4.13 and Appendix Tables C.34- C.35).

Rituximab: Current use of Rituximab was associated with an increased frequency of skin and nail infection with increases more than 34 to 40 times in moderate and mild infection, but among all biologics, the use of Rituximab was associated with a reduced risk in mild (p value 0.0032) and moderate (p value 0.0149) skin and nail infection. (Table 4.12-4.13 and Appendix Tables C.34- C.35).

Methotrexate (plus Folic acid): Currently taking Methotrexate (plus Folic acid) can reduce the frequency of both mild and moderate skin and nail infection. The reduction in mild infection is up to almost 26 times lower than in those who have never taken Methotrexate (plus Folic acid). In moderate skin and nail infection, it is approximately 20 times lower (Table 4.12-4.13 and Appendix Tables C.34- C.35).

Sulphasalazine: Currently taking SAS reduces the frequency of mild skin and nail infection. According to points estimate, the frequency of skin and nail infection is 29 times (1/0.710) less than that for patients who have never taken this agent (CI:0.566 to 0.891) (Table 4.12-4.13 and Appendix Tables C.34- C.35).

Leflunomide: Currently taking LEF increases the risk of mild and moderate skin and nail infection. According to points estimate the frequency of skin and nail infection is 35 times

153

greater for mild (CI:1.131 to 1.629) and 25 times greater for moderate infection (CI:1.015 to 1.528) (Table 4.12-4.13 and Appendix Tables C.34- C.35).

Prednisolone: Currently taking Prednisolone increases the frequency of severe infection. According to points estimate, this infection rate is more than 160 times more than that for patients who have never taken prednisolone (CI: 1.688 to 4.055) (Table 4.12-4.13 and Appendix Tables C.34- C.35).

In summary, they use of multiple biologic and conventional synthetic DMARDs was found to be associated with increased rates of mild and moderate skin and nail infections, whereas prednisolone use and infliximab were associated with an increased frequency of severe skin and nail infections.

154

Table 4.12 Analysis of maximum likelihood estimate in skin and nail infection Analysis of Maximum Likelihood Estimates Parameter Medication Skin and DF Estimate Standard Wald Pr > ChiSq Status Nail Error Chi- Infection Square Etanercept currently Mild 1.00 0.001 0.08 0.001 0.99 taking Moderate 1.00 -0.21 0.09 5.22 0.02 Severe 1.00 -0.35 0.15 5.06 0.02 Adalimumab currently Mild 1.00 0.22 0.08 7.13 0.01 taking Moderate 1.00 0.01 0.09 0.01 0.93 Severe 1.00 -0.08 0.16 0.29 0.59 Infliximab currently Mild 1.00 0.22 0.18 1.59 0.21 taking Moderate 1.00 -0.11 0.21 0.29 0.59 Severe 1.00 0.55 0.27 4.20 0.04 Rituximab currently Mild 1.00 -0.50 0.17 8.70 0.001 taking Moderate 1.00 -0.41 0.17 5.93 0.01 Severe 1.00 -0.45 0.26 3.02 0.08 Methotrexate currently Mild 1.00 -0.30 0.07 17.08 <.0001 (plus Folic taking Moderate 1.00 -0.18 0.08 5.37 0.02 acid) Severe 1.00 0.21 0.12 3.27 0.07

Sulphasalazine currently Mild 1.00 -0.34 0.12 8.77 0.001 taking Moderate 1.00 -0.11 0.12 0.86 0.35 Severe 1.00 0.09 0.19 0.21 0.65 Leflunomide currently Mild 1.00 0.31 0.09 10.74 0.001 taking Moderate 1.00 0.22 0.10 4.40 0.04 Severe 1.00 0.18 0.17 1.14 0.29 Prednisolone currently Mild 1.00 0.10 0.09 1.31 0.25 taking Moderate 1.00 0.01 0.10 0.02 0.90 Severe 1.00 0.96 0.22 18.49 <.0001

155

Conclusion: Differential effects on the frequency of skin and nail infections were observed amongst users of csDMARDs and bDMARDs. Prednisolone substantially increased the frequency of severe skin and nail infections, whereas less consistent effects were observed with other non-biologic agents. Amongst bDMARDs, severe skin and nail infections were higher in recipients of Infliximab, whereas lower skin and nail infection rates were observed with Etanercept. Taking Leflunomide can also increase the risk of mild and moderate skin and nail infection.

Table 4.13 Estimation of odds ratios in skin and nail infection Odds Ratio Estimates Effect Skin and Nail Point 95% Wald Infection Estimate Confidence Limits Etanercept - currently taking vs Mild 1.00 0.85 1.18 never taken Moderate 0.81 0.68 0.97 Severe 0.71 0.52 0.96 Adalimumab - currently taking vs Mild 1.25 1.06 1.47 never taken Moderate 1.01 0.84 1.21 Severe 0.92 0.68 1.25 Infliximab - currently taking vs Mild 1.25 0.89 1.76 never taken Moderate 0.89 0.59 1.34 Severe 1.73 1.02 2.91 Rituximab - currently taking vs Mild 0.61 0.44 0.85 never taken Moderate 0.67 0.48 0.92 Severe 0.64 0.38 1.06 Methotrexate/ Methotrexate Mild 0.74 0.64 0.85 (plus Folic acid) - currently taking Moderate 0.84 0.72 0.97 vs never taken Severe 1.24 0.98 1.56

Sulphasalazine - currently taking Mild 0.71 0.57 0.89 vs never taken Moderate 0.90 0.71 1.13 Severe 1.09 0.75 1.60 Leflunomide - currently taking vs Mild 1.36 1.13 1.63 never taken Moderate 1.25 1.02 1.53 Severe 1.20 0.86 1.67 Prednisolone - currently taking vs Mild 1.11 0.93 1.33 never taken Moderate 1.01 0.83 1.23 Severe 2.62 1.69 4.06

3.5. Artificial (Prosthetic) Joint infection - analysis of Anti-RA medicines Amongst 21506 respondent observations, 19/21506 (0.088 %) reported mild prosthetic joint infection, 39/21506 (0.18%) self-reported moderate prosthetic joint infection, and 78/21506 (0.36%) reported severe prosthetic joint infection. For 21370/21506 (99 %) patient-visits, no infections were reported.

156

In this model, the different categories of prosthesis infection were compared to the control group (RA respondents who did not have prosthetic joint infection). A multinomial logistic regression model was used. The reason for using this model was because the outcome was a non-binary categorical variable. Using pairwise Chi-square test without using the model could increase potential error, therefore we use regression model.

The model convergence status table shows that the test meets the criteria for accuracy and the variables fit the model of statistics. Overall, the test shows that medications have significant different impacts on causing artificial joint infection in RA (lr Chi-square of 208.9481with a P value of 0.0018) .

The model convergence status table (Appendix Tables D.5-D.7) shows that the test meets the criteria for accuracy and the variables fit the statistical model. Overall, the test shows significant differential effects of anti-RA medications on prosthetic joint infection that (lr Chi-square of 208.9481with a P value of 0.0018) on the Artificial joint infection [20] (Appendix table D.7). In the model fit statistics (Table 4.14), the likelihood ratio or lr is 208.411 and the P value is significant. This shows that a model with covariates is making the test more robust and that covariates are actual impacting cofactors in respect to prosthetic joint infection. Other tests, such as SC and AIC, are also used to recheck this conclusion [22] (Table 4.14 and Appendix table D.6).

Table 4.14 Estimation of the impact of confounders Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 1,913.34 2,010.39 SC 1,937.27 3,254.66 -2 Log L 1,907.34 1,698.39

The Wald Chi-square for overall test is also highly significant (0.0001) with a Chi-square of almost 200.8923 among 153 degrees of freedom. In the other words, the impact on the Artificial Joint infection is not the same in different groups. This Chi-square P value is almost equivalent to the p value in the overall Pearson test. Indeed, the logistic regression result is much same as

157

the frequency table (Table 3.53-3.54) result because it is a large sample. As our model is logistic regression and not a linear regression, we are using Chi-square test for our comparison. [21] (Appendix D.7).

The Score test (Lagrange multiplier test) requires estimating only a single model. The test statistic is calculated based on the slope of the likelihood function at the observed values of the variables in the model. Usually we compare the score tests when we add parameters and it gives us an estimation of how far the accuracy of test improves by adding new parameters or deleting existing parameters. If we compare this test in backward regression, the major drop is happening in Cyclosporine, Azathioprine, Tocilizumab and Prednisolone [21] (Appendix D.7).

During the backward stepwise procedure in the next part of the model, the effects of the medications are dropped, one by one, to see how much change happen in the Chi-square and to get an estimation of the association between the medication and changes in the artificial joint infection [22] (Appendix D.8-D.31).

3.5.1. Wald Chi-square, Likelihood ratio test and Score test to test significance of differences As the size of the study population in our study is enough, we can use any of these three tests, but if the size of the sample is small then we have to check all these three tests to increase reliability of the conclusions. Among these tests, likelihood ratio test is the most reliable test, because it stays unchanged even if we reparametrize what we are testing [22] (Appendix D.7).

3.5.2. Effects of different medications on artificial (prosthetic) joint infection As the medication effects in this model are all qualitative, it is possible to work out the degree of effect (impact) on prosthetic joint infection by comparing these categorical variables.

For this section, a backward procedure in multinominal logistic regression was preferred (Table 4.15-4.16). Logistic regression was required because the outcome is categorical and as artificial joint infection has three categories of severity, notably mild, moderate and severe and a no infection report, multinominal logistic regression is the appropriate model. Also, backward stepwise is used here because it is more accurate than the forward procedure and considers the accumulating effect of all variables and starts with a bigger model (Appendix Table D.4). As the model fit statistics (Appendix Table D.6) show that using this large model is still fitted to

158

the data, it can be used appropriately. Also, there is not any collinearity and none of any two variables are identical. This makes it easier to use the backward model [22].

Table 4.15 Estimation of fitness of tests in artificial joint infection Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 208.95 153.00 0.001 Score 312.91 153.00 <.0001 Wald 200.89 153.00 0.01

According to the summary of the backward procedure (Table 4.16 and Appendix Table D.32), the least significant effect is from Certolizumab followed by Golimumab, Anakinra, Abatacept, Methotrexate (plus Folic acid), Azathioprine, Infliximab, Sulphasalazine, Prednisolone, Etanercept, Leflunomide, Penicillamine, Hydroxychloroquine, Tocilizumab, and Cyclosporine. However, the effect of all these medications was found to be minimal and they were dropped from the model.

159

Table 4.16 Summary of backward elimination of anti-RA medications and risk of artificial joint infection Summary of backward elimination Step Effect DF Number Wald Pr > ChiSq Removed In Chi-Square 1.00 Certolizumab 9.00 17.00 0.001 1.00 2.00 Golimumab 6.00 16.00 0.02 1.00 3.00 Anakinra 9.00 15.00 1.66 1.00 4.00 Abatacept 9.00 14.00 2.96 0.97 5.00 Folic acid plus 3.00 13.00 0.90 0.82 Methotrexate 6.00 Azathioprine 9.00 12.00 5.69 0.77 7.00 Infliximab 9.00 11.00 4.45 0.88 8.00 Sulphasalazine 9.00 10.00 5.63 0.78 9.00 Prednisolone 9.00 9.00 7.02 0.64 10.00 Etanercept 9.00 8.00 8.48 0.49 11.00 Leflunomide 9.00 7.00 7.73 0.56 12.00 Penicillamine 9.00 6.00 9.13 0.43 13.00 Hydroxychloroquine 9.00 5.00 10.32 0.33 14.00 Tocilizumab 9.00 4.00 12.22 0.20 15.00 Cyclosporine 9.00 3.00 16.02 0.07

According to type 3 analysis of effects (Table 4.17), the following medications had significant association with either increasing or reducing the risk of artificial joint infection in RA. These medications were: Adalimumab, Rituximab, and IM Gold injection (Table 4.17 and Appendix Table D.32).

Table 4.17 Medications implicated in the development of artificial (prosthetic) joint infection Type 3 Analysis of Effects Effect DF Wald Pr > ChiSq Chi-Square Adalimumab 9.00 17.24 0.05 Rituximab 9.00 17.54 0.04 IM Gold injection 9.00 30.83 0.001

160

In the analysis of maximum likelihood, the effect of each medication was investigated in more detail:

Adalimumab: Taking Adalimumab is associated with reduction in the risk of moderate artificial joint infection (P value 0.035). However, if patient does not need to take adalimumab, he / she will have a lower risk for infection (by up to 70 times) (CI: 0.083 to 0.916). (Table 4.18-4.19 and Appendix Tables D.34- D.35).

Rituximab: Currently taking Rituximab is not associated with either a reduction or increase in prosthetic joint infection (Table 4.18-4.19 and Appendix Tables D.34- D.35).

IM Gold injection: Currently taking parenteral Gold is not associated with either an increase or decrease in prosthetic joint infection (Table 4.18-4.19 and Appendix Tables D.34- D.35).

Table 4.18- Analysis of Maximum likelihood estimate in artificial joint infection Analysis of Maximum Likelihood Estimates Parameter Medication Artificial DF Estimate Standard Wald Pr > ChiSq Status Joint Error Chi- Infection Square Adalimumab currently Mild 1.00 -0.38 0.66 0.34 0.56 taking Moderate 1.00 -1.29 0.61 4.42 0.04 Severe 1.00 -0.07 0.28 0.06 0.81 Rituximab currently Mild 1.00 0.54 0.81 0.44 0.51 taking Moderate 1.00 -0.97 1.03 0.88 0.35 Severe 1.00 -0.46 0.74 0.38 0.54 IM Gold currently Mild 1.00 -11.94 1,057.40 0.001 0.99 injection taking Moderate 1.00 -12.03 728.10 0.003 0.99 Severe 1.00 -11.57 520.60 0.002 0.98

Conclusion: According to the above information and analysis, Adalimumab has a significant reduction impact in comparison with other bDMARDs. In people with risk of artificial joint infection, Adalimumab is the safest (Table 4.18-4.19 and Appendix Tables D.1- D.35).

161

Table 4.19- Estimation of Odds ratios in Artificial Joint infection Odds Ratio Estimates Effect Artificial Point 95% Wald joint Estimate Confidence Limits infection Adalimumab currently taking Versus never Mild 0.682 0.187 2.488 taken Adalimumab currently taking Versus never Moderate 0.276 0.083 0.916 taken Adalimumab currently taking Versus never Severe 0.937 0.545 1.612 taken Rituximab currently taking Versus never taken Mild 1.710 0.352 8.307 Rituximab currently taking Versus never taken Moderate 0.378 0.050 2.870 Rituximab currently taking Versus never taken Severe 0.633 0.149 2.690 IM Gold injection currently taking Versus Mild <0.001 <0.001 >999.999 never taken IM Gold injection currently taking Versus Moderate <0.001 <0.001 >999.999 never taken IM Gold injection currently taking Versus Severe <0.001 <0.001 >999.999 never taken

3.6. Bone, joint and muscle (BJM) infection - analysis of anti-RA medicines Amongst 21506 observations, 82/21506 (0.38 %) self-reported mild infection, 213/21506 (0.99%) self-reported moderate infection, 243/21506 (1.12%) self-reported severe infection, and for 20968/21506 (97.49 %) patient-visits, no infections were reported. In this model, ARAD participants who self-reported bone, joint and muscle infection of differing severity were compared with ARAD participants in whom there was no such infection. The statistical model used is the multinomial logistic regression model (Appendix Tables D1-D4). The reason for using this model is because the outcome is a non-binary categorical variable. Using pairwise Chi-square test without using the model can increase potential mistakes because the number of comparisons is large[20]. The model convergence status table (Appendix Tables D.5-D.7) shows that the test meets the criteria for accuracy and the variables fit the model of statistics. Overall, the test shows significant differences (lr Chi-square of 283.1804 with a P value of <.0001) between variable effects on bone, joint, and muscle infections (Appendix table D.7).

In the model fit statistics table (Appendix table D.7), the likelihood ratio or lr (difference between -2 Log L or Deviance in the model which just contains the intercept and one which contains both the intercept and covariates) is 283.1804. The P value is highly significant (Appendix table D.1-D.7). This shows that a model with covariates is strengthening the test and demonstrates that covariates are the actual impacting cofactors in BJM bone, joint and muscle 162

infection. Other tests such as SC and AIC are also used to recheck this conclusion (Appendix table D.6). Wald Chi-square for overall test is also highly significant (0.0001) with a Chi-square of almost 274 among 153 degrees of freedom. In other words, the impact on BJM bone, joint and muscle infection is not the same in different groups. This Chi-square P value is almost equivalent to the p value in the overall Pearson test. Indeed, the logistic regression result is much the same as in the frequency table (Tables 3.49-3.50) result, because it is a large sample. The model used is a logistic regression model and not a linear regression, Chi-square test was used for comparison. [21] (Appendix D.7).

The Score test (Lagrange multiplier test) requires estimating only a single model. The test statistic is calculated based on the slope of the likelihood function at the observed values of the variables in the model. Usually the score tests are compared when parameters are added, which provides an estimation of how far the accuracy of the test improves by adding new parameters or deleting existing parameters [21] (Appendix D.7). During the backward stepwise procedure in the next part of the model, the effects of the medications are dropped, one by one, to see how much there is a change in the Chi-square and to get an estimation of the amount of impact of that medication in increasing bone, joint and muscle (BJM) infection [22] (Appendix D8- D.31).

3.6.1. Wald Chi-squared, Likelihood ratio test and Score test to test significance of differences As the size of the study population in this study is large enough, any of these three tests can be used, but if the size of the sample were small, then it would be necessary to check all three tests to increase the reliability of the conclusions. Among these tests, the likelihood ratio test is the most reliable test, because it stays unchanged even if the data under analysis is reparametrized [22] (Appendix D.7).

3.6.2. Effects of different medications on bone, joint and muscle infection As the medication effects in this model are all qualitative, the degree of effect (impact) on an infection can be easily determined by comparing these categorical variables [21]. For this section, a backward procedure in multinominal logistic regression was preferred. Logistic regression was used because the outcome is categorical. As bone, joint and muscle infection has three categories of severity, notably: mild, moderate and severe together with a no infection category, multinominal logistic regression is an appropriate model to use (Appendix Table 163

D.4). Also, backward stepwise is used here because it is more accurate than is a forward procedure. It also considers the accumulating effect of all variables and starts with a bigger model. As the model fit statistics (Appendix Table D.6) show that using this large model is still fitted to the data, it can be appropriately used. Also, there is no collinearity and no two variables are identical. This makes it easier to use the backward model [22].

According to the summary table in the backward procedure, the least significant effect is from Certolizumab followed by Azathioprine, Anakinra, Golimumab, Tocilizumab, Cyclosporine, Methotrexate, Rituximab, Abatacept, Sulphasalazine, Etanercept, Adalimumab and IM Gold injection. However, the effect of all these medications was found to be minimal, so they were dropped from the model (Table 4.20 and Appendix Table D.32).

164

Table 4.20- Summary of backward elimination of anti-RA medications and risk of Bone, Joint and Muscle infections

Summary of Backward Elimination

Effect Number Wald Variable Step Removed DF In Chi-Square Pr > ChiSq Label

1 Certolizumab 9 18 1.6031 0.9963 Certolizumab

2 Azathioprine 9 17 3.8457 0.9213 Azathioprine

3 Anakinra 9 16 4.3211 0.8890

4 Golimumab 6 15 3.0321 0.8048 Golimumab

5 Tocilizumab 9 14 5.4188 0.7964 Tocilizumab

6 Cyclosporin 9 13 6.0931 0.7306 Cyclosporin

7 Methotrexate 9 12 8.0079 0.5334 Methotrexate

8 Rituximab 9 11 10.8068 0.2892 Rituximab

9 Abatacept 9 10 11.4004 0.2493 Abatacept

10 Sulphasalazine 9 9 14.0763 0.1196 Sulphasalazin e

11 Etanercept 9 8 15.3543 0.0817

12 Adalimumab 9 7 15.0322 0.0901

13 IM Gold 9 6 16.1101 0.0646 IM Gold

According to type 3 analysis of effects table, the following medications have significant impact on either increasing or reducing the risk of bone, joint and muscle (BJM) infection in RA. These medications include: Infliximab, Methotrexate (plus Folic acid), Hydroxychloroquine, Leflunomide, Prednisolone and Penicillamine (Table 4.21 and Appendix Table D.33).

165

Table 4.21 Effect of medications in causing bone, joint and muscle infection

Type 3 Analysis of Effects

Wald Effect DF Chi-Square Pr > ChiSq

Infliximab 9 24.8305 0.0032

Methotrexate (plus Folic acid) 3 8.3854 0.0387

Hydroxychloroquine 9 25.4841 0.0025

Leflunomide 9 35.2574 <.0001

Prednisolone 9 33.2572 0.0001

Penicillamine 9 28.5823 0.0008

In the analysis of maximum likelihood table, the effect of each medication was investigated in more detail. The results are outline below:

Infliximab: Currently taking Infliximab was found to be marginally associated with an increased frequency of severe BJM infection (P value: 0.07). The amount of this increase can reach 82 times more than patient who were not treated with this medicine at all (Table 4.22- 4.23 and Appendix Tables D.34- D.35).

Methotrexate (plus Folic acid): Currently taking MTX / Folic acid was found to reduce moderate BJM infection, significantly (P value: 0.02). The amount of this reduction is almost 66% of patient who did not need to take MTX, at all. (CI: 0.46 to 0.94) (Table 4.22-4.23 and Appendix Tables D.34- D.35).

Hydroxychloroquine: There is no evidence that currently taking this medication can affect BJM infection (Table 4.22-4.23 and Appendix Tables D.34- D.35).

Leflunomide: Currently taking Leflunomide was associated with an increased frequency of severe BJM infection (P value: 0.0014). The amount of this increase is 87 times greater than in patients who have never taken this medication at all (CI: 1.276 to 2.759) (Table 4.22-4.23 and Appendix Tables D.34- D.35).

166

Prednisolone: Currently taking Prednisolone was associated with significant increased frequency of moderate and severe bone, joint and muscle infection. The amount of this increase is in turn 87 and 152 times more than in patients who have never taken prednisolone at all (Table 4.22-4.23 and Appendix Tables D.34- D.35).

Penicillamine: There is no evidence that currently taking Penicillamine can affect BJM infection (Table 4.22-4.23 and Appendix Tables D.34- D.35).

In summary, the use of infliximab, leflunomide and prednisolone were associated with statistically significant increases in the frequency of severe BJM infections.

167

Table 4.22-Analysis of maximum likelihood estimate in bone, joint and muscle (BJM) infection

Analysis of Maximum Likelihood Estimates Bone/Joint/M Standard Wald Parameter uscle infection DF Estimate Error Chi-Square Pr > ChiSq Infliximab currently taking 1 1 -0.2314 0.7289 0.1007 0.7510 Infliximab currently taking 2 1 -0.0882 0.4578 0.0372 0.8471 Infliximab currently taking 3 1 0.5992 0.3308 3.2819 0.0700 Methotrexate currently taking 1 1 -0.4576 0.2957 2.3950 0.1217 (plus Folic acid) Methotrexate currently taking 2 1 -0.4093 0.1788 5.2389 0.0221 (plus Folic acid) Methotrexate currently taking 3 1 0.1239 0.1458 0.7217 0.3956 (plus Folic acid) Hydroxychloro currently taking 1 1 -0.2784 0.3205 0.7543 0.3851 quine Hydroxychloro currently taking 2 1 0.2671 0.1893 1.9918 0.1582 quine Hydroxychloro currently taking 3 1 0.1313 0.1708 0.5916 0.4418 quine Arava currently taking 1 1 0.1898 0.3336 0.3236 0.5694 (Leflunomide) Arava currently taking 2 1 0.1851 0.2077 0.7935 0.3730 (Leflunomide) Arava currently taking 3 1 0.6292 0.1968 10.2160 0.0014 (Leflunomide) Prednisolone currently taking 1 1 0.0363 0.3124 0.0135 0.9075 Prednisolone currently taking 2 1 0.6298 0.2394 6.9233 0.0085 Prednisolone currently taking 3 1 0.9273 0.2630 12.4305 0.0004 Penicillamine currently taking 1 1 1.6260 1.0242 2.5204 0.1124 Penicillamine currently taking 2 1 - 913.3 0.0002 0.9886 13.0777 Penicillamine currently taking 3 1 - 847.3 0.0002 0.9878 12.9395

168

Conclusion:

Differential effects on the frequency of BJM infections were observed amongst users of csDMARDs and bDMARDs. Amongst csDMARDs, Prednisolone and leflunomide were associated with a significant increased frequency of severe BJM infections, whereas methotrexate was associated with reduced frequency of moderate BJM infections. Amongst bDMARDs, infliximab was associated with an increased frequency of BJM infections in most categories and severe infection type was significant. The data eres less clear for hydroxychloroquine and penicillamine, but low numbers may have limited the capacity for

analysis (Table 4.22-4.23 and Appendix Tables D.1- D.35).

Table 4.23- Estimation of Odds ratios in Bone, Joint and Muscle infection Odds Ratio Estimates Bone/Joint/M Point 95% Wald Effect uscle infection Estimate Confidence Limits Infliximab currently taking Versus never taking 1 0.793 0.190 3.311 Infliximab currently taking Versus never taking 2 0.916 0.373 2.246 Infliximab currently taking Versus never taking 3 1.821 0.952 3.482 Methotrexate (plus Folic acid) currently taking Mild 0.633 0.354 1.130 Versus never taking Methotrexate (plus Folic acid) currently taking Moderate 0.664 0.468 0.943 Versus never taking Methotrexate (plus Folic acid) currently taking Severe 1.132 0.851 1.506 Versus never taking Hydroxychloroquine currently taking Versus never Mild 0.757 0.404 1.419 taking Hydroxychloroquine currently taking Versus never Moderate 1.306 0.901 1.893 taking Hydroxychloroquine currently taking Versus never Severe 1.140 0.816 1.594 taking Arava (Leflunomide) currently taking Versus never Mild 1.209 0.629 2.325 taking Arava (Leflunomide) currently taking Versus never Moderate 1.203 0.801 1.808 taking Arava (Leflunomide) currently taking Versus never Severe 1.876 1.276 2.759 taking Prednisolone currently taking Versus never taking Mild 1.037 0.562 1.913 Prednisolone currently taking Versus never taking Moderate 1.877 1.174 3.001 Prednisolone currently taking Versus never taking Severe 2.528 1.510 4.232 Penicillamine currently taking Versus never taking Mild 5.084 0.683 37.844 Penicillamine currently taking Versus never taking Moderate <0.001 <0.001 >999.999 Penicillamine currently taking Versus never taking Severe <0.001 <0.001 >999.999

169

3.7. Blood infection - analysis of Anti-RA medicines Amongst 21506 observations, 21/21506 (0.097 %) self-reported mild infection, 70/21506 (0.32%) self-reported moderate infection, 111/21506 (0.51%) self-reported severe infection, whereas 21304/21506 (99.06 %) reported no infection. In this model, participants with different categories of blood infection (presumed sepsis or septicaemia) were compared to those who did not report any infection (Appendix tables E1-E3). A multinomial logistic regression model was used. The reason for using this model is that the outcome is a non-binary categorical variable. Using pairwise Chi-square test without using the model can increase potential mistakes because the number of comparisons is large [20].

Table 4.24 Estimation of fitness of tests in blood infection Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2,670.26 2,718.38 SC 2,694.19 3,962.65 -2 Log L 2,664.26 2,406.38

Table 4.25 Estimation of fitness of tests in blood infection Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 257.8845 153 <.0001 Score 284.8579 153 <.0001 Wald 233.2076 153 <.0001

The model convergence status table (Appendix Tables E.5-E.7) shows that the test meets the criteria for accuracy and the variables fit the model of statistics. Overall the test shows a significant difference (lr Chi-square of 257.8845 with a P value of <.0001) between variable effects on blood infection [20] (Appendix table E.7).

In the model fit statistics table (Appendix table E.7), the likelihood ratio or lr (difference between -2 Log L or Deviance in the model which contains just the intercept and the one which contains both the intercept and covariates) is 257. The P value is highly significant (Appendix table E.1-E.7). This shows that in a model with covariates, the test strengthened. Furthermore, 170

the covariates were found to be impacting cofactors in blood infection. Other tests such as SC and AIC were also used to recheck this conclusion [22] (Appendix table E.6).

Wald Chi-Square for overall test is also highly significant (<.0001) with a Chi-square of almost 233 among 153 degrees of freedom. In other words, the impact on the blood infection is not the same for different groups. This Chi-square P value is almost equivalent to the p value in the overall Pearson test. Indeed, the logistic regression result is much the same as the frequency table (Tables 3.59-3.60) result because it is a large sample. As the model is a logistic regression and not a linear regression, the Chi-square test has been used to examine the comparison. [21] (Appendix E.7).

The Score test (Lagrange multiplier test) requires estimating only a single model. The test statistic is calculated based on the slope of the likelihood function at the observed values of the variables in the model. Usually the score tests are compared when parameters are added. They provide an estimation of how far the accuracy of the test improves when new parameters are added to or removed from existing parameters [21] (Appendix E.7). During the backward stepwise model in the next part of the model, the effects of the medications are dropped one by one to see how much change occurs in the chi-square and to get an estimate of the amount of impact of that medication in increasing blood infection [22] (Appendix E.8-E.31).

3.7.1. Wald Chi-square, Likelihood ratio test and Score test to test the significance of differences As the size of the study population in this study was large enough, any of these three tests can be used, but if the size of the sample were small, then it would be necessary to check all three tests to increase the reliability of the conclusions. Among these tests, the likelihood ratio test is the most reliable test, because it stays unchanged even if re-parametrization of the data is undertaken [22] (Appendix E.7).

3.7.2. Effects of different medications on blood infection

As the medication effects in this model are all qualitative, the degree of effect (impact) on blood infection can be easily determined by comparing these categorical variables [21].

171

For this section, a backward procedure in multinominal logistic regression was preferred. Logistic regression is required because the outcome is categorical and also because blood infection is divided into three categories of severity, viz: mild, moderate and severe. Moreover, there is a no infection category, (self-reported absence of infection). Accordingly, a multi nominal logistic regression model was deemed most appropriate (Appendix Table E.4). Furthermore, a backward stepwise approach is used here too, because it is more accurate than a forward procedure and considers the accumulating effect of all variables and starts with a bigger model. As the model fit statistics (Appendix Table E.6) show that using this large model is still well fitted to the data, it can be used appropriately. Additionally, there is no collinearity and none of any two variables are identical. This makes it easier to use the backward model (Tables 4.26-4.29).

According to the summary table in the backward procedure (Table 4.26 and Appendix Table E.32), the least significant effect is from Anakinra followed by Certolizumab, Infliximab, Rituximab, Leflunomide, Penicillamine, Golimumab, Cyclosporine, Sulphasalazine, Azathioprine, Abatacept, Tocilizumab, Methotrexate (plus Folic acid), IM Gold injection, Adalimumab and Etanercept. However, the effect of all of these medications was minimal and so they were dropped from the model.

172

Table 4.26 Summary of backward elimination of anti-RA medications in respect to Blood infection (sepsis or septicaemia) Summary of Backward Elimination Step Effect Removed DF Number In Wald Chi-Square Pr > ChiSq 1.00 Anakinra 9.00 17.00 0.28 1.00 2.00 Certolizumab 9.00 16.00 1.68 1.00 3.00 Infliximab 9.00 15.00 3.55 0.94 4.00 Rituximab 9.00 14.00 4.87 0.85 5.00 Leflunomide 9.00 13.00 5.34 0.80 6.00 Penicillamine 9.00 12.00 5.65 0.77 7.00 Golimumab 6.00 11.00 3.51 0.74 8.00 Cyclosporine 9.00 10.00 7.10 0.63 9.00 Sulphasalazine 9.00 9.00 7.13 0.62 10.00 Azathioprine 9.00 8.00 9.52 0.39 11.00 Abatacept 9.00 7.00 12.24 0.20 12.00 Tocilizumab 9.00 6.00 12.21 0.20 13.00 Methotrexate (plus Folic 3.00 5.00 4.80 0.19 acid) 14.00 IM Gold injection 9.00 4.00 15.55 0.08 15.00 Adalimumab 9.00 3.00 16.53 0.06 16.00 Etanercept 9.00 2.00 14.05 0.12

The only medications for which there was evidence of an effect on blood infection were: Hydroxychloroquine and Prednisolone (Table 4.27 and Appendix Table A.32).

Table 4.27 Effect of medications in causing Blood infection Type 3 Analysis of Effects Effect DF Wald Pr > ChiSq Chi-Square Hydroxychloroquine 9.00 18.10 0.03 Prednisolone 9.00 49.54 <.0001

173

Hydroxychloroquine: Taking Hydroxychloroquine was associated with a significant reduction in reporting blood infection. Although taking this medication (CI: 0.22 to 0.84 P value: 0.03) (Table 4.27-4.28 and Appendix Tables E.34- E.35).

Prednisolone: Use of prednisolone is strongly associated with an increased frequency of severe blood infections. These severe blood infections were increased by up to almost 431 times compared to participants who were never users of prednisolone (CI: 2.147 to 13.142) (Table 4.27-4.28 and Appendix Tables E.34- E.35).

In summary, use of Prednisolone carries considerable risk in respect to septicaemia.

Table 4.28 Analysis of maximum likelihood estimate in blood infection

Analysis of Maximum Likelihood Estimates Parameter Medicati Blood DF Estimat Standar Wald Pr > ChiS on Status Infectio e d Error Chi- q n Squar e currently Mild 1.0 -1.94 1.03 3.51 0.06 Hydroxychloroquine taking 0 Moderat 1.0 -0.29 0.34 0.72 0.40 e 0 Severe 1.0 -0.86 0.35 6.07 0.01 0 Mild 1.0 -0.40 0.56 0.50 0.48 Prednisolone 0 currently Moderat 1.0 0.23 0.33 0.47 0.49 taking e 0 Severe 1.0 1.67 0.46 13.06 0.001 0

174

Conclusion: Hydroxychloroquine and especially prednisolone substantially increase rates of severe blood infection and have a significant association with risk of blood infection. While hydroxychloroquine is associated with a reduction in this risk, taking prednisolone is significantly associated with a sharp increase in this risk (Table 4.28-4.29 and Appendix Tables E.1- E.35).

Table 4.29 Estimation of Odds ratios in blood infection Odds Ratio Estimates Effect Blood Point 95% Wald Infection Estimate Confidence Limits Hydroxychloroquine - currently taking vs 1.00 0.14 0.02 1.09 never taken 2.00 0.75 0.39 1.45 3.00 0.43 0.22 0.84 Prednisolone - currently taking vs never taken 1.00 0.67 0.22 2.02 2.00 1.25 0.66 2.39 3.00 5.31 2.15 13.14

175

3.8. Gastro-intestinal tract infection - analysis of medication confounders Among 21506 observations, 118/21506 (0.54 %) self-reported mild infection, 241/21506 (1.12%) self-reported moderate infection and 155/21506 (0.72%) self-reported severe infection, whereas 20992/21506 (97.6 %) reported no infection. In this model, participants with different categories of gastrointestinal infection were compared with ARAD participants, who did not self-report this type of infection (Appendix tables F1-F3). The statistical model used was Multinomial logistic regression. The reason for using this model is that the outcome is a non- binary categorical variable. Using pairwise Chi-square test without using the model can increase potential mistakes because the number of comparisons is very large [20].

The model convergence status table (Appendix Tables F.5-F.7) shows that the test meets the criteria for accuracy and the variables fit the statistical model. Overall the test shows significant effects (lr Chi-square of 233.3227 with a P value of <.0001) from anti RA medicines on GIT infection [20] (Appendix table F.7).

In the model fit statistics table (Appendix table F.7), the likelihood ratio or lr (difference between -2 Log L or Deviance in the model which contains just the intercept and the one which contains both the intercept and covariates) is 233.3227. The P value is highly significant (Table 4.30 and Appendix table F.1-F.7). This shows that a model with covariates is strengthening the test and that the covariates are actually impacting cofactors in GIT infections. Other tests such as SC and AIC are also used to recheck this conclusion [22] (Table 4.30 and Appendix table F.6).

Table 4.30 Estimation of the impact of confounders Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5,944.02 6,016.70 SC 5,967.95 7,260.97 -2 Log L 5,938.02 5,704.70

176

Wald Chi-square for overall test is also highly significant (0.0001) with a Chi-square of almost 233.3227 among 153 degrees of freedom. In other words, the impact on the GIT infection is not the same in different groups. This Chi-square P value is almost equivalent to the p value in the overall Pearson test. Indeed, the logistic regression result is much the same as the frequency table (Tables 3.59-3.60) result because it is a large sample. As a logistic regression and not a linear regression has been used, the Chi-Square test has been used for comparison. [21] (Table 4.31 and Appendix F.7).

Table 4.31 Estimation of fitness of tests in GIT infection Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 233.3227 153 <.0001 Score 267.8657 153 <.0001 Wald 231.4222 153 <.0001

During the backward stepwise phase in the next part of the model, the effects of the medications are dropped one by one to see how much there is a change in the Chi-square and to get an estimate of the impact of that medication with regard to increasing or decreasing GIT infection [22] (Appendix F8-F.31).

177

3.8.1. Wald Chi-square, Likelihood ratio test and Score test As the size of the study population in this study is large enough, any of these three tests can be used, but if the size of the sample is small then all three tests need to be used to increase the reliability of the conclusions. Among these tests, the likelihood ratio test is the most reliable test, because it remains unchanged even if the data is reparametrized [22] (Appendix F.7).

3.8.2. Effects of different medications on GIT infections

As the medication effects in this model are all qualitative, the degree of effect (impact) on any infection can be easily worked out by comparing these categorical variables [21].

For this section, a backward procedure in multinominal logistic regression is preferred. Logistic regression is required because the outcome is categorical and as GIT infection has three categories, notably mild, moderate and severe as well as a no infection category, multinominal logistic regression is an appropriate model (Appendix Table F.4). Also, backward stepwise is used here because it is more accurate than a forward procedure and considers the accumulating effect of all variables and starts with a bigger model. As the model fit statistics (Appendix Table F.6) show that using this large model is still fitted to the data, it can be used appropriately. Also, there is no collinearity and none of any two variables are identical. This makes it easier to use the backward model [22].

According to the summary of the backward procedure (Table 4.32), the least significant effect is from Certolizumab followed by Azathioprine, IM Gold injection, Golimumab, Tocilizumab, Etanercept, Leflunomide, Anakinra, Penicillamine, Abatacept, Methotrexate (plus Folic acid), Rituximab, Hydroxychloroquine, and Sulphasalazine (Table 4.32 and Appendix Table F.32).

178

Table 4.32 Summary of backward elimination of anti RA medications and risk of GIT infection Summary of backward elimination Step Effect Removed DF Number In Wald Chi-Square Pr > ChiSq 1.00 Certolizumab 9.00 17.00 1.04 1.00 2.00 Azathioprine 9.00 16.00 3.60 0.94 3.00 IM Gold injections 9.00 15.00 4.16 0.90 4.00 Golimumab 6.00 14.00 2.60 0.86 5.00 Tocilizumab 9.00 13.00 4.03 0.91 6.00 Etanercept 9.00 12.00 4.74 0.86 7.00 Leflunomide 9.00 11.00 7.01 0.64 8.00 Anakinra 9.00 10.00 8.19 0.52 9.00 Penicillamine 9.00 9.00 9.03 0.44 10.00 Abatacept 9.00 8.00 9.71 0.37 11.00 Folic acid plus Methotrexate 3.00 7.00 3.37 0.34 12.00 Rituximab 9.00 6.00 11.20 0.26 13.00 Hydroxychloroquine 9.00 5.00 12.07 0.21 14.00 Sulphasalazine 9.00 4.00 13.32 0.15

However, the effect of all these medications was minimal, so they were dropped from the model. The only medications with significant effects were: Adalimumab, Infliximab, Cyclosporine, and Prednisolone (Table 4.33 and Appendix Table F.33).

Table 4.33 Medications associated with either increased or decreased GIT infection Type 3 Analysis of Effects Effect DF Wald Pr > ChiSq Chi-Square Adalimumab 9.00 17.65 0.04 Infliximab 9.00 20.52 0.02 Cyclosporine 9.00 45.78 <.0001 Prednisolone 9.00 21.30 0.01

179

In the analysis of maximum likelihood table, the effect of each medication was examined in more detail:

Adalimumab: Currently taking Adalimumab was found to be significantly associated with a reduction in the chance of mild GIT infection. However, if patient does not need to take Adalimumab at all, the risk could be even less and can get up to 48 times less (CI: 0.314 to 0.888) (Table 4.34-4.35 and Appendix Tables F.34- F.35).

Infliximab: Currently taking Infliximab is associated with a significant increase (P Value: 0.0235) in the risk of moderate infection. The amount of this increase is almost 99 times more than patients who never took this medication (Table 4.34-4.35 and Appendix Tables F.34- F.35).

Cyclosporine: Currently taking this medication is associated with a higher risk of mild (P value of <0.0001) and moderate (P value of <0.0001) GIT infection. This approaches 500 times compared to patients who don’t take this medication (Table 4.34-4.35 and Appendix Tables F.34- F.35).

Prednisolone: Currently taking prednisolone is associated with an increase in the risk of severe GIT infection (P value: 0.0505) and there is marginal evidence (P value 0.065) that it can also increase moderate GIT infection. This increase is more than 50 times compared to patients who have never taken Prednisolone (Table 4.34-4.35 and Appendix Tables F.34- F.35).

In summary, use of Prednisolone was associated with an increased frequency of severe GIT infection and use of both Infliximab and Cyclosporin were associated with increased rates of moderate GIT infection

180

Table 4.34 Analysis of Maximum likelihood estimates in GIT infection Analysis of Maximum Likelihood Estimates Parameter Medication GIT DF Estimate Standard Wald Pr > ChiSq Status Infection Error Chi- Square Adalimumab currently 1.00 1.00 -0.64 0.26 5.81 0.02 taking 2.00 1.00 0.13 0.17 0.62 0.43 3.00 1.00 -0.10 0.22 0.22 0.64 Infliximab currently 1.00 1.00 -0.61 0.72 0.71 0.40 taking 2.00 1.00 0.69 0.31 5.13 0.02 3.00 1.00 0.24 0.46 0.27 0.61 Cyclosporine currently 1.00 1.00 1.89 0.47 16.32 <.0001 taking 2.00 1.00 1.83 0.36 26.26 <.0001 3.00 1.00 0.75 0.72 1.09 0.30 Prednisolone currently 1.00 1.00 0.19 0.30 0.39 0.53 taking 2.00 1.00 0.44 0.24 3.39 0.07 3.00 1.00 0.54 0.28 3.83 0.05

Conclusion: Differential effects on the frequency of GIT infections were observed amongst users of csDMARDs and bDMARDs. Amongst csDMARDs, Cyclosporine was associated with an increase in mild and moderate self-reported GIT infections, whereas prednisolone was associated with an increase in severe self-reported GIT infections. Amongst bDMARDs, use of infliximab was associated with an increase in moderate self-reported GIT infections, whereas adalimumab was associated with a protective effect for mild, but not moderate or severe GIT infections. The clinical relevance of this latter finding is uncertain. Once again, the potential for corticosteroid therapy to confer risk for severe infection in multiple systems was evident.

181

Table 4.35 Estimation of odds ratios in GIT infection Odds Ratio Estimates Effect GIT Point Estimate 95% Wald Infection Confidence Limits Adalimumab - currently taking vs never taking 1.00 0.53 0.31 0.89 2.00 1.14 0.82 1.58 3.00 0.90 0.58 1.40 Infliximab - currently taking vs never taking 1.00 0.55 0.13 2.22 2.00 2.00 1.10 3.64 3.00 1.27 0.51 3.16 Cyclosporine - currently taking vs never taking 1.00 6.64 2.65 16.65 2.00 6.21 3.09 12.48 3.00 2.12 0.52 8.71 Prednisolone - currently taking vs never taking 1.00 1.21 0.67 2.17 2.00 1.55 0.97 2.48 3.00 1.72 1.00 2.97

3.9. Nervous System infection - analysis of medication confounders Amongst 21506 observations, 9/21506 (0.0418 %) self-reported mild infection, 9/21506 (0.0418%) self-reported moderate infection, 12/21506 (0.055%) self-reported severe infection, whereas 21476/21506 (99.86 %) reported no infection. In this model, nervous system infections of different severity were compared to those in participants who did not have such infection (Appendix tables G1-G3). The model used was a Multinomial logistic regression model. The reason for using this model is because the outcome is a non-binary categorical variable. Using pairwise Chi-square test without using the model can increase potential mistakes because the number of comparisons is very large [20]. The model convergence status table (Appendix Tables G.5-G.7) shows that the test does not meet the criteria for accuracy and the variables do not fit the model of statistics.

Table 4.36 Estimation of the impact of confounders. Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.79 719.40 SC 549.71 1,963.67 -2 Log L 519.79 407.40

182

Overall, the test does not show any significant effect of anti-RA medicines on nervous system infections (lr Chi-Square of 112.3885 with a P value of 0.9943) [20] (Table 4.36-4.37 and Appendix table G.7).

Table 4.37 Estimation of fitness of tests in nervous system infection Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 112.39 153.00 0.99 Score 170.52 153.00 0.16 Wald 83.37 153.00 1.00

This means that the test is not reliable due to several potential reasons, including sample size, and so it is not possible to draw reliable conclusions from the analysis.

3.10. TB infection - analysis of medication confounders A multinomial logistic regression model was used to compare different the severities of TB infection with a control group (respondents who have not had TB infection). The reason for using this model was because the outcome was a non-binary categorical variable. The results indicate that amongst 21506 observations 3/21506 (0.013 %) had mild TB infection, 6/21506 (0.027%) had moderate TB infection and 2/21506 (0.0092%) reported severe TB infection, whereas 21495/21506 (99.94 %) reported no infection at all. The model convergence status table shows that the test does not meet the criteria for accuracy and the variables do not fit the model of statistics. Overall the test does not show significant difference (lr Chi-square of 93.7402 with a P value of 1) in between variable effects on the TB infection. This means that the test is not reliable due to several potential reasons including sample size and we cannot judge the conclusion out of such analysis.

Table 4.38 Estimation of the impact of confounders Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 216.60 428.86 SC 240.53 1,673.13 -2 Log L 210.60 116.86

183

Overall, the test does not show a significant difference (lr Chi-square of 93.7402 with a P value of 1) between variable effects on TB infection (Table 4.39 and Appendix table H.7).

Table 4.39 Estimation of fitness of tests in TB infection Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 93.74 153.00 1.00 Score 191.37 153.00 0.02 Wald 26.52 153.00 1.00

This means that the test is not reliable due to several potential reasons including sample size and so it is not possible to draw reliable conclusions from the analysis.

3.11. Urinary tract infection - analysis of medication confounders Amongst 21506 observations, 290/21506 (1.34 %) self-reported mild infection, 833/21506 (3.87%) self-reported moderate infection, and 256/21506 (1.19%) self-reported severe infection, whereas 20127/21506 (93.58 %) reported no infection. In this model, persons with different categories of urinary tract infection are compared with people who don’t have this type of infection (Appendix tables I1-I3). A multinomial logistic regression model was employed to analyse the data. The reason for using this model is because the outcome is a non- binary categorical variable. Using pairwise Chi-square test without using the model can increase potential mistakes because the number of comparisons is very large [20]. The model convergence status table (Appendix Tables I.5-I.7) shows that the test meets the criteria for accuracy and the variables fit the model of statistics. Overall the test shows a significant difference (lr Chi-square of 442.0070 with a P value of less than 0.0001) between variable effects on urinary tract infection [20] (Table 4.40-4.41 and Appendix table I.7).

Table 4.40 Estimation of the impact of confounders Model Fit Statistics Criterion Intercept Only Intercept and covariates AIC 12856.1 12720.09 SC 12880.03 13964.36 -2Log L 12850.1 12408.09

184

In the model fit statistics table (Appendix table I.7), the likelihood ratio or lr (difference between -2 Log L or Deviance in the model which contains just the intercept and the one which contains both the intercept and covariates) is 442. The P value is highly significant (Appendix table I.1-I.7). This shows that a model with covariates strengthens the test and that the covariates are actually impacting cofactors in urinary tract infection. Other tests such as SC and AIC are also used to recheck this conclusion [22] (Tables 4.40-4.41 and Appendix table I.6).

Table 4.41 Estimation of fitness of tests in Urinary tract infection Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 442.0070 153 <.0001 Score 494.1448 153 <.0001 Wald 442.5757 153 <.0001

The Wald Chi-Square for overall test is also highly significant (0.0001) with a Chi-square of almost 442 among 153 degrees of freedom. In other words, the impact on urinary tract infection is not the same in different groups. This Chi-Square P value is almost equivalent to the p value in the overall Pearson test. Indeed, the logistic regression result is much the same as in the frequency table (Table 3.47) result, because it is a large sample. As a logistic regression and not a linear regression model was used, Chi-square test for comparison. [21] (Appendix table I.7).

185

3.11.1. Wald Chi-square, Likelihood ratio test and Score test to test significance of differences As the size of the study population in this study is enough, we can use any of these three tests, but if the size of the sample is small then it is necessary to check all three tests to increase the reliability of the conclusions. Among these tests, the likelihood ratio test is the most reliable test, because it stays unchanged even if the testing is reparametrized 3.10.1. Effects of different medications on urinary tract infection [22] (Appendix I.7).

3.11.2. Effects of medications on Urinary tract infection As the medication effects in this model are all qualitative, the degree of effect (impact) on urinary tract infections infection can be easily worked out by comparing the categorical variables [21].

For this section, a backward procedure in multinominal logistic regression was preferred. Logistic regression is required because the outcome is categorical and as urinary tract infection has three categories, notably mild, moderate and severe infection as well as a no infection report, multinominal logistic regression is an appropriate model (Appendix Table I.4). The backward stepwise procedure is used here because it is more accurate than a forward procedure and considers the accumulating effect of all variables and starts with a bigger model. As the model fit statistics (Appendix Table I.6) show that this large model is still fitted to the data, it is appropriate for use. Also, there is no collinearity and no two variables are identical. This makes it easier to use the backward model [22].

According to the summary table in the backward procedure (Table 4.42), the least significant effect is from Abatacept followed by Anakinra, Certolizumab, Golimumab, Tocilizumab, Sulphasalazine, Adalimumab, Rituximab, and Folic acid plus Methotrexate (plus Folic acid). However, the effect of all these medications was found to be minimal and they were therefore dropped from the model. There was evidence for an effect of the following medications in respect to Urinary tract infection: Etanercept, Infliximab, Hydroxychloroquine, Leflunomide, Azathioprine, Cyclosporine, Prednisolone, IM Gold Injection, and Penicillamine (Table 4.42 and Appendix Table I.32).

186

Table 4.42 Summary of backward elimination of anti RA medications and risk of urinary tract infection

Summary of Backward Elimination Step Effect DF Number Wald Pr > ChiSq Removed In Chi-Square 1.00 Abatacept 9.00 17.00 2.97 0.97 2.00 Anakinra 9.00 16.00 3.68 0.93 3.00 Certolizumab 9.00 15.00 6.06 0.73 4.00 Golimumab 6.00 14.00 4.11 0.66 5.00 Tocilizumab 9.00 13.00 9.34 0.41 6.00 Sulphasalazine 9.00 12.00 10.71 0.30 7.00 Adalimumab 9.00 11.00 13.75 0.13 8.00 Rituximab 9.00 10.00 15.73 0.07 9.00 Folic acid plus 3.00 9.00 7.47 0.06 Methotrexate

In the analysis of maximum likelihood table, the effect of each medication was examined in more detail.

187

Table 4.43 Analysis of Maximum likelihood estimate in Urinary tract infection Analysis of Maximum Likelihood Estimates Parameter Medication Urinary DF Estimate Standard Wald Pr > ChiSq Status system Error Chi- Infection Square Infliximab currently Mild 1.00 0.33 0.34 0.98 0.32 taking Moderate 1.00 0.02 0.22 0.01 0.91 Severe 1.00 -1.49 0.74 4.06 0.04 Hydroxychloroquine currently Mild 1.00 0.07 0.17 0.16 0.69 taking Moderate 1.00 0.08 0.11 0.53 0.47 Severe 1.00 -0.18 0.20 0.81 0.37 Leflunomide currently Mild 1.00 -0.21 0.19 1.17 0.28 taking Moderate 1.00 -0.14 0.12 1.19 0.28 Severe 1.00 -0.53 0.24 5.01 0.03 Azathioprine currently Mild 1.00 -11.88 269.90 0.001 0.96 taking Moderate 1.00 -0.80 0.59 1.84 0.17 Severe 1.00 -11.93 256.60 0.001 0.96 Cyclosporine currently Mild 1.00 1.52 0.38 16.37 <.0001 taking Moderate 1.00 1.06 0.29 13.23 0.001 Severe 1.00 0.69 0.53 1.68 0.20 Prednisolone currently Mild 1.00 -0.03 0.19 0.03 0.86 taking Moderate 1.00 0.36 0.12 9.16 0.001 Severe 1.00 0.78 0.24 10.51 0.001 IM Gold injection currently Mild 1.00 -0.03 0.72 0.001 0.97 taking Moderate 1.00 0.48 0.33 2.10 0.15 Severe 1.00 1.11 0.44 6.56 0.01 Penicillamine currently Mild 1.00 -12.05 451.70 0.001 0.98 taking Moderate 1.00 -12.12 270.80 0.001 0.96 Severe 1.00 -12.08 478.60 0.001 0.98

Based on the likelihood estimate table and odd’s ratio the following results are concluded:

Etanercept: There is no evidence that currently taking Etanercept is associated with urinary tract infection (Table 4.43-4.45 and Appendix Tables I.34- I.35).

188

Infliximab: Currently taking Infliximab is associated with a reduced (P Value: 0.0438) frequency of Urinary tract infection. However, patients who have never taken Infliximab have almost 80 times less chance for Urinary tract infection (CI: 0.053 to 0.960) (Table 4.43-4.45 and Appendix Tables I.34- I.35).

Hydroxychloroquine: There is no evidence that currently taking Hydroxychloroquine is associated with urinary tract infection (Table 4.43-4.45 and Appendix Tables I.34- I.35).

Leflunomide: Currently using Leflunomide is associated with a reduced reports of severe urinary tract infection. The amount of this reduction was almost 58% compared to non-users of LEF (CI: 0.367 to 0.936 P value: 0.0252) (Table 4.43-4.45 and Appendix Tables I.34- I.35).

Azathioprine: There is no evidence that currently taking Azathioprine is associated with urinary tract infection (Table 4.43-4.45 and Appendix Tables I.34- I.35).

Cyclosporine: Currently taking Cyclosporine is associated with an increased frequency of mild and moderate urinary tract infection. Compared to participants who have never taken Cyclosporine, the extent of this increase is 358-fold for mild and 187-fold for moderate urinary tract infection.

Prednisolone: Currently taking Prednisolone is associated with an increased frequency of moderate and severe urinary tract infection. Compared to participants who have never taken Prednisolone, the extent of the increase is 43-fold for moderate and 117-fold for severe urinary tract infection (Table 4.43-4.45 and Appendix Tables I.34- I.35).

IM Gold Injection: Currently receiving parenteral Gold is associated with an increase in the frequency of severe urinary tract infection (P value: 0.0104). Compared to participants who have never received parenteral Gold, the extent of the increase is 204-fold (CI: 1.299 to 7.151) (Table 4.43-4.45 and Appendix Tables I.34- I.35).

Penicillamine: There was no evidence that currently taking Penicillamine is associated with an increased frequency of urinary tract infection (Table 4.43-4.44 and Appendix Tables I.34- I.35).

Conclusion: In summary use of Prednisolone and Cyclosporin was found to be associated with moderate and severe UTIs in the case of the former and moderate and mild, but not severe UTIs in the case of the latter. Use of Infliximab and Leflunomide appeared to have protective effects in respect to UTI.

189

Table 4.44 Estimation of Odds ratios in participants with urinary tract infection Odds Ratio Estimates Effect Urinary system Point 95% Wald Infection Estimate Confidence Limits Etanercept - currently taking vs Mild 1.39 1.04 1.84 never taken Moderate 0.90 0.75 1.06 Severe 0.91 0.65 1.26 Infliximab - currently taking vs Mild 1.39 0.72 2.69 never taken Moderate 1.03 0.67 1.58 Severe 0.23 0.05 0.96 Hydroxychloroquine - currently Mild 1.07 0.76 1.51 taking vs never taken Moderate 1.08 0.87 1.34 Severe 0.84 0.57 1.23 Leflunomide - currently taking vs Mild 0.81 0.56 1.19 never taken Moderate 0.87 0.68 1.12 Severe 0.59 0.37 0.94 Azathioprine - currently taking vs Mild <0.001 <0.001 >999.99 never taken Moderate 0.45 0.14 1.43 Severe <0.001 <0.001 >999.99 Cyclosporine - currently taking vs Mild 4.59 2.19 9.60 never taken Moderate 2.88 1.63 5.09 Severe 1.99 0.70 5.66 Prednisolone - currently taking vs Mild 0.97 0.67 1.40 never taken Moderate 1.43 1.14 1.81 Severe 2.17 1.36 3.47 IM Gold - currently taking vs never Mild 0.97 0.24 3.98 taken Moderate 1.62 0.84 3.11 Severe 3.05 1.30 7.15 Penicillamine - currently taking vs Mild <0.001 <0.001 >999.99 never taken Moderate <0.001 <0.001 >999.99 Severe <0.001 <0.001 >999.99

As the anatomy of urinary system is different between two sexes it will be appropriate to assess if these differences can modify the effect of tablets. In other word if the effect of Anti RA

190

medication on UTI is the same for male and female sexes. In table 4.45 the interaction between anti RA medication and sex in UTI has been assessed.

Table 4.45 Estimation of interaction between anti RA medication and sex

Type 3 Analysis of Effects

Effect DF Wald Pr > ChiSq Chi-Square

Interaction with Etanercept 2 17.4683 0.0002

Interaction with Tocilizumab 2 11.0376 0.0040

Interaction with Hydroxychloroquine 3 8.9707 0.0297

Interaction with Leflunomide 2 9.9919 0.0068

Interaction with Prednisolone 2 15.1467 0.0005

Table 4.45 shows that sex modifies effects of Etanercept, Tocilizumab, Hydroxychloroquine, Leflunomide and Prednisolone. In order to find the differences in more details, table 4.46 presents the pairwise test in each anti RA medication. Based on the results of this table Leflunomide is associated with increased rate of infection in male sex, but all other discussed anti RA medications (Etanercept, Tocilizumab, Hydroxychloroquine and prednisolone) were associated with less frequent UTI in male than in female (Table 4.46).

With Etanercept (odds ratio 0.10) and Tocilizumab (odds ratio 0.0044) and prednisolone (odds ratio 0.049), the effect of these medications is stronger in the female sex than the male sex. However, Leflunomide (odd’s ratio of 0.88) can significantly increase UTI in the male sex.

191

Table 4.46 Significant Sex interactions with anti-RA medications

Analysis of Maximum Likelihood Estimates

Parameter Wald Standard DF Estimate Chi- Pr > ChiSq Error Square

Intercept male 1 -1.2366 0.2185 32.0290 <.0001

Intercept female 1 1.7962 0.2227 65.0781 <.0001

Sex male 1 0.3896 0.4426 0.7748 0.3787

Sex female 0 0 . . .

Interaction with currently male 1 -1.0522 0.4816 4.7733 0.0289 Etanercept taking

Interaction with currently male 1 -4.1742 1.3291 9.8643 0.0017 Tocilizumab taking

Interaction with currently male 1 -0.5432 0.5263 1.0652 0.3020 Hydroxychloroquine taking

Interaction with currently male 1 1.1168 0.5347 4.3624 0.0367 Leflunomide taking

Interaction with currently male 1 -1.7637 0.5440 10.5125 0.0012 Prednisolone taking

Conclusion:

Differential effects on the frequency of urinary tract infections were observed amongst users of csDMARDs and bDMARDs. Cyclosporine, IM Gold and Prednisolone increased the risk for urinary tract infections, whereas Leflunomide protected against severe urinary tract infections mainly in female sex but can significantly increase risk of UTI in male sex. None of the evaluated biologic agents increased the frequency of UTIs in both sexes and Infliximab had an unequivocal protective effect, the clinical significance of which is uncertain.

192

3.12. Viral infection - analysis of medication confounders Amongst 21506 observations, 435/21506 (2.022 %) self-reported mild infection, 837/21506 (3.89%) self-reported moderate viral infection; 305/21506 (1.41%) self-reported severe viral infection, whereas for 19929/21506 (92.66 %) patient-visits, no infections were reported. In this model different categories of viral infection were compared with people who did not self- report such infection (Table 4.47) (Appendix tables I1-I3). A multinomial logistic regression model was used to analyse the data. The reason for using this model is because the outcome is a non-binary categorical variable. Using pairwise Chi-square test without using the model can increase potential mistakes because the number of comparisons is very large [20].

Table 4.47 Estimation of the impact of confounders Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14,465.34 14,470.56 SC 14,489.27 15,714.83 -2 Log L 14,459.34 14,158.56

The model convergence status table (Appendix Tables I.5-I.7) shows that the test meets the criteria for accuracy and the variables fit the model of statistics. Overall the test shows significant difference (lr Chi-square of 300.7784 with a P value of less than 0.0001) between variable effects on VIRAL infections [20] (Table 4.47) (Appendix table I.7).

Table 4.48 Estimation of fitness of tests in viral infection Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 300.7784 153 <.0001 Score 331.3978 153 <.0001 Wald 311.3192 153 <.0001 In the model fit statistics table (Appendix table I.7), the likelihood ratio or lr (difference between -2 Log L or deviance in the model (which just contains intercept and the one which contains intercept and covariates) is 300.7784 (Table 4.48). The P value is highly significant (Table 4.48) (Appendix table I.1-I.7). This shows that a model with covariates is making the test more significant and covariates are actually impacting cofactors in respect to VIRAL

193

infections. Other tests such as SC and AIC are also used to recheck this conclusion [22](Appendix table I.6).

Wald Chi-square for overall test is also highly significant (0.0001) with a Chi-square of almost 300 among 153 degrees of freedom. In other words, the impact on VIRAL infection is not the same in different groups. This Chi-square P value is almost equivalent to the p value in the overall Pearson test. Indeed, the logistic regression result is much the same as it is in the frequency table (Tables 3.63-3.64) because it is a large sample. As the model is a logistic regression and not a linear regression model, the Chi-square test has been used for comparison. [21] (Table 4.48) (Appendix table I.7).

The Score test (Lagrange multiplier test) requires estimating only a single model. The test statistic is calculated based on the slope of the likelihood function at the observed values of the variables in the model. Usually, the score tests are compared when new parameters are added. It gives an estimate of how far the accuracy of the test improves when new parameters are added, or existing parameters are deleted [21] (Table 4.48) (Appendix table I.7). During the backward stepwise phase in the next part of the model, the effects of the medications are dropped one by one to see how much there is a change in the Chi-square and to get an estimate of the impact of that medication with regard to increasing or decreasing VIRAL infection [22] (Appendix table I.8-I.31).

3.12.1. Selection between Wald Chi-square, Likelihood ratio test and Score test to test significance of differences

As the size of the study population in this study is large enough, any of these three tests can be used, but if the size of the sample is small then it is necessary to check all three tests to increase the reliability of the conclusions. Among these tests, the likelihood ratio test is the most reliable test, because it stays unchanged even if the data being tested is reparametrized what we are testing [22] (Table 4.49) (Appendix table I.7).

3.12.2. Effects of different medications on the frequency of viral infections

As the variables are all categorical variables, the effects of each variable on viral infection can be examined by studying its coefficient, directly [21].

194

For this section, a backward procedure in multinominal logistic regression was preferred. Logistic regression is required because the outcome is categorical and as viral infection has three categories, notably: mild, moderate and severe infection as well as no infection report, multinominal logistic regression is an appropriate model (Table 4.49) (Appendix Table I.4). Also, a backward stepwise procedure is used here because it is more accurate than a forward procedure and considers the accumulating effect of all variables and starts with a bigger model. As the model fit statistics (Appendix Table I.6) show that this large model is still fitted to the data, it is appropriate for use. Furthermore, there is no collinearity and no two variables are identical. This makes it easier to use the backward model [22].

As can be seen in the summary table for the backward procedure analysis (Table 4.49 and Appendix Table I.32), the least significant effect is from Penicillamine, Certolizumab, Golimumab, Leflunomide, Abatacept, IM Gold injection, Azathioprine, Sulphasalazine, Anakinra, Tocilizumab, Adalimumab, Rituximab, and Infliximab.

Table 4.49 Summary of backward elimination of anti-RA medications and risk of viral infection

Step Effect DF Number Wald Chi-Square Pr > ChiSq Removed In 1 Penicillamine 9 17 4.95 0.84 2 Certolizumab 9 16 8.19 0.51 3 Golimumab 6 15 5.47 0.48 4 Leflunomide 9 14 7.71 0.56 5 Abatacept 9 13 10.03 0.35 6 IM Gold injection 9 12 10.95 0.28 7 Azathioprine 9 11 11.21 0.26 8 Sulphasalazine 9 10 9.91 0.36 9 Anakinra 9 9 10.14 0.34 10 Tocilizumab 9 8 11.97 0.21 11 Adalimumab 9 7 15.06 0.09 12 Rituximab 9 6 16.70 0.05 13 Infliximab 9 5 13.14 0.15

195

However, the effect of all these medications was minimal and they were, therefore, dropped from the model. According to type 3 analysis of effects (Table 4.32), medications which significantly affected the frequency of viral infection in RA include: Etanercept, Methotrexate (plus Folic acid), Hydroxychloroquine, Cyclosporine, and Prednisolone (Table 4.51 and Appendix Table 4.32).

Table 4.50Medications which increase the frequency of viral infection

Type 3 Analysis of Effects Effect DF Wald Pr > ChiSq Chi-Square Etanercept 9.00 43.98 <.0001 Folic acid plus Methotrexate 3.00 13.78 0.001 Hydroxychloroquine 9.00 29.30 0.001 Cyclosporine 9.00 29.31 0.001 Prednisolone 9.00 25.64 0.001

In the analysis of maximum likelihood table, the effect of each medication is examined in more detail:

 Etanercept: There is marginal evidence that currently taking Etanercept is associated with an increase in mild viral infection (P value 0.0573). This increase is almost 24 times more than in people who never taken this medication (CI: 0.993 to 1.550) (Table 4.51-4.52 and Appendix Tables I.34- I.35).

 Methotrexate (plus Folic acid): Currently taking Methotrexate (plus Folic acid) is associated with a reduction in mild and moderate viral infection. However, if the patient does not take this medication at all, the risk will be almost 25 times less (Table 4.51- 4.52 and Appendix Tables I.34- I.35).

 Hydroxychloroquine: Currently taking Hydroxychloroquine is associated with an increase in moderate viral infection (P value 0.0117). This risk is almost 30 times more than in people who never took this medication (Table 4.51-4.52 and Appendix Tables I.34- I.35).

 Cyclosporine: Currently taking Cyclosporine is strongly associated with an increase in the risk of moderate viral infection (P value: 0.0008). The amount of this increase is 167

196

times more than in patients who have never taken Cyclosporine (CI 1.503 to 4.777) (Table 4.51-4.52 and Appendix Tables I.34- I.35).

 Prednisolone: Currently taking Prednisolone is associated with increased rates of moderate and severe viral infection. The extent of this increase is 41-fold greater in the case of moderate and 108-fold greater in the case of severe viral infection respectively (Table 4.51-4.52 and Appendix Tables I.34- I.35).

Table 4.51 Analysis of maximum likelihood estimate in Viral infection Analysis of Maximum Likelihood Estimates Parameter Medication Viral DF Estimate Standard Wald Chi- Pr > C Status Infection Error Square hiSq Mild 1.00 0.22 0.11 3.61 0.06 Etanercept currently Moderate 1.00 -0.07 0.09 0.66 0.42 taking Severe 1.00 0.02 0.14 0.01 0.91

Mild 1.00 -0.28 0.12 5.25 0.02 Methotrexate (plus currently Moderate 1.00 -0.25 0.09 8.16 0.001 Folic acid) taking Severe 1.00 -0.13 0.14 0.88 0.35

Mild 1.00 0.22 0.14 2.55 0.11 Hydroxychloroquine currently Moderate 1.00 0.26 0.11 6.35 0.01 taking Severe 1.00 0.14 0.17 0.66 0.42

Mild 1.00 0.001 0.59 0.001 1.00 Cyclosporine currently Moderate 1.00 0.99 0.30 11.16 0.001 taking Severe 1.00 0.06 0.72 0.01 0.94

Mild 1.00 0.001 0.15 0.001 0.98 Prednisolone currently Moderate 1.00 0.35 0.12 7.81 0.01 taking Severe 1.00 0.74 0.22 11.18 0.001

Conclusion:

Differential effects on the frequency of viral infections were observed amongst users of csDMARDs and bDMARDs. Amongst csDMARDs recipients, Cyclosporine and Prednisolone substantially increased the risk for viral infections, whereas only a modest increase was 197

observed with Methotrexate (plus Folic acid) and Hydroxychloroquine. Amongst bDMARDs recipients, Etanercept increased viral infections slightly. Overall no major effect was observed amongst bDMARDs users (Table 4.51-4.52 and Appendix Tables I.34- I.35).

Table 4.52 Estimation of odds ratios in viral infection Odds Ratio Estimates Effect Viral Point 95% Wald Infection Estimate Confidence Limits Hydroxychloroquine - currently Mild 1.25 0.95 1.64 taking vs never taken Moderate 1.30 1.06 1.60 Severe 1.15 0.82 1.60 Cyclosporine - currently taking vs Mild 1.00 0.32 3.16 never taken Moderate 2.68 1.50 4.78 Severe 1.06 0.26 4.31 Prednisolone - currently taking vs Mild 1.00 0.74 1.34 never taken Moderate 1.41 1.11 1.80 Severe 2.09 1.36 3.22

3.12.3 Chapter Conclusion Infections of diverse severity were observed commonly in the ARAD cohort, in keeping with the high rates expected for active RA in a rheumatoid population biased toward the higher end of the age spectrum, where moderately high levels of functional impairment are operative and comorbidities are common. Based on ARAD data from 2001 to 2014, the highest to lowest rates of major organ infections reported by questionnaire respondents receiving diverse therapies were: EENT infection, skin and nail infection, lung infection, viral infection, kidney and urinary tract infection (Figure 4.1). As might be expected, the frequency of use of one or more than one anti-RA medications of various types was high amongst RA participants in ARAD. This accords with the targeting of patients who were about to commence a biologic therapy and in whom prior usage of csDMARDs was government-mandated, thereby ensuring use of multiple therapies not only in the quest for disease control, but also to satisfy prescribing restrictions.

198

Anti-RA medications were found to be used with divergent frequencies across the full range of RA respondents. From the most frequent to the least frequent medication, anti-RA medications used were Etanercept, Adalimumab, Methotrexate, Hydroxychloroquine, Sulphasalazine, Rituximab, Abatacept, Prednisolone, Tocilizumab, Infliximab and Leflunomide respectively (Table 4.1). Overall, although a number of differences exist between current RA treatment guidelines, there are some general principles. Remission or low disease activity is the preferred target. csDMARDs should be started soon after diagnosis and, usually, methotrexate is in the first line. It is important to monitor disease activity regularly and, if disease remains active persistently, biologics therapies should be used, as well[15].

Strengths of this study include large numbers of participants derived from real-world experience in community clinical practice and the considerable number of sequential visits. Substantial confidence concerning the primary diagnosis of RA is a further strength, since subsets of patients have been classified under ACR criteria with strong diagnostic concordance apparent. The large numbers aid in statistical analysis and the robustness of the statistical modelling strengthens the analysis.

This study has several limitations. Importantly, infections were self-reported and unvalidated, so their veracity cannot be substantiated. No microbiological reports or family physician corroborations were available. A different form of categorisation, whilst readily understood (notably: mild or moderate or severe), was, on the one hand, helpful but, on the other hand, not, since it precluded comparison with the more widely utilised categories of (SI, frequently found in publications and meta-analyses. Furthermore, SIs could not be easily deduced, since hospitalisation data was not available across the full duration of the study period, notably 2001 to 2014. A further pitfall was the lack of comparability between users of bDMARDs and those who had not taken such medication. Those who did not progress to bDMARDs had less severe disease that was, for the most part, amenable to simpler therapy and, thus, differences in respect to type and severity of infection may relate to differences in disease severity and not the type of medications taken. The agents used for treatment varied considerably and reflected the timing of introduction for clinical use and also prescriber preferences. Thus, the number of bDMARDs for different categories is skewed toward TNF inhibitors, which in a number of cases limited comparability with other bDMARDs due to uneven numbers of users

199

and sometimes very small numbers of recipients. Furthermore, this study was carried out in the pre-orally active csDMARDs era.

All in all, the reports for infection in RA came to a total of 757/1947 (38.88 %) in comparison to 1190/1947 (61.11%) in whom there was no infection. • In EENT infection, most participants reported infection of moderate severity.

• In lung infection, most participants reported infections in a moderate (6.41%) or severe category (2.9%).

• In skin and nail infection, most participants reported either mild (5.82%) or moderate (4.83%) infection.

• In artificial joint infection, most participants reported infections in either a severe (0.36%) or moderate (0.18%) category.

• In bone, joint and muscle infections, most participants reported severe (1.12%) or moderate (0.99%) infections.

• In blood infection, most participants reported infections of either severe (0.51%) or moderate (0.31%) severity.

• In GIT infection, most participants reported infections in either a moderate (1.12%) or severe (0.72%) category.

• In urinary tract infection, most participants reported infections in either a mild (1.34%) or moderate (3.87%) category.

• In viral infection, most participants reported infections in either a mild (2.022%) or moderate (3.89%) category.

• Nervous system and mycobacterium tuberculosis (TB) infections were very rare.

Based on the findings in this study, the csDMARDs and bDMARDs drugs that either protect against or predispose to infection in RA can be tabulated as shown in the table below (Table 4.53). Details are provided for different organ systems. Prednisolone was found to strongly predispose to moderate or severe infections in multiple organ systems, notably: EENT; lung; skin and nail; bone, joint and muscle; blood; GIT; urinary tract and in respect to viral infections. This finding accords with clinical experience and the well documented capacity of corticosteroids to predispose to infection in many systems. Other csDMARDs associated strongly with moderate or severe infection were cyclosporine for multiple systems (EENT, 200

lung, GIT, urinary tract and viral infections), hydroxychloroquine for blood infection alone and parenteral Gold for urinary tract infection alone. As cyclosporine has potent immunosuppressive properties, it is not surprising that it is implicated in infections in multiple organ systems. Whether the hydroxychloroquine and parenteral gold observations are clinically important is uncertain.

Amongst bDMARDs, only Infliximab was associated with moderate or severe infections in multiple organ systems (EENT, skin and nail, GIT). Adalimumab was associated with moderate or severe infection in the skin and nails alone. Etanercept, Certolizumab, Golimumab, Abatacept, Tocilizumab and Rituximab were not associated with moderate or severe infection in any system. The possibility that these agents do predispose to moderate or severe infection when used in combination with other agents cannot be discounted. Since, for some of these infection,s the numbers of patients who contracted such infections was less than 100 (Blood infection, GIT infection, Nervous system infection) and the number of observations over time correspondingly smaller, there is a need to exercise caution in applying these sample results to the wider population.

Several csDMARDs appear to protect against infections in selected systems. For example, methotrexate was associated with a protective effect in EENT and viral infections, leflunomide was associated with a protective effect in urinary tract infection alone and hydroxychloroquine against viral infection alone. Whether such effects are clinically meaningful is also uncertain, but somewhat doubtful. Amongst bDMARDs, etanercept appeared to protect against EENT, lung and viral infection, whereas adalimumab protected against lung, artificial joint and GIT infection. Infliximab was associated with a protective effect in urinary tract infection. Again, such effects are of uncertain significance and need to be independently validated.

201

Table 4.53 Summary of anti RA medication impacts on different types of infection Type of infection Safest Medications Medications associated with higher rates of moderate or severe infections EENT Etanercept, Methotrexate Cyclosporine, Prednisolone, Infliximab Lung Etanercept and Adalimumab Cyclosporine and Prednisolone Skin and nail Leflunomide Prednisolone, Infliximab, Adalimumab Artificial Joint Adalimumab Not enough information Bone, Joint and muscle Not enough information Leflunomide and Prednisolone Blood infection Not enough information Hydroxychloroquine and Prednisolone GIT Adalimumab Cyclosporine, Prednisolone, Infliximab. Nervous system Not enough information Not enough information TB Not enough information Not enough information Urinary tract infection Leflunomide and Infliximab Cyclosporine, Prednisolone and IM Gold Viral Infection Etanercept, Methotrexate Cyclosporine and Prednisolone and Hydroxychloroquine

These findings need to be validated in independent studies. However, they do provide new insights into the likely differential effects of csDMARDs and bDMARDs on diverse infections. They confirm known and suspected risks associated with corticosteroid use and potent TNF inhibitors, such as infliximab. These different effects were observed across multiple anatomical systems. Some protective effects were observed which, if confirmed in further studies, might allow an opportunity for selection of one agent over another, particularly where a high risk for infections is known to apply, based on comorbidities or previous infection history. Thus, it may be possible, taking all factors into consideration, to make an informed choice, rather than one more arbitrary, thereby enhancing patient safety without compromising clinical outcomes. This application of a personalised medicine approach has the potential to reduce the morbidity and mortality associated with non-serious and very importantly with serious infections of diverse aetiologies.

202

References: [1] M. Feely, “New and emerging therapies for the treatment of rheumatoid arthritis,” OARRR, p. 35, Jul. 2010, doi: 10.2147/OARRR.S6868.

[2] “New and emerging therapies for the treatment of rheumatoid arthritis.” https://www- ncbi-nlm-nih-gov.ezproxy.csu.edu.au/pmc/articles/PMC5074771/ (accessed Apr. 20, 2019).

[3] F. C. Breedveld and B. Combe, “Understanding emerging treatment paradigms in rheumatoid arthritis,” Arthritis Research & Therapy, vol. 13, no. 1, p. S3, May 2011, doi: 10.1186/1478-6354-13-S1-S3.

[4] Hootman J, Bolen J, Helmick C, Langmaid G, “Prevalence of Doctor-Diagnosed Arthritis and Arthritis-Attributable Activity Limitation --- United States, 2003--2005,” 2006. https://www.cdc.gov/mmwr/preview/mmwrhtml/mm5540a2.htm (accessed Apr. 08, 2020).

[5] N. B. Klarenbeek et al., “Clinical synovitis in a particular joint is associated with progression of erosions and joint space narrowing in that same joint, but not in patients initially treated with infliximab,” Ann. Rheum. Dis., vol. 69, no. 12, pp. 2107–2113, Dec. 2010, doi: 10.1136/ard.2010.131201.

[6] P. E. Lipsky et al., “Infliximab and Methotrexate in the Treatment of Rheumatoid Arthritis,” New England Journal of Medicine, vol. 343, no. 22, pp. 1594–1602, Nov. 2000, doi: 10.1056/NEJM200011303432202.

[7] V. Chiurchiù and M. Maccarrone, “Chronic inflammatory disorders and their redox control: from molecular mechanisms to therapeutic opportunities,” Antioxid. Redox Signal., vol. 15, no. 9, pp. 2605–2641, Nov. 2011, doi: 10.1089/ars.2010.3547.

[8] R. A. Myllykangas-Luosujirvi, K. Aho, and H. A. Isomiiki, “Mortality in Rheumatoid Arthritis,” p. 10.

203

[9] D. De Cock and K. Hyrich, “Malignancy and rheumatoid arthritis: Epidemiology, risk factors and management,” Best Practice & Research Clinical Rheumatology, vol. 32, no. 6, pp. 869–886, Dec. 2018, doi: 10.1016/j.berh.2019.03.011.

[10] D. S. Dalal et al., “Efficacy and safety of biological agents in the older rheumatoid arthritis patients compared to Young: A systematic review and meta-analysis,” Seminars in Arthritis and Rheumatism, vol. 48, no. 5, pp. 799–807, Apr. 2019, doi: 10.1016/j.semarthrit.2018.07.009.

[11] K. Thomas and D. Vassilopoulos, “Individual Drugs in Rheumatology and the Risk of Infection,” in The Microbiome in Rheumatic Diseases and Infection, G. Ragab, T. P. Atkinson, and M. L. Stoll, Eds. Cham: Springer International Publishing, 2018, pp. 445– 464.

[12] K. Grønning, S. Lim, and O. Bratås, “Health status and self-management in patients with inflammatory arthritis-A five-year follow-up study after nurse-led patient education,” Nurs Open, vol. 7, no. 1, pp. 326–333, Jan. 2020, doi: 10.1002/nop2.394.

[13] N. Jung, J.-L. Bueb, F. Tolle, and S. Bréchard, “Regulation of neutrophil pro- inflammatory functions sheds new light on the pathogenesis of rheumatoid arthritis,” Biochemical Pharmacology, vol. 165, pp. 170–180, Jul. 2019, doi: 10.1016/j.bcp.2019.03.010.

[14] National Health and Medical Research Council (Australia) and Royal Australian College of General Practitioners, Clinical guideline for the diagnosis and management of early rheumatoid arthritis. South Melbourne, Vic.: Royal Australian College of General Practitioners, 2009.

[15] A. Mian, F. Ibrahim, and D. L. Scott, “A systematic review of guidelines for managing rheumatoid arthritis,” BMC Rheumatology, vol. 3, no. 1, p. 42, Oct. 2019, doi: 10.1186/s41927-019-0090-7.

[16] J. S. Smolen et al., “EULAR recommendations for the management of rheumatoid arthritis with synthetic and biological disease-modifying antirheumatic drugs: 2016 update,” Ann Rheum Dis, vol. 76, no. 6, pp. 960–977, Jun. 2017, doi: 10.1136/annrheumdis-2016-210715. 204

[17] J. A. Singh et al., “2015 American College of Rheumatology Guideline for the Treatment of Rheumatoid Arthritis,” Arthritis & Rheumatology, vol. 68, no. 1, pp. 1–26, 2016, doi: 10.1002/art.39480.

[18] A. J. MacGregor et al., “Characterizing the quantitative genetic contribution to rheumatoid arthritis using data from twins,” Arthritis Rheum., vol. 43, no. 1, pp. 30–37, Jan. 2000, doi: 10.1002/1529-0131(200001)43:1<30::AID-ANR5>3.0.CO;2-B.

[19] D. F. Liew, “Therapeutic Guidelines: Rheumatology. Version 3.,” doi: 10.18773/austprescr.2017.059.

[20] J. Starkweather and A. K. Moske, “Multinomial logistic regression,” Consulted page at September 10th: http://www. unt. edu/rss/class/Jon/Benchmarks/MLR_JDS_Aug2011. pdf, vol. 29, pp. 2825–2830, 2011.

[21] A. Satorra and P. M. Bentler, “A scaled difference chi-square test statistic for moment structure analysis,” Psychometrika, vol. 66, no. 4, pp. 507–514, Dec. 2001, doi: 10.1007/BF02296192.

[22] E. McCrum-Gardner, “Which is the correct statistical test to use?,” British Journal of Oral and Maxillofacial Surgery, vol. 46, no. 1, pp. 38–41, Jan. 2008, doi: 10.1016/j.bjoms.2007.09.002.

205

CHAPTER 5

Serious Infections in Rheumatoid Arthritis

206

Abstract

Objective: The main goal of this study is to evaluate self-reported infections in patients suffering from rheumatoid arthritis and to determine the level of impact from different csDMARDs and bDMARDs as potential risk factors.

Method: ARAD reports were collected from 2001 to 2014 and cleaned by deleting all duplicated answers, single answers and faulty/incomplete patients reports. Overall 27,709 visits from 3110 patients during 2001 to 2014 were collected. Based on our definition for serious infection, patients’ reports were searched for evidence of hospitalisation or IV infusion for infection.

Results: Out of 27,709 visits during 2001 to 2014, 811 patients had reported serious infection, a prevalence of almost 3 %. Also, among all patient who took bDMARDs, adalimumab and etanercept were the most common medications with association with serious infection. Other factors such as age and gender, alcohol consumption, biologics, prednisolone, diseases such as diabetes, kidney disease, liver disease, heart attack and sometimes previous coronary artery bypass grafting (CABG) were all shown to have contribution in the development of SIs. These risk factors have been used to generate an equation which assists in predicting the development of SIs due to a range of risk factors.

Conclusion: There is clearly an increasing trend for serious infection (SI) among patients who were treated with biologics.

207

1. Introduction Rheumatoid arthritis (RA) is a chronic multisystem, immuno-inflammatory disease, the cardinal features of which are joint deformity and damage in most, but not all cases. Destructive polyarthritis is common and can be severely disabling and diminish quality of life. The major clinical manifestation of RA is persistent and progressive synovitis, mostly in the peripheral joints, leading to resorption of cartilage and subchondral bone. Joint disease in RA is usually symmetrical, polyarticular and when destructive the joint damage is usually irreversible [1]. The prevalence of the disease increases with increasing age, but it may happen at any age, with the peak incidence between the fourth and sixth decades. RA may be diagnosed as early as 3 months from onset up to 2 years when the disease is established. Depending on the diagnosis the prevalence of RA is up to 0.5–1% of the world’s population. The female sex is usually up to three times more susceptible to the disease than the male sex [1][2].

RA can cause chronic pain and joint destruction, premature mortality, and elevated risk of disability, with high costs for victims and for society. It is a heterogeneous disease comprising several subsets of patients with variations in pathogenesis, but it usually stems from due to a sustained specific immune response directed against unknown self-antigens. The characteristic of this autoimmune reaction is a cellular infiltration and synovial inflammation resulting in tissue damage. The major pathophysiological events in RA include mononuclear cell infiltration in the sub intimal layer, hyperplastic changes in synovial lining cells and formation of a destructive type of synovial tissue known as pannus that invades the interface between cartilage and bone. Chronic synovitis can progress to the destruction of adjacent bone and cartilage, leading to joint deformity and disability [1] [3].

All aspects of RA treatment have changed in the past 25 years. The pathogenic basis of RA also plays a role in its treatment. As early onset of structural damages is usual in RA, late treatment can cause more than 50% disability in this disease. Therefore, early treatment of the disease is an important objective [4]. Treatment in RA usually has three main goals including elimination of pain, prevention of joint damage and improvement of joint function. Usually treatment plans change depending on the disease activity, severity of symptoms, signs and prognosis [5].

208

Disease-modifying anti-rheumatic drugs (DMARDs) are medicines which are generally used to control RA. They interfere with the immune system to suppress the overactive immune system in RA, decreasing inflammation and progression of the disease process. These drugs are categorized as biologics and non-biologics, where non-biologics are indirect and nonspecific immune suppressants and biologics interfere with a specific aspect of the immune system (Chiurchiù & Maccarrone, 2011)[6]. [7].

Non-biologic medicines suppress the immune system indirectly, while biologics suppress immune system directly by interfering with a specific mediator. For example, methotrexate acts as a folic acid analogue to inhibit different pathways in the immune system, but the inhibition of tumor necrosis factor (TNF) ) by biologic DMARDs is achieved with a monoclonal antibody such as infliximab (Remicade), adalimumab (Humira), certolizumab pegol (Cimzia), and golimumab (Simponi), or with a circulating receptor fusion protein such as etanercept (Etanercept or Brenzys)[8] [6][7].

All the currently available conventional synthetic DMARDs are associated with limited efficacy and many of them cause important side effects. Due to these side effects, the majority of patients have to stop non biologic DMARDs within 1-2 years. Aletaha and Smolen found that, among 593 patients with RA, comprising 1319 courses of DMARD therapy over 2378 patient-years of treatment, retention rates were less than 24 months in most cases and treatment courses were terminated mostly due to adverse effects and toxicity (42%) and sometimes lack of sufficient efficacy (37%) [6][7].

1.1. Aims The aims of this study were to:

• determine the frequency of serious infections (SIs) amongst ARAD participants who were mostly taking bDMARDs;

• Investigate for apparent differences, if present, amongst patients taking bDMARDs and those taking csDMARDs;

• Identify and evaluate potential clinical risk factors for infections e.g. age, comorbidities, use of corticosteroids, synthetic DMARDs and biologic DMARDs, and assess ARAD data to investigate possible roles of behavioural, environmental and genetic risk factors in the development of serious infections in RA patients. 209

1.2. Hypothesis

The aims are based on the following hypotheses: • Infections and specifically serious infections are common in RA and likely modulated by medication use;

• It is possible to predict the approximate risk of infection based on risk factors that a patient possesses; and

• Type of medication, duration of medication usage and dosage of the medication can all impact the frequency of infection.

In order to assess these points, the ARAD data has been statistically analysed for the rate of serious infections (SIs). This study includes the rate, as well as the type of infection in more than 3000 RA patients.

The following points will be discussed: (i) Descriptive analysis of infection in RA including the comparison of the SIs between patients on csDMARDs and bDMARDs;

(ii) Comparison of the frequency of use of different biologics in serious infection;

(iii) Assessing the status of prednisolone and methotrexate in producing serious infection;

(iv) Discussing the different potential risk factors for serious infection in more details; and

(v) Predicting risk of serious infection based on the impacts from different risk factors.

2. Methods

2.1. Data Collection The data were collected from the ARAD, in which a cohort of 3569 RA patients (960 males and 2609 females) who had completed related questionnaires 28176 times (during 2001-2014) were investigated for the development of SIs associated with a range of risk factors. Among the 3569 patients, 459 patients were eliminated because they had filled out the questionnaire only once. We were therefore left with 3110 patients. After deducing eight duplications, at the end we came up to 27709 visits and, amongst these visits, 811 were classified as having developed

210

SIs and these were used to calculate SIs/100 patient-years. Using the ARAD database, RA patients with serious infections were identified, and their age, sex, type of infection, and type of medications used were all extracted for further statistical analysis.

2.2. Statistical Analysis This data was subjected to a series of descriptive and inferential statistical analyses, including the determination of summaries of the frequencies, ratios, proportions, incidence rates, and the rate of possible complications, as well as age and gender. In addition, a series of descriptions, including comparison of central location in frequencies and dispersion in both intervention and control variables, were undertaken. To facilitate the discussion around possible clinical risk factors for the development of serious infections in RA patients, relative risks and odds ratios were also calculated. In appropriate circumstances, these calculations can examine the association between variables and side effects.

We calculated patient-years of treatment for diverse treatment categories by dividing the number of SIs associated with each therapeutic by the sum of the lengths of time or cumulative exposure time during which each patient was taking that medicine. This enables assessment of the association between different bDMARDs and SIs while considering the number of patients and the duration of the period that they were taking these therapies. Furthermore, categorical statistical analysis using Chi-squared test between medications such as bDMARDs and a range of potential risk factors were calculated. The results of the Chi-squared tests helped determine the true association between intervention factors and SIs in this research. In this study we reviewed the following risk factors: gender, alcohol intake, prednisolone, diabetes (non-insulin- dependent diabetes mellitus (T2DM) as well as insulin dependent diabetes mellitus (T1DM), lung, kidney, and liver disease, heart attack, coronary artery bypass grafting (CABG), and stenting status. From these chi-squared values, the relevant p-values were calculated.

A generalized linear mixed model was applied to the data from the 28679 visits by the 3110 patients who visited more than once. Based on that, a model was constructed which expressed the natural logarithm of the odds for SI in terms of five predictor variables: age (in years), sex (M or F), alcohol, biologics (currently on any biologic? yes/no) and prednisolone (currently taking prednisolone? yes/no).

211

A histogram was created to demonstrate the age (in years) of each participant on his/her first visit. A series of statistical analyses was carried out to determine the number and the percentage of male/female patients who had self-reported their use of various biologics as (i) Never taken or Don't know, (ii) Currently taking, or (iii) Stopped taking.

Wherever DMARDs are discussed in this study, DMARDs is divided into csDMARDs and bDMARDs. csDMARDs or conventional synthetic DMARDs include: 1- Methotrexate – oral or parenteral, 2- Hydroxychloroquine (Hydroxychloroquine), 3- Sulphasalazine, 4- Leflunomide, 5- Azathioprine, 6- Cyclosporine. bDMARDs or Biologics or biological DMARDs include: 1- Humira/Adalimumab, 2- Etanercept/Etanercept, 3- Kineret/Anakinra, 4- Remicade/Infliximab, 5- Mabthera/Rituximab, 6- Orencia/Abatacept, 7- Actemra/Tocilizumab, 8- Simponi/Golimumab, 9- Cimzia/Certolizumab Pegol. Prednisolone, IM gold and Penicillamine do not belong to any group and are studied separately.

3.0 Results and discussion Amongst the 27709 visits made by RA patients who had taken part in the study and had filled in the questionnaire more than once, 811 visits were confirmed to relate to patients who had developed serious infections, a prevalence of 2.92 %. Our data were examined for the impact of several predictor variables: medications, age, gender and length of time in program. Some combinations of these variables were also considered. This is described below.

3.1. Analysis of Rheumatoid Arthritis (RA) and Serious Infections (SIs) in Australia Analysis of the association between the frequency of taking different anti RA medications and development of serious infection. Among all different anti RA medications, it seems that taking most of the medication between RA and RA with SIs are same, except for Adalimumab (Arrow in the table) which is the most frequent medication in Sis followed by Etanercept (figure 5.1).

212

Duration in years Figure 5.1 Differences in biologics between Rheumatoid Arthritis and Rheumatoid Arthritis with Serious infection. Adalimumab is indicated by arrow

Figure 5.2 shows the absolute number of self-reported SIs amongst recipients of prednisolone. The three coloured lines depict the categories of prednisolone usage, notably current, previous and never used. Here, we are comparing frequency of taking prednisolone in patients with serious infection compare to RA group. As it is apparent in the serious infection group, ‘currently taking prednisolone’ has the highest rate, followed by’ stopped taking’. The frequency of ‘Never taking’ prednisolone is almost same in both group and its graph has become flat (Figure 5.2). This indicates a significant role for prednisolone in serious infection.

213

Duration in years

Figure 5.2 Serious infection and prednisolone usage

The situation changes with methotrexate. Figure 5.3 is comparing the frequency of taking methotrexate in patients with serious infection compare to RA group. As it is apparent in serious infection group never taken methotrexate has the highest rate followed by patients who were taking Methoteraxate and recently stopped taking. The frequency of “currently taking” Methotrexate is almost same in both groups and its graph has become almost flat (Figure 5.3). This indicates a small role for methotrexate in serious infection (Figure 5).

Figure 5.3 Methotrexate status in Serious infection compare to RA

214

3.2. Age and gender In the database, there were 3569 participants (960 males and 2609 females) who completed a relevant questionnaire 28168 times. A boxplot of the ages at first visit for the two sexes is shown in Figure 5.4.

Figure 5.4 Boxplot showing the ages of patients at their first visit, broken down by gender

Sex1= Male Sex 2= Female Figure 5. 5 Sex distribution in RA, Sample size: 3111

215

The mean and standard deviation of the ages of the males were 59.3 and 11.7 years respectively and the mean and standard deviation of the ages of the females were 56.1 and 12.9 years respectively. The differences were statistically significant (P value <0.0001). A boxplot of the ages of the two genders appears in Figure 5.6. As can be seen in the boxplot, the age distribution was very similar in the two sexes. However, disease occurs slightly earlier and with a wider age dispersion among females.

Figure 5.6. Boxplot of the ages (in years) at the time of entry to the registry, broken down by the two sexes

According to Figure 5.7, disease onset in most patients was reported between the ages of 50 and 60 years.

216

Figure 5.7 Histogram of frequency of age groups for RA

3.3. Length of time in the program A histogram showing the number of months during which participants reported is shown in Figure 5.8. The longest report was for almost 150 months.

Sex (1= Male, 2= Female) Figure 5.8. Boxplot of participation Time in the ARAD Program, broken down by gender

217

3.4. Time in the program as a function of Gender The mean and standard deviation of the times in the program for males were 52.3 and 32.3 months, respectively, and the mean and standard deviation for the duration of participation for ages of females were 53.9 and 38.9 months, respectively. A boxplot of participation times in the program for the two genders appears in Figure 5.8. According to this boxplot, with almost similar standard deviation between both sexes, on average, in this study, females participated for a longer duration.

3.5. Distribution of age groups In this section, the mean and standard deviation in respect to age for csDMARDs with bDMARDs recipients and among three main participant groups of bDMARDs are compared. Among three main participant groups of bDMARDs 64.5 and 14.0 for bDMARDs treated participants, 59.1 and 17.4 for biologic 2, and finally 59.35 and 15.35 for biologic 3. A boxplot of the age upon entry and the biologic status at the last visit appears in Figure 5.9. According to this boxplot there are small differences in the age distribution among patients who take biologics. These differences can potentially increase risk of type 1 error in this study.

Figure 5.9 Boxplot of distribution of ages for the main three biologic groups 218

3.6. Incidence and rate of SIs. A cohort of 3569 RA patients (960 males and 2609 females), who had completed related questionnaires 28176 times (during 2001-2014), were investigated for the development of SIs associated with previously identified risk factors. After eliminating eight duplicate visit records, the records of 28168 visits remained. The data were studied from two perspectives; the incidence of SIs and the rate of SIs.

Visits were classified as indicating the presence of an SI or not. An investigation of the incidence of SIs examines each individual visit and so, potentially, it looked at all 28168 visits (for the 3569 patients). However, to look at the rate of SIs per 100 patient years requires a count of the number of SIs over a period that is not instantaneous, so records from at least two visits by a patient are needed to measure the passage of time. As 459 patients had completed the questionnaire only once, they were eliminated from the analysis of the rate of SIs. This left 3110 patients (2275 females and 835 males) and the records of 27709 visits. In these records, 811 visits were identified where the patient had an SI. (Figure 5.10)

Sex1= Male Sex 2= Female

Figure 5. 10 Sex distribution in RA with serious infection

219

3.7. Incidence of SIs According to the registries from other countries, such as Britain, the rate of SIs in bDMARDs recipients is highest in the first year. Therefore, in ARAD we are separating the first year’s visit to assess bDMARDs in the first year. There were 9087 visits, overall. Each visit was classified based on the presence of SI and descriptions for that infection including infected organ, type of DMARD and statistical significance of the difference. The following table shows the various combinations of SI/no SI with ever/never having taken bDMARDs (Table 5.1).

Table 5.1 Relationship between bDMARDs and Serious infections

Serious Infection Biologics Yes No Total Yes 191 98 289 No 6308 2490 8798 Total 6499 2588 9087

A Pearson’s chi-squared test of independence was performed to assess the data. The null hypothesis that the two variables are independent is rejected at the 5% level of significance (X- squared = 4.0495, df = 1, p-value = 0.04418). There is slight evidence of an association between whether a visit is classified as “SI” and whether the patient has ever had a biologic.

In addition, the statistics for taking bDMARDs based on the gender of the patient has been calculated in table 5.2. Based on the large Chi square result and insignificant p-value (0.38) at level of 0.05, the frequency of taking bDMARDs is similar in between male and female sex (Table 5.2).

Table 5.2 Biologic status of the patients in the study Biologic status Never Taken Currently Stopped Total or Don't Know Using taking Number and % of 521 2160 429 3110 patients (16.75%) (69.45%) (13.8%) Male 151 565 119 835 Gender Female 370 1595 310 2275 Pearson's Chi-squared test data: X-squared = 1.9067, df = 2, p-value = 0.3854

220

3.7.1. Rates of serious infections

When considering rates of SI, at least two visits by a patient are required; among the 3569 patients, 459 patients were eliminated because they had completed a questionnaire only once. We were therefore left with 3110 patients (2275 females and 835 males). Amongst these visits, 811 RA patients with serious infections were identified. The SI patients did not differ appreciably from the overall group with regard to gender or distribution of ages at the first visit. In the table 5.5 the status of bDMARDs treatment and status of Sis has been demonstrated. In order to calculate 100 patient year, the number of patients was divided by duration in years multiplied by 100. The reason for using patient 100 year is to provide more accurate comparisons among groups when follow-up time (i.e., patient exposure time) is not the same in all groups (Table 5.3).

Table 5.3 Numbers of SIs, total elapsed time (in months) between first and final visits for the patients, and the corresponding rate of SIs per 100 patient-years

bDMARDs status Never Taken or Currently Using Stopped taking Don't Know No. of SIs 89 585 137 Total time (in months) 27600 115764 22821 Rate of SIs/100 patient years 3.870 6.06 7.20

The risk of SI in ARAD is 26 % or 811 SI out of 3110 patients with RA. In the following table we review some of the differences between these two populations (Table 5.4).

221

Table 5.4 Demographic characteristics of participants who self-reported an SI and participants who did not (Data collected from ARAD)

Variable Mean SD Median

Age in non-SI 61.48 12.31 63.00

Age in SI reporters 59.73 12.22 61.00

Number of Cigarettes / D in non-SI 14.89 13.23 15.00

Number of Cigarettes /D in SI 19.20 15.63 15.00 reporters

Duration of Smoking in non-SI 17.26 13.95 16.00

Duration of Smoking in SI reporters 21.41 12.28 20.00

ALCOHOL CONSUMPTION 1.32 0.47 1.00 >2 or <2 in non-SI

ALCOHOL CONSUMPTION 0.66 0.47 1.00 >2 or <2 in SI reporters

3.7.2. Predictor variables

The predictor variables considered possibly to influence the rate of SIs per patient were age (in years) at first visit, gender, alcohol (ever taken/never taken), bDMARDs use (ever/never taken any of Anakinra, Etanercept, Adalimumab, Infliximab, Certolizumab, Golimumab, Rituximab, Abatacept or Tocilizumab), prednisolone (ever/never taken), diabetes (ever/never had non-insulin- dependent diabetes mellitus (T2DM) or insulin dependent diabetes mellitus (T1DM)), lung disease (ever/never suffered), kidney disease (ever/never), liver disease (ever/never), heart attack (ever/never), coronary artery bypass grafting (CABG) (ever/never), and stenting status (ever/never).

One patient’s records had to be removed from this analysis, as there were contradictory answers to alcohol status over her various visits. This left 3109 patients (2274 females and 835 males). The overall SI rate among these 3109 patients was 5.8597 SIs per 100 patient years (PYs). For those who had ever taken a bDMARDs, the rate was 6.2610 SIs per 100 years. For those who had never taken bDMARDs, the rate was 3.8386 SIs per 100 PYs.

222

It is reasonable to assume that the number of SIs experienced by a patient in the time of observation might follow a Poisson distribution, with the rate of incidence varying from patient to patient as a function of the potential risk factors described in the previous paragraph. A Generalized Linear Model with a logarithmic link was used to model the number of SIs in terms of total time observed and the predictor variables listed in the previous paragraph. The number of SIs per patient was found to be slightly less variable than would be expected of data from a Poisson distribution, so appropriate adjustments were made to the analysis when testing for significance of terms in the model. The variable (ever had CABG) was found not to have a significant effect on the number of SIs, either alone or in conjunction with other variables, and was deleted from the model.

The following model was selected as providing the best fit to the data: log(rate) = 0.5562 + 0.003984 × (age at first visit) – 0.9394 (if male) + 0.6160 (if ever taken a biologic) + 0.3044 (if ever taken prednisolone) + 0.08837 (if ever had diabetes) + 0.2388 (if ever had lung disease) + 0.2926 (if ever had liver disease) + 0.5205 (if ever had heart attack) + 0.4225 (if ever had stenting) + 0.01517 × (age at first visit) (if male) + 0.2727 (if male and had ever had lung disease) – 0.5768 (if male, and had ever had liver disease) – 0.4945 (if ever taken biologic and ever had heart attack) + 0.5872 (if ever taken a biologic and ever had heart attack) + 0.5872 (if ever had diabetes and ever had a heart attack) – 1.0865 (if ever had diabetes and ever had stenting). (eq. 1)

The equation is an expression for the natural logarithm of the rate of SI per 100 PYs. It can be converted to an expression for the rate by taking antilogarithms[9]: Rate = e0.5562 × e0.003984 × (age at first visit) × e–0.9394 (if male) × e0.6160 (if ever taken a biologic) × e0.3044 (if ever taken prednisolone) × e0.08837 (if ever had diabetes) × e0.2388 (if ever had lung disease) × e0.2926 (if ever had liver disease) × e0.5205 (if ever had heart attack) × e0.4225 (if ever had stenting) × e0.01517 × (age at first visit) (if male) × e0.2727 (if male and had ever had lung disease) × e–0.5768 (if male, and had ever had liver disease) × e–0.4945 (if ever taken biologic and ever had heart attack) × e0.5872 (if ever had diabetes and ever had a heart attack) × e–1.0865 (if ever had diabetes and ever had stenting). (eq. 2)

Note that, except for the first two terms on the right-hand side, all other terms on the right-hand side appear only if one or two conditions are satisfied (simultaneously). For example, “-0.9394 (if male)” means that 0.9394 is subtracted from the right-hand side if the patient is male; if the patient is female, nothing is done. The expression “+ 0.2727 (if male and had ever had lung disease)” 223

means that 0.2727 is added to the right-hand side if the patient is male and had ever had lung disease; if the patient is female or had never had lung disease, nothing is done.

In order to predict the rate of serious infection in each patient we need to have an estimation of the risk factors for serious infection. In the following table we summarize the connection between different risk factors (predictors) and biologic medication.

3.8. Prediction of Serious infection A generalised linear model (GLM) was utilised to calculate the frequency of SI based on the estimated impact from each risk factor. For visits up to 12 months, there were 9087 visits overall. The following equation was used in the model: In (P/1-P) = 〆+B1X1+B2X2+B3X3+B4X4 where P stands for prevalence. log(p/(1-p)) = - 4.293 – 0.042 × initial age - 0.358 (if patient is male) – 3.284 (if patient has ever taken a biologic) + 0.263 (if patient drinks alcohol every day) + 2.626 (if patient has ever taken prednisone) – 0.909 (if patient has ever had diabetes) + 4.877 (if patient has ever had lung disease) + 9.341 (if patient has ever had kidney disease) + 2.885 (if patient has ever had liver disease) – 6.317 (if patient has ever had heart attack) – 2.673 (if patient has ever had angioplasty) + 0.039 × initial age (if patient has ever had a bDMARDs) – 0.089 × initial age (if patient has ever had lung disease) – 0.156 × initial age (if patient has ever had kidney disease) + 0.118 × initial age (if patient has ever had a heart attack) + 2.083 (if patient is male and has ever had lung disease) + 2.091 (if patient is male and has ever had kidney disease) + 3.465 (if patient is male and has ever had a heart attack) – 3.075 (if patient is male and has ever had angioplasty) – 1.494 (if patient has ever had a biologic and has ever had prednisone) – 0.932 (if patient has ever had a biologic and has ever had lung disease) + 1.076 (if patient has ever had a biologic and has ever had kidney disease) + 2.641 (if patient has ever had a biologic and has ever had liver disease) – 4.076 (if patient has ever had a biologic and has ever had a heart attack) + 6.305 (if patient has ever had a bDMARD and has ever had angioplasty) + 1.368 (if patient drinks alcohol every day and has ever had lung disease) – 1.542 (if patient has ever had prednisone and has ever had kidney disease) – 5.710 (if patient has ever had prednisone and has ever had liver disease) – 4.055 (if patient has ever had diabetes and has ever had a heart attack) + 1.605 (if patient has ever had lung disease and has ever had kidney disease) – 5.113 (if patient has ever had lung disease and has ever had a heart attack) + 5.009 (if patient has ever had liver disease and has ever had a heart attack).

224

The model does not contain anything involving “if the patient has ever had a graft”, because no term involving this predictor variable was found to be statistically significant. For example, a male patient who was 62 years on his first visit, drinks alcohol every day, has taken a bDMARD but has never taken prednisolone, has had lung disease but none of the other diseases considered and has had a heart attack, but has not had angioplasty.

Then we have: log(p/(1-p)) = - 4.293 – 0.042 × 62 - 0.358 – 3.284 + 0.263 + 4.877 – 6.317 + 0.039 × 62 – 0.089 × 62 + 0.118 × 62 + 2.083 + 3.465 – 0.932 – 4.076 + 1.368 – 5.113 = -10.705, and so, odds = p/(1-p) = e-10.705 = 2.24 × 10-5. P=e/1+e= 2.24 × 10-5/1+2.24 × 10-5

Here is a table that shows the actual SI status on each visit, and the predicted status (odds less than 1 implies “Not SI”, odds greater than 1 implies “SI”).

Predicted

Yes No

SI Yes 3 286 289 No 14 8784 8798 17 9070 9087

The model was very reliable in predicting a “non-SI” when the patient did not have an SI but was virtually useless in predicting an SI when the patient had an SI.

This is a plot of the predicted probability that a visit will be an “SI” vs the actual result of the visit. Probabilities less than 0.5 correspond to odds of less than 1, while probabilities greater than 1 correspond to odds of more than 1. We can see that the probabilities cover a very wide range even though we would like them to be very close to 0 or 1 (for “non-SI and “SI” respectively). For visits after 12 months, the following table of SI/not SI vs biologic ever/never can be constructed:

bDMARD

SI yes no

yes 480 69 549 no 15105 3426 18531 15585 3495 19080

225

There is a significant association between SI and bDMARD use (X-squared = 12.095, df = 1, p- value = 0.0005055). Of the people who have ever taken a bDMARD, the proportion whose visit is associated with an SI is 480/15585 = 3.08%, whereas, of the people who have never taken a biologic, the proportion whose visit is associated with an SI is 69/3495 = 1.83%. There is a greater proportion of visits associated with an SI amongst those who have ever taken a biologic than amongst those who have never taken a biologic.

When the full statistical analysis involving all predictor variables was performed, the best model was found to be log(p/(1-p)) = - 7.968 + 0.025 × initial age - 2.043 (if patient is male) + 0.579 (if patient has ever taken a biologic) – 1.062 (if patient drinks alcohol every day) + 0.719 (if patient has ever taken prednisone) – 2.747 (if patient has ever had diabetes) + 0.978 (if patient has ever had lung disease) + 1.772 (if patient has ever had kidney disease) - 0.258 (if patient has ever had liver disease) + 1.711 (if patient has ever had a heart attack) – 0.813 (if patient has ever had a graft) + 3.692 (if patient has ever had angioplasty) + 0.033 × initial age (if patient is male) + 0.043 × initial age (if patient has ever had diabetes) – 0.060 × initial age (if patient has ever had angioplasty) – 1.987 (if patient is male and has ever had kidney disease) -2.653 (if patient is male and has ever had a graft) + 0.893 (if patient is male and has ever had angioplasty) + 1.057 (if patient has ever had a biologic and drinks alcohol every day) + 1.004 (if patient has ever had a biologic and has ever had diabetes) - 1.850 (if patient has ever had a biologic and has ever had a heart attack) – 0.651 (if patient drinks alcohol every day and has ever had lung disease) – 1.432 (if patient drinks alcohol every day and has ever had kidney disease) + 1.657 (if patient drinks alcohol every day and has ever had liver disease) + 0.937 (if patient drinks alcohol every day and has ever had a heart attack) + 2.691 (if patient drinks alcohol every day and has ever had a graft) – 1.383 (if patient drinks alcohol every day and has ever had angioplasty) + 3.464 (if patient has ever had prednisone and has ever had a graft) – 1.154 (if patient has ever had diabetes and has ever had lung disease) + 1.299 (if patient has ever had diabetes and has ever had liver disease) – 0.880 (if patient has ever had diabetes and has ever had a heart attack) + 0.695 (if patient has ever had lung disease and has ever had a heart attack) - 1.806 (if patient has ever had lung disease and has ever had a graft) + 0.983 (if patient has ever had a heart attack and has ever had angioplasty).

226

The following is the table of predictions made by the model for each visit:

Predicted

Yes No

SI Yes 1 548 549 No 2 18529 18531

Total 3 19077 19080

The model predicted virtually every visit to be a “non-SI”, whether an SI was present. Fig. 5.12 displays a plot of predicted probabilities for a given visit being an “SI” vs the actual result of the visit. It may seem that the models for predicting that a visit is associated with an SI, due to small amount of available data are not of much value, but as SI is potentially fatal and has dangerous consequences, it is worthwhile to try and predict it.

4. Discussion The model for those 2966 patients who had at least two visits in the first 12 months, for the rate (in SIs per 100 PYs) is: log(rate) = 0.1583 + 0.0244 × initial age - 0.5759 (if patient is male) + 0.2543 (if has ever taken a biologic) + (1.8543 - 0.02356 × initial age) (if has ever taken prednisone) + 0.7903 (if male and has ever taken a biologic).

We could take antilog and rewrite this equation as: rate = e 0.1583 × e 0.0244 × initial age × e- 0.5759 (if patient is male) × e 0.2543 (if has ever taken a biologic) × e (1.8543 - 0.02356 × initial age) (if has ever taken prednisone) + e 0.7903 (if male and has ever taken a biologic)

In this model, the rate increases with increasing age with or without taking prednisolone, it increases if the patient has ever taken a bDMARD, and it increases further if the patient is a male who has ever taken a biologic (Figure 5.11).

227

Figure 5. 11 A plot of the number of SIs predicted by the model for the first 12 months against the observed (actual) number of SIs

Guide Table for figure 5.11 Frequency of predicted serious infections No. of SIs 0 1 2 3 Total Frequency 2726 225 14 1 2966

In figure 5.12 the highest individual predicted number is about 0.14, whereas one person had three SIs. (That patient was female, aged 70 at the initial visit, and had taken biologics and prednisolone (Figure 5.12).

Figure 5.12 Rates of SIs per 100 patient years vs age at initial visit (in years) for males and females based on infections in the first 12 months after the initial visit

228

The effects of bDMARDs on serious infections may vary according to duration of bDMARD use. Therefore, predicted risk has been studied in the first year and then also after the first year. In Figure 5.13, the plot of the predicted number of SIs for the patients versus the observed numbers of SIs, after the first year of exposure to bDMARD medication has been demonstrated (Figure 5.13).

Figure 5.13 A plot of the number of SIs predicted by the model for more than 12 months exposure against the observed (actual) number of SIs.

Guide Table for figure 5.13 - Frequency of predicted serious infections.

No. of SIs 0 1 2 3 4 5 Total

Frequency 2283 346 74 14 2 1 2720

As can be seen in figure 5.14, the shape of the graph for the frequency of serious infection in females rises to only a modest extent with age, whereas in males it tends to increase sharply with age (Figure 5.13).

229

Figure 5.14 Rates of SIs per 100 patient years vs age at initial visit (in years) for males and females based on infections more than 12 months after the initial visit These rates of SIs during the first year and then after the first-year show that in the first 12 months of treatment with bDMARDs, the rates are about twice (10-12 per 100PYs) those observed for the whole period of exposure (approximately 6 per 100PYs). From those 2720 patients who had at least one visit after the first 12 months, the rate (in SIs per 100 patient years) is: log(rate) = 0.3939 + 0.0075 × initial age + (-1.2990 + 0.0228 × initial age) (if patient is male) + 0.4017 (if has ever taken a bDMARD) + 0.4619 (if has ever taken prednisone).

If we take antilog, we can rewrite this equation as: rate = e 0.3939 × e 0.0075 × initial age × e (-1.2990 + 0.0228 × initial age) (if patient is male) × e 0.4017 (if has ever taken a bDMARD) × e 0.4619 (if has ever taken prednisolone),

5. Chapter conclusion

The risk of serious infections among patients in ARAD during 2001 to 2014 is about 3%. In this chapter the level of impact from different biologics as potential risk factors for serious infection was studied. Data suggest that the rates of SI increases with increasing age with or without taking prednisolone, it increases if the patient has ever taken a biologic, and it increases further if the patient is a male who has ever taken a biologic. However, medication is not the only risk for SI and other risks, such as having chronic diseases or other factors which compromise immune system, can play a potential role and change the results, as well.

230

In compare to other registries, the rate of SI in Australia is higher. For instance, in south American countries according to a study which published in Aug. 2019 almost 2591 out of 13380 patient/years were taking bDMARDs and 1126 treated with csDMARDs. The SI IR was 30.54 (CI 27.18-34.30) for all bDMARDs and 5.15 (CI 3.36-7.89) for csDMARDs. In this study the aIRR between the two groups was 2.03 ([1.05, 3.9] p = 0.034) for the first 6 months of treatment but subsequently increased to 8.26 ([4.32, 15.76] p < 0.001). The SI IR for bDMARDs decreased over time in both registries, dropping from 36.59 (28.41-47.12) in 2012 to 7.27 (4.79-11.05) in 2016.[10]

In another study from British Society for Rheumatology Biologics Register - Rheumatoid Arthritis in total, 5289 subjects 19 431 patient-years had at least one SI. The baseline annual rate of first SI was 4.6% (95% CI: 4.5, 4.7), increasing to 14.1% (95% CI: 13.5, 14.8) following an index infection. Respiratory infections were the most frequent (44% of all events). Recurrent infections mirrored the organ class of the index infection. Sepsis, increasing age and polypharmacy were significant predictors of infection recurrence in a fully adjusted model. The system class of index infection was associated with the risk of a recurrent event; subjects who experienced sepsis had the highest risk of subsequent SI within 12 months, 19.7% (95% CI: 15.1, 25.7). [11]

Finally, in a study from five different registries (USA, Sweden, UK, Japan, and CORRONA International (multiple countries)) from 2000 to 2017 the results showed that age/sex-standardised rates of hospitalised infection were quite consistent across registries (range 1.14-1.62 per 100 patient-years). Higher and more consistent rates were observed when adding standardisation for HAQ score (registry range 1.86-2.18, trials rate 2.92) or restricting to a treatment initiation sub cohort followed for 18 months (registry range 0.99-2.84, trials rate 2.74).[12]

231

THESIS SUMMARY AND REMARKS

232

Summary of main findings In this model, the rate of SIs increases with increasing age, but at a greater rate (from a smaller start) if the patient is male. It also increases if the patient has ever taken a bDMARD and increases if the patient has ever taken prednisolone. Figure 5.12 shows a plot of the predicted individual rate of SIs versus the observed individual rate of SIs. Once again, we have a very large number of individuals who had no SIs is observed, so the model tends to predict a low rate[13]. A Generalized Linear Mixed Model was applied to the data from the 28168 visits by the 3569 patients in the study. A model was constructed which expressed the natural logarithm of the odds for SI in terms of x predictor variables. The number of SIs is greatest for Etanercept, Adalimumab and Abatacept respectively, but when the rate per 100 PYs is considered, this order changes to Adalimumab, Etanercept and Anakinra[14].

It may mean that Etanercept is potentially more dangerous than Etanercept, but in practice the dosage of medicine also plays a role. If a bDMARDs is prescribed for a shorter time at a lower dosage, it puts the patient potentially at lower risk and is safer. However, even one SI report in this situation is statistically more significant than the same report for another medicine with a longer duration and higher dosage. Also, doctors’ preferences in prescribing one medication and the selection of patients for these biologics all can contribute to the results.

In addition, for Anakinra there was only a limited number of patients available and this can reduce the reliability of the results[15]. For Anakinra, there was only a small number of patients exposed to the drug (n= ZZ), which calls into question the validity of the findings for this agent. Finally, medication is not the only risk for SI in most of the reports and other risks such as having chronic diseases or other factors which compromise immune system can play a potential role and change the results (Table 5.5)[16]. Around 83% of the patients have received a biologic. There are more women than men in each biologic status category, and the ratios of women to men are not significantly different across the three categories (Table 5.5)[16].

The results for alcohol consumption are likely relevant. Patients who were using a bDMARD treatment tended to consume alcohol more often (Table 5.5, P = QQ). Possible reasons for this observation include the use of alcohol to combat pain in those with more active disease, who are more likely to progress to bDMARDs.

233

A further possibility is that disease activity may be greater in alcohol users and so they may progress to bDMARDs more quickly or more often. Liver enzyme induction in higher consumers of alcohol may diminish responsiveness to csDMARDs and lead to more frequent progression to bDMARDs. Germane to this possible explanation is the tendency for anti-TB drugs, such as Rifampicin to induce liver enzymes and reduce responsiveness to corticosteroids, which in turn can lead to more active RA in patients who were previously stable[17].

There is also a statistically significant association between taking bDMARDs and taking prednisolone. Most of the patients who were taking bDMARDs were also taking prednisolone concurrently or had taken prednisolone previously. This is most likely explained on the basis of rheumatoid disease severity. RA patients not well controlled on multiple or sequential csDMARDs are more likely to have received adjunctive corticosteroids prior to qualifying for a bDMARD. Furthermore, depending on the clinical response to the bDMARD, they may or may not have been able to discontinue prednisolone. In any event, since this parameter includes previous usage, the associated use would have been captured[17].

Increased disease severity in RA is a risk factor for SIs but this risk cannot be easily disentangled from other risk factors. In other conditions, such as insulin dependent diabetes mellitus (T1DM) and non-insulin dependent diabetes mellitus (T2DM), lung disease and in patients with previous stent operations for coronary heart disease, the frequency with which bDMARDs were used is lower (Table 5.5). Any association between bDMARD status and the other factors is not statistically significant. One possible reason for this lack of association is the relatively small numbers of participants with these conditions in this RA cohort. However, the possibility of confounding by indication also exists, since prescribers may have avoided bDMARD usage in certain patients with concerning comorbidities. For example, the known propensity for diabetics to develop infections might have led to treating Rheumatologists exercising restraint in respect to prescribing bDMARDs in the context of T2DM and T1DM[17].

234

Table 5. 5- Relationship between bDMARD use and potential cofactors

Predictor Variables bDMARDs Never taken Taken at some time Total Never/Don’t know 135 785 920 Alcohol current 297 1501 1789 past 89 303 392 total 521 2589 3110 Never/Don’t know 180 308 488 Prednisolone current 242 99 521 past 1766 515 2589 total 2008 614 3110 Never/Don’t know 494 2431 2925 T1DM current 26 148 174 past 1 10 11 total 521 2589 3110 Never/Don’t know 475 2290 2765 T2DM current 41 256 299 past 46 299 46 total 521 2589 3110 Never/Don’t know 373 1792 2165 Lung Disease current 112 536 648 past 36 261 2589 total 521 2589 3110 Never/Don’t know 477 2365 2842 current 21 111 132 Kidney Disease past 23 113 136 total 521 2589 3110 Liver Disease Never/Don’t know 497 2386 2883 current 9 93 102 past 15 110 125 total 521 2589 3110 MI Never/Don’t know 464 2394 2858 current 16 56 72 past 41 139 180 total 521 2589 3110 CABG Never/Don’t know 499 2528 3027 current 8 14 22 past 14 47 61 total 521 2589 3110 C Stenting Never/Don’t know 480 2448 2928 current 20 69 89 past 21 72 93 total 3110 Pearson's Chi-squared test X-squared = 96.663, df = 2, p-value < 2.2e-16X-squared = 173.82, df = 2, p-value < 2.2e-16 A large number of people, who are taking biologics, are simultaneously taking prednisolone Insulin dependent Diabetes (T1DM) X-squared = 0.91075, df = 2, p-value = 0.6342None Insulin dependent Diabetes (T2DM) X-squared = 5.0186, df = 2, p-value = 0.08132Lung Disease X-squared = 5.051, df = 2, p-value = 0.08002 Kidney Disease X-squared = 0.071819, df = 2, p-value = 0.9647 Liver disease X-squared = 7.1119, df = 2, p-value = 0.02855 MI: Myocardial infarction X-squared = 6.7789, df = 2, p-value = 0.03373 Coronary artery bypass grafting (CABG) X-squared = 7.9029, df = 2, p-value = 0.01923

Page 235 of 577

Coronary stenting; X-squared = 4.6234, df = 2, p-value = 0.0

Concluding remarks

In the descriptive statistics, there is always a possibility that people who contribute to the research are not randomly selected or one sex contributes more than the other sex in answering the questions. In some studies, the number of patients who contribute to the study also may cause limitations. However, in this project, there was access to an adequate number of participants. Furthermore, there were no selection criteria applied by ARAD designers that might have led to preferential selection of some patients instead of others. Still, criteria such as the contribution of one gender more than the other or the ability of patients to answer questionnaires may have contributed to some bias[11].

RA in Australia is predominantly a female disease. The mean and median age for male patients is greater than the corresponding ages for female patients, which means that RA in females tends to begin at a younger age than in males. The prevalence of RA peaks in the 60s or seventh decade of life. The relevance of RA among those in the population younger than 60 is greater than in the population over 60. Many patients are diagnosed with RA when they are in their 50s[11].

Most of the patients with RA who participated in the ARA database were already taking or were about to begin bDMARDs, however, this is because the database was conceived as a bDMARD registry and non-bDMARD recipients were recruited later to supplement the cohort. The prevalence of serious infection in ARAD participants was 2.92 %. Rates were appreciably higher in the first year of treatment at around 12 per 100PYs in this study. Similar increased rates in the first year have been observed in other registries. Thereafter, rates were about half that in the first year at approximately 6 per 100PYs, which again accords with that reported in other registries. Males had higher rates of SI and this increased sharply with age, whereas in females there was a modest, but steady increase with age and even in advanced age, the rates in females did not approach those in males over 65 years. The major risk factors contributing to high rates of SI were advancing age, use of bDMARDs and use of Prednisolone. Comorbidities were not found to be major contributors to SIs[17].

As indicated previously, it is difficult to determine the rate of SIs in RA, since it is a function of disease severity as well as other factors. Within ARAD at least, the csDMARDs group was recruited post-hoc and likely consists of patients unmatched for disease activity, since they did

Page 236 of 577

not require bDMARDs. Moreover, there may have been different rates of corticosteroid use in csDMARDs users. Nevertheless, serious infections in this group were in the order of 3-4 per 100PYs, which is clearly lower than that in the first year of bDMARD therapy and lower than that in long term bDMARD recipients. These differences cannot be taken as proof of a substantial difference, but they are consistent with an increased propensity for SIs in bDMARD users[12].

Strengths of this analysis include the large size of the database and the randomness or lack of bias in recruitment, the opportunity to assess participants over a relatively long period of time (2001-2014) and the capture of events that might have been overlooked if reporting were not done by the participants, but rather by busy clinicians, whose reporting compliance may have been suboptimal.

Limitations include the unmatched nature of csDMARDs and bDMARD participants, thus confounding valid comparisons, the inability to verify self-reported infections of any severity (no input from family practitioners, no hospital records available, no microbiological confirmation of infections), but particularly SIs and the inability to capture SIs that resulted in death or severe disability that precluded further reporting.

Future studies both within and without ARAD have the potential to verify the findings reported here and to extend them. For example, the extent to which SIs increase in participants who transition from csDMARDs to bDMARDs within ARAD could be compared as there will have been an adequate period of observation during the pre-bDMARD era in these patients. Such a study would have the added benefit that the participants could be their own control, which in turn would provide greater rigor. Whether SI rates decline in bDMARD recipients after 5-10 years could also be examined. Deeper analysis of newly discovered clinical risk factors might also be possible. With linkage in time to biobanks, it may also become possible to examine the role of genetic and acquired immunity[17].

Page 237 of 577

References:

[1] D. H et al., “Clonal V alpha 12.1+ T cell expansions in the peripheral blood of rheumatoid arthritis patients.,” J Exp Med, vol. 177, no. 6, pp. 1623–1631, Jun. 1993, doi: 10.1084/jem.177.6.1623.

[2] D. F. McWilliams, P. D. W. Kiely, A. Young, and D. A. Walsh, “Baseline factors predicting change from the initial DMARD treatment during the first 2 years of rheumatoid arthritis: experience in the ERAN inception cohort,” BMC Musculoskelet Disord, vol. 14, p. 153, May 2013, doi: 10.1186/1471-2474-14-153.

[3] R. F. van Vollenhoven, “Sex differences in rheumatoid arthritis: more than meets the eye...,” BMC Medicine, vol. 7, no. 1, p. 12, Mar. 2009, doi: 10.1186/1741-7015-7-12.

[4] A. M. Abdel-Nasser, J. J. Rasker, and H. A. Valkenburg, “Epidemiological and clinical aspects relating to the variability of rheumatoid arthritis,” Semin. Arthritis Rheum., vol. 27, no. 2, pp. 123–140, Oct. 1997, doi: 10.1016/s0049-0172(97)80012-1.

[5] S. Cohen, “2012 challenges in rheumatoid arthritis care,” Rheumatology, vol. 51, no. suppl 6, pp. vi3–vi4, Dec. 2012, doi: 10.1093/rheumatology/kes284.

[6] A. I. Rutherford, S. Subesinghe, K. L. Hyrich, and J. B. Galloway, “Serious infection across biologic-treated patients with rheumatoid arthritis: results from the British Society for Rheumatology Biologics Register for Rheumatoid Arthritis,” Ann. Rheum. Dis., vol. 77, no. 6, pp. 905–910, 2018, doi: 10.1136/annrheumdis-2017-212825.

[7] J. A. Singh et al., “Risk of serious infection in biological treatment of patients with rheumatoid arthritis: a systematic review and meta-analysis,” Lancet, vol. 386, no. 9990, pp. 258–265, Jul. 2015, doi: 10.1016/S0140-6736(14)61704-9.

[8] J. S. Smolen, D. Aletaha, M. Koeller, M. H. Weisman, and P. Emery, “New therapies for treatment of rheumatoid arthritis,” The Lancet, vol. 370, no. 9602, pp. 1861–1874, Dec. 2007, doi: 10.1016/S0140-6736(07)60784-3.

[9] P. Cohen, “Protein kinases--the major drug targets of the twenty-first century?,” Nat Rev Drug Discov, vol. 1, no. 4, pp. 309–315, Apr. 2002, doi: 10.1038/nrd773.

Page 238 of 577

[10] R. Ranza et al., “Changing rate of serious infections in biologic-exposed rheumatoid arthritis patients. Data from South American registries BIOBADABRASIL and BIOBADASAR,” Clin Rheumatol, vol. 38, no. 8, pp. 2129–2139, Aug. 2019, doi: 10.1007/s10067-019-04516-2.

[11] S. Subesinghe, A. I. Rutherford, R. Byng-Maddick, K. Leanne Hyrich, and J. Benjamin Galloway, “Recurrent serious infections in patients with rheumatoid arthritis-results from the British Society for Rheumatology Biologics Register,” Rheumatology (Oxford), vol. 57, no. 4, pp. 651–655, 01 2018, doi: 10.1093/rheumatology/kex469.

[12] H. Yamanaka et al., “Infection rates in patients from five rheumatoid arthritis (RA) registries: contextualising an RA clinical trial programme,” RMD Open, vol. 3, no. 2, p. e000498, 2017, doi: 10.1136/rmdopen-2017-000498.

[13] M. Schoels, T. Kapral, T. Stamm, J. S. Smolen, and D. Aletaha, “Step-up combination versus switching of non-biological disease-modifying antirheumatic drugs in rheumatoid arthritis: results from a retrospective observational study,” Ann. Rheum. Dis., vol. 66, no. 8, pp. 1059–1065, Aug. 2007, doi: 10.1136/ard.2006.061820.

[14] J. B. Galloway et al., “The risk of serious infections in patients receiving anakinra for rheumatoid arthritis: results from the British Society for Rheumatology Biologics Register,” Rheumatology (Oxford), vol. 50, no. 7, pp. 1341–1342, Jul. 2011, doi: 10.1093/rheumatology/ker146.

[15] M. Lahiri and W. G. Dixon, “Risk of infection with biologic antirheumatic therapies in patients with rheumatoid arthritis,” Best Practice & Research Clinical Rheumatology, vol. 29, no. 2, pp. 290–305, Apr. 2015, doi: 10.1016/j.berh.2015.05.009.

[16] A. N. Lau et al., “Occurrence of Serious Infection in Patients with Rheumatoid Arthritis Treated with Biologics and Denosumab Observed in a Clinical Setting,” J. Rheumatol., vol. 45, no. 2, pp. 170–176, Feb. 2018, doi: 10.3899/jrheum.161270.

[17] J. B. Galloway et al., “Anti-TNF therapy is associated with an increased risk of serious infections in patients with rheumatoid arthritis especially in the first 6 months of treatment: updated results from the British Society for Rheumatology Biologics Register

Page 239 of 577

with special emphasis on risks in the elderly,” Rheumatology (Oxford), vol. 50, no. 1, pp. 124–131, Jan. 2011, doi: 10.1093/rheumatology/keq242.

Page 240 of 577

APPENDICES

Page 241 of 577

Appendices 241 Description of data in appendix ...... 243 Taking different medication levels ...... 243 Response levels ...... 243

Appendix A: Output of SAS for EENT Infection 244

Appendix B: OUTPUT of SAS for Lung Infection 268

Appendix C: Output of SAS for Nail and skin infection 301

Appendix D: Output of SAS for artificial joint infection 328

Appendix E: Output of SAS for bone muscle joint infection 351

Appendix F: Output of SAS for blood infection 385

Appendix G: Output of SAS for GIT Infection 411

Appendix H: Output of SAS for Nervous system infection 433

Appendix I: Output of SAS for TB infection 461

Appendix J: Output of SAS for Urinary Tract Infection 485

Appendix K: Output of SAS for viral infection 509

Appendix L: Ethical approval for the thesis 535

APPENDIX M: Sample of ARAD questionnaire 536

Page 242 of 577

Description of data in appendix

Taking different medication levels 1=Never taking 2=Currently taking 3=Stopped taking 4=Don’t know

Response levels 1=Mild 2=Moderate 3=Severe 4=Missing

Page 243 of 577

APPENDIX A: OUTPUT OF SAS FOR EENT INFECTION

Table A.1- Model information for EENT infection

Model Information Data Set WORK.IMPORT2 Response Variable InfEent InfEent Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table A.2- Observation status for EENT infection

Number of Observations Read 27711 Number of Observations Used 21506

Table A.3- response value for EENT infection

Response Profile Ordered Total Value InfEent Frequency Mild 1 1050 Moderate 2 1829 Severe 3 406 Missing 4 18221

Logits modelled use InfEent='4' as the reference category. Note: 6205 observations were deleted due to missing values for the response or explanatory variables.

Page 244 of 577

Table A.4- Backward Elimination Procedure for EENT infection Backward Elimination Procedure Class Level Information Class Value Design Variables Etanercept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Adalimumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Anakinra 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Infliximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Rituximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Abatacept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Tocilizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Golimumab 3 1 0 0 currently taking 0 1 0 never taking 0 0 1 Certolizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Folic Acid currently taking 1 0 never taking 0 1 Hydroxychloroquine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Sulphasalazine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Arava (Leflunomide) 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Page 245 of 577

Class Level Information Class Value Design Variables Azathioprine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Cyclosporin 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Prednisolone 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 IM Gold injection 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1 Penicillamine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Step 0. The following effects were entered: Intercept Etanercept Adalimumab Anakinra Infliximab Rituximab Abatacept Tocilizumab Golimumab Certolizumab Folic Acid Hydroxychloroquine Sulphasalazine Arava (Leflunomide) Azathioprine Cyclosporin Prednisolone IM Gold injection Penicillamine

Table A.5- Model Convergence status for EENT infection

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Page 246 of 577

Table A.6- Model Fit statistics for EENT infection

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24501.128 SC 24650.284 25745.398 -2 Log L 24620.355 24189.128

Page 247 of 577

Table A.7- Testing null hypothesis for EENT infection

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq 153 Likelihood Ratio 431.2272 <.0001 153 Score 463.0664 <.0001 153 Wald 419.5882 <.0001

Table A.8- Model Fit statistics for removing covariant step 1

Step 1. Effect Azathioprine is removed:

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table A.9- Model Fit statistics for removing covariant step 1

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24488.566 SC 24650.284 25661.051 -2 Log L 24620.355 24194.566

Page 248 of 577

Table A.10- Testing Null hypothesis after removing covariant step 1

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 425.7897 144 <.0001 Score 457.8861 144 <.0001 Wald 415.1007 144 <.0001

Table A.11- Residual removing covariant step 1 Residual Chi-Square Test Chi-Square DF Pr > ChiSq 5.0524 9 0.8297

Table A.12- Model Fit statistics for removing covariant step 2

Step 2. Effect Certolizumab is removed:

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Page 249 of 577

Table A.13- Model Fit statistics after removing covariant step 2 Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24476.658 SC 24650.284 25577.358 -2 Log L 24620.355 24200.658

Table A.14- Testing Null hypothesis after removing covariant step 2 Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 419.6974 135 <.0001 Score 450.9468 135 <.0001 Wald 408.7712 135 <.0001

Table A.15- Residual removing covariant step 2 Residual Chi-Square Test Chi-Square DF Pr > ChiSq 10.9161 18 0.8979

Page 250 of 577

Table A.16- Model Fit statistics for removing covariant step 3

Step 3. Effect Penicillamine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table A.17- Model Fit statistics after removing covariant step 3 Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.35 24473.673 SC 24650.28 25502.589 -2 Log L 24620.35 24215.673

Table A.18- Testing Null hypothesis after removing covariant step 3 Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 404.6821 126 <.0001 Score 440.0301 126 <.0001 Wald 401.9656 126 <.0001

Page 251 of 577

Table A.19- Residual removing covariant step 3 Residual Chi-Square Test Chi-Square DF Pr > ChiSq 22.0077 27 0.7370

Table A.20- Model Fit statistics for removing covariant step 4

Step 4. Effect IM Gold injection is removed:

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table A.21- Model Fit statistics after removing covariant step 4 Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24465.880 SC 24650.284 25423.011 -2 Log L 24620.355 24225.880

Page 252 of 577

Table A.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 394.4751 117 <.0001 Score 430.4313 117 <.0001 Wald 392.2553 117 <.0001

Table A.23- Residual removing covariant step 4 Residual Chi-Square Test Chi-Square DF Pr > ChiSq 31.2787 36 0.6926

Table A.24- Model Fit statistics for removing covariant step 5

Step 5. Effect Rituximab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Page 253 of 577

Table A.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24461.305 SC 24650.284 25346.650 -2 Log L 24620.355 24239.305

Table A.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 381.0509 108 <.0001 Score 415.9895 108 <.0001 Wald 378.4582 108 <.0001

Table A.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 44.9975 45 0.4721

Page 254 of 577

Table A.28- Model Fit statistics for removing covariant step 6

Step 6. Effect Golimumab is removed:

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table A.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24462.511 SC 24650.284 25300.000 -2 Log L 24620.355 24252.511

Table A.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 367.8443 102 <.0001 Score 403.4935 102 <.0001 Wald 366.8141 102 <.0001

Page 255 of 577

Table A.31- Residual removing covariant step 7

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 57.2672 51 0.2539

Table A.32- Summary of backward elimination in EENT

Note: No (additional) effects met the 0.05 significance level for removal from the model.

Summary of Backward Elimination Step Effect DF Number Wald Pr > ChiSq Variable Removed In Chi-Square Label

1 Azathioprine 9 17 4.9893 0.8352 Azathioprine 2 Certolizumab 9 16 5.4537 0.7931 Certolizumab 3 Penicillamine 9 15 7.1956 0.6168 Penicillamine

4 IM Gold injection 9 14 9.1915 0.4198 IM Gold injection 5 Rituximab 9 13 13.6536 0.1352 Rituximab 6 Golimumab 6 12 11.2165 0.0819 Golimumab

Page 256 of 577

Table A.33- Type 3 analysis of effects in EENT

Type 3 Analysis of Effects Wald Effect DF Chi-Square Pr > ChiSq Etanercept 9 52.1431 <.0001 Adalimumab 9 22.4139 0.0077 Anakinra 9 18.2690 0.0322 Infliximab 9 31.0160 0.0003 Abatacept 9 18.0153 0.0350 Tocilizumab 9 18.1032 0.0340 Folic Acid 3 9.4165 0.0242 Hydroxychloroquine 9 23.3663 0.0054 Sulphasalazine 9 26.7402 0.0015 Arava (Leflunomide) 9 17.5339 0.0410 Cyclosporin 9 47.3358 <.0001 Prednisolone 9 29.4764 0.0005

Table A.34- Analysis of maximum likelihood estimates in EENT

Analysis of Maximum Likelihood Estimates Wald Eent Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Intercept Mild 1 -3.4872 0.1190 859.2759 <.0001 Intercept Mod 1 -2.9786 0.0928 1031.0501 <.0001 Intercept Severe 1 -4.3609 0.1917 517.2695 <.0001 Abatacept 3 Mild 1 0.5166 0.1769 8.5260 0.0035 Abatacept 3 Mod 1 0.1147 0.1534 0.5587 0.4548 Abatacept 3 Severe 1 -0.4339 0.3573 1.4747 0.2246 Abatacept 4 Mild 1 0.2751 0.8455 0.1059 0.7449 Abatacept 4 Mod 1 -0.6022 0.6388 0.8887 0.3458 Abatacept 4 Severe 1 1.3251 1.0615 1.5582 0.2119 Abatacept currently Mild 1 0.3362 0.1582 4.5176 0.0335 taking

Page 257 of 577

Analysis of Maximum Likelihood Estimates Wald Eent Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Abatacept currently Mod 1 0.1491 0.1240 1.4460 0.2292 taking Abatacept currently Severe 1 -0.2016 0.2673 0.5686 0.4508 taking Abatacept never taking Mild 0 0 . . . Abatacept never taking Mod 0 0 . . . Abatacept never taking Severe 0 0 . . . Adalimumab 3 Mild 1 0.0104 0.0914 0.0129 0.9094 Adalimumab 3 Mod 1 0.1823 0.0686 7.0504 0.0079 Adalimumab 3 Severe 1 0.1418 0.1403 1.0222 0.3120 Adalimumab 4 Mild 1 -0.5402 0.6756 0.6394 0.4239 Adalimumab 4 Mod 1 - 0.4440 0.0000 0.9984 0.00090 Adalimumab 4 Severe 1 - 147.6 0.0048 0.9447 10.2462 Adalimumab currently Mild 1 0.2887 0.0941 9.4206 0.0021 taking Adalimumab currently Mod 1 0.1847 0.0737 6.2813 0.0122 taking Adalimumab currently Severe 1 -0.0798 0.1470 0.2946 0.5873 taking Adalimumab never taking Mild 0 0 . . . Adalimumab never taking Mod 0 0 . . . Adalimumab never taking Severe 0 0 . . . Anakinra 3 Mild 1 0.1448 0.2523 0.3295 0.5659 Anakinra 3 Mod 1 -0.0761 0.2191 0.1205 0.7285 Anakinra 3 Severe 1 0.4597 0.3413 1.8146 0.1780 Anakinra 4 Mild 1 -0.7187 0.6297 1.3026 0.2537 Anakinra 4 Mod 1 0.0275 0.4047 0.0046 0.9459 Anakinra 4 Severe 1 -0.4484 1.0513 0.1819 0.6697

Page 258 of 577

Analysis of Maximum Likelihood Estimates Wald Eent Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Anakinra currently Mild 1 - 512.1 0.0005 0.9814 taking 11.9697 Anakinra currently Mod 1 1.7999 0.4745 14.3901 0.0001 taking Anakinra currently Severe 1 - 809.5 0.0002 0.9879 taking 12.2422 Anakinra never taking Mild 0 0 . . . Anakinra never taking Mod 0 0 . . . Anakinra never taking Severe 0 0 . . . Arava (Leflunomide) 3 Mild 1 0.1098 0.0935 1.3804 0.2400 Arava (Leflunomide) 3 Mod 1 0.1933 0.0729 7.0343 0.0080 Arava (Leflunomide) 3 Severe 1 0.1484 0.1434 1.0712 0.3007 Arava (Leflunomide) 4 Mild 1 -0.2856 0.5428 0.2768 0.5988 Arava (Leflunomide) 4 Mod 1 0.4582 0.3091 2.1967 0.1383 Arava (Leflunomide) 4 Severe 1 -0.7134 1.0581 0.4546 0.5002 Arava (Leflunomide) currently Mild 1 0.2705 0.1060 6.5075 0.0107 taking Arava (Leflunomide) currently Mod 1 0.1492 0.0858 3.0250 0.0820 taking Arava (Leflunomide) currently Severe 1 0.00639 0.1726 0.0014 0.9705 taking Arava (Leflunomide) never taking Mild 0 0 . . . Arava (Leflunomide) never taking Mod 0 0 . . . Arava (Leflunomide) never taking Severe 0 0 . . . Cyclosporin 3 Mild 1 0.0263 0.0937 0.0789 0.7788 Cyclosporin 3 Mod 1 0.2042 0.0692 8.7084 0.0032 Cyclosporin 3 Severe 1 0.4662 0.1325 12.3789 0.0004 Cyclosporin 4 Mild 1 -0.2418 0.3214 0.5662 0.4518 Cyclosporin 4 Mod 1 0.0673 0.2205 0.0931 0.7603 Cyclosporin 4 Severe 1 -1.0526 0.7303 2.0770 0.1495

Page 259 of 577

Analysis of Maximum Likelihood Estimates Wald Eent Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Cyclosporin currently Mild 1 0.5290 0.3373 2.4603 0.1168 taking Cyclosporin currently Mod 1 1.0439 0.2236 21.7983 <.0001 taking Cyclosporin currently Severe 1 1.0216 0.4398 5.3965 0.0202 taking Cyclosporin never taking Mild 0 0 . . . Cyclosporin never taking Mod 0 0 . . . Cyclosporin never taking Severe 0 0 . . . Etanercept 3 Mild 1 -0.0509 0.0911 0.3118 0.5766 Etanercept 3 Mod 1 -0.0713 0.0705 1.0220 0.3120 Etanercept 3 Severe 1 -0.3981 0.1457 7.4653 0.0063 Etanercept 4 Mild 1 1.3033 0.5444 5.7307 0.0167 Etanercept 4 Mod 1 1.9227 0.3431 31.3968 <.0001 Etanercept 4 Severe 1 1.3439 0.8633 2.4234 0.1195 Etanercept currently Mild 1 0.1730 0.0941 3.3831 0.0659 taking Etanercept currently Mod 1 0.0891 0.0722 1.5232 0.2171 taking Etanercept currently Severe 1 -0.3383 0.1446 5.4736 0.0193 taking Etanercept never taking Mild 0 0 . . . Etanercept never taking Mod 0 0 . . . Etanercept never taking Severe 0 0 . . . Folic Acid and currently Mild 1 -0.1059 0.0761 1.9365 0.1641 Methotrexate taking Folic Acid and currently Mod 1 -0.1683 0.0598 7.9220 0.0049 Methotrexate taking Folic Acid and currently Severe 1 -0.0493 0.1190 0.1713 0.6789 Methotrexate taking

Page 260 of 577

Analysis of Maximum Likelihood Estimates Wald Eent Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Folic Acid and never taking Mild 0 0 . . . Methotrexate Folic Acid and never taking Mod 0 0 . . . Methotrexate Folic Acid and never taking Severe 0 0 . . . Methotrexate Hydroxychloroquine 3 Mild 1 0.1431 0.0736 3.7794 0.0519 Hydroxychloroquine 3 Mod 1 0.2299 0.0575 15.9873 <.0001 Hydroxychloroquine 3 Severe 1 0.1860 0.1175 2.5057 0.1134 Hydroxychloroquine 4 Mild 1 -0.0695 0.4165 0.0278 0.8676 Hydroxychloroquine 4 Mod 1 -0.0273 0.3338 0.0067 0.9348 Hydroxychloroquine 4 Severe 1 0.6305 0.5074 1.5444 0.2140 Hydroxychloroquine currently Mild 1 0.0100 0.0960 0.0109 0.9168 taking Hydroxychloroquine currently Mod 1 0.0789 0.0753 1.0991 0.2945 taking Hydroxychloroquine currently Severe 1 0.0332 0.1546 0.0463 0.8297 taking Hydroxychloroquine never taking Mild 0 0 . . . Hydroxychloroquine never taking Mod 0 0 . . . Hydroxychloroquine never taking Severe 0 0 . . . Infliximab 3 Mild 1 0.0552 0.1337 0.1707 0.6795 Infliximab 3 Mod 1 -0.2055 0.1098 3.5047 0.0612 Infliximab 3 Severe 1 0.0478 0.1974 0.0585 0.8088 Infliximab 4 Mild 1 0.4422 0.4291 1.0621 0.3027 Infliximab 4 Mod 1 -0.1974 0.3745 0.2779 0.5981 Infliximab 4 Severe 1 -0.8062 0.8952 0.8110 0.3678 Infliximab currently Mild 1 0.6440 0.1747 13.5909 0.0002 taking Infliximab currently Mod 1 0.4727 0.1396 11.4614 0.0007 taking

Page 261 of 577

Analysis of Maximum Likelihood Estimates Wald Eent Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Infliximab currently Severe 1 -0.3706 0.3711 0.9971 0.3180 taking Infliximab never taking Mild 0 0 . . . Infliximab never taking Mod 0 0 . . . Infliximab never taking Severe 0 0 . . . Prednisolone 3 Mild 1 0.3310 0.1083 9.3359 0.0022 Prednisolone 3 Mod 1 0.2552 0.0838 9.2693 0.0023 Prednisolone 3 Severe 1 0.4980 0.1834 7.3738 0.0066 Prednisolone 4 Mild 1 0.7838 0.5610 1.9520 0.1624 Prednisolone 4 Mod 1 0.5466 0.4327 1.5961 0.2065 Prednisolone 4 Severe 1 0.7162 1.0389 0.4753 0.4906 Prednisolone currently Mild 1 0.1671 0.1087 2.3642 0.1241 taking Prednisolone currently Mod 1 0.1308 0.0838 2.4345 0.1187 taking Prednisolone currently Severe 1 0.3911 0.1833 4.5509 0.0329 taking Prednisolone never taking Mild 0 0 . . . Prednisolone never taking Mod 0 0 . . . Prednisolone never taking Severe 0 0 . . . Sulphasalazine 3 Mild 1 0.0933 0.0714 1.7093 0.1911 Sulphasalazine 3 Mod 1 0.2229 0.0554 16.1883 <.0001 Sulphasalazine 3 Severe 1 0.1577 0.1136 1.9284 0.1649 Sulphasalazine 4 Mild 1 0.1273 0.3002 0.1799 0.6714 Sulphasalazine 4 Mod 1 -0.2181 0.2582 0.7132 0.3984 Sulphasalazine 4 Severe 1 0.6855 0.3912 3.0697 0.0798 Sulphasalazine currently Mild 1 0.1470 0.1112 1.7468 0.1863 taking Sulphasalazine currently Mod 1 0.00403 0.0926 0.0019 0.9653 taking

Page 262 of 577

Analysis of Maximum Likelihood Estimates Wald Eent Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Sulphasalazine currently Severe 1 -0.0445 0.1896 0.0551 0.8144 taking Sulphasalazine never taking Mild 0 0 . . . Sulphasalazine never taking Mod 0 0 . . . Sulphasalazine never taking Severe 0 0 . . . Tocilizumab 3 Mild 1 0.1595 0.2454 0.4224 0.5157 Tocilizumab 3 Mod 1 0.1835 0.1951 0.8847 0.3469 Tocilizumab 3 Severe 1 0.7127 0.3269 4.7534 0.0292 Tocilizumab 4 Mild 1 - 529.0 0.0005 0.9827 11.4739 Tocilizumab 4 Mod 1 - 222.6 0.0027 0.9587 11.5154 Tocilizumab 4 Severe 1 - 820.5 0.0002 0.9894 10.9097 Tocilizumab currently Mild 1 0.4933 0.1695 8.4682 0.0036 taking Tocilizumab currently Mod 1 0.3301 0.1348 5.9962 0.0143 taking Tocilizumab currently Severe 1 0.1795 0.2814 0.4069 0.5236 taking Tocilizumab never taking Mild 0 0 . . . Tocilizumab never taking Mod 0 0 . . . Tocilizumab never taking Severe 0 0 . . .

Page 263 of 577

Table A.35- Odds ratio estimates in EENT

Odds Ratio Estimates 95% Wald Effect InfEent Point Estimate Confidence Limits Etanercept 3 vs never taking 1 0.950 0.795 1.136 Etanercept 3 vs never taking 2 0.931 0.811 1.069 Etanercept 3 vs never taking 3 0.672 0.505 0.894 Etanercept 4 vs never taking 1 3.682 1.266 10.702 Etanercept 4 vs never taking 2 6.840 3.491 13.400 Etanercept 4 vs never taking 3 3.834 0.706 20.819 Etanercept currently taking vs never taking 1 1.189 0.989 1.430 Etanercept currently taking vs never taking 2 1.093 0.949 1.259 Etanercept currently taking vs never taking 3 0.713 0.537 0.947 Adalimumab 3 vs never taking 1 1.010 0.845 1.209 Adalimumab 3 vs never taking 2 1.200 1.049 1.373 Adalimumab 3 vs never taking 3 1.152 0.875 1.517 Adalimumab 4 vs never taking 1 0.583 0.155 2.190 Adalimumab 4 vs never taking 2 0.999 0.419 2.385 Adalimumab 4 vs never taking 3 <0.001 <0.001 >999.999 Adalimumab currently taking vs never taking 1 1.335 1.110 1.605 Adalimumab currently taking vs never taking 2 1.203 1.041 1.390 Adalimumab currently taking vs never taking 3 0.923 0.692 1.232 Anakinra 3 vs never taking 1 1.156 0.705 1.895 Anakinra 3 vs never taking 2 0.927 0.603 1.424 Anakinra 3 vs never taking 3 1.584 0.811 3.091 Anakinra 4 vs never taking 1 0.487 0.142 1.675 Anakinra 4 vs never taking 2 1.028 0.465 2.272 Anakinra 4 vs never taking 3 0.639 0.081 5.013 Anakinra currently taking vs never taking 1 <0.001 <0.001 >999.999 Anakinra currently taking vs never taking 2 6.049 2.387 15.330 Anakinra currently taking vs never taking 3 <0.001 <0.001 >999.999 Infliximab 3 vs never taking 1 1.057 0.813 1.373 Infliximab 3 vs never taking 2 0.814 0.657 1.010

Page 264 of 577

Odds Ratio Estimates 95% Wald Effect InfEent Point Estimate Confidence Limits Infliximab 3 vs never taking 3 1.049 0.712 1.544 Infliximab 4 vs never taking 1 1.556 0.671 3.608 Infliximab 4 vs never taking 2 0.821 0.394 1.710 Infliximab 4 vs never taking 3 0.447 0.077 2.582 Infliximab currently taking vs never taking 1 1.904 1.352 2.682 Infliximab currently taking vs never taking 2 1.604 1.220 2.109 Infliximab currently taking vs never taking 3 0.690 0.334 1.429 Abatacept 3 vs never taking 1 1.676 1.185 2.371 Abatacept 3 vs never taking 2 1.122 0.830 1.515 Abatacept 3 vs never taking 3 0.648 0.322 1.305 Abatacept 4 vs never taking 1 1.317 0.251 6.905 Abatacept 4 vs never taking 2 0.548 0.157 1.915 Abatacept 4 vs never taking 3 3.763 0.470 30.134 Abatacept currently taking vs never taking 1 1.400 1.027 1.908 Abatacept currently taking vs never taking 2 1.161 0.910 1.480 Abatacept currently taking vs never taking 3 0.817 0.484 1.380 Tocilizumab 3 vs never taking 1 1.173 0.725 1.897 Tocilizumab 3 vs never taking 2 1.201 0.820 1.761 Tocilizumab 3 vs never taking 3 2.039 1.075 3.870 Tocilizumab 4 vs never taking 1 <0.001 <0.001 >999.999 Tocilizumab 4 vs never taking 2 <0.001 <0.001 >999.999 Tocilizumab 4 vs never taking 3 <0.001 <0.001 >999.999 Tocilizumab currently taking vs never taking 1 1.638 1.175 2.283 Tocilizumab currently taking vs never taking 2 1.391 1.068 1.812 Tocilizumab currently taking vs never taking 3 1.197 0.689 2.077 Folic Acid currently taking vs never taking 1 0.899 0.775 1.044 Folic Acid currently taking vs never taking 2 0.845 0.752 0.950 Folic Acid currently taking vs never taking 3 0.952 0.754 1.202 Hydroxychloroquine 3 vs never taking 1 1.154 0.999 1.333 Hydroxychloroquine 3 vs never taking 2 1.259 1.124 1.409 Hydroxychloroquine 3 vs never taking 3 1.204 0.957 1.516 Hydroxychloroquine 4 vs never taking 1 0.933 0.412 2.110

Page 265 of 577

Odds Ratio Estimates 95% Wald Effect InfEent Point Estimate Confidence Limits Hydroxychloroquine 4 vs never taking 2 0.973 0.506 1.872 Hydroxychloroquine 4 vs never taking 3 1.879 0.695 5.078 Hydroxychloroquine currently taking vs never taking 1 1.010 0.837 1.219 Hydroxychloroquine currently taking vs never taking 2 1.082 0.934 1.254 Hydroxychloroquine currently taking vs never taking 3 1.034 0.764 1.400 Sulphasalazine 3 vs never taking 1 1.098 0.955 1.263 Sulphasalazine 3 vs never taking 2 1.250 1.121 1.393 Sulphasalazine 3 vs never taking 3 1.171 0.937 1.463 Sulphasalazine 4 vs never taking 1 1.136 0.631 2.046 Sulphasalazine 4 vs never taking 2 0.804 0.485 1.334 Sulphasalazine 4 vs never taking 3 1.985 0.922 4.273 Sulphasalazine currently taking vs never taking 1 1.158 0.931 1.440 Sulphasalazine currently taking vs never taking 2 1.004 0.837 1.204 Sulphasalazine currently taking vs never taking 3 0.956 0.660 1.387 Arava (Leflunomide) 3 vs never taking 1 1.116 0.929 1.341 Arava (Leflunomide) 3 vs never taking 2 1.213 1.052 1.399 Arava (Leflunomide) 3 vs never taking 3 1.160 0.876 1.536 Arava (Leflunomide) 4 vs never taking 1 0.752 0.259 2.178 Arava (Leflunomide) 4 vs never taking 2 1.581 0.863 2.898 Arava (Leflunomide) 4 vs never taking 3 0.490 0.062 3.898 Arava (Leflunomide) currently taking vs never taking 1 1.311 1.065 1.613 Arava (Leflunomide) currently taking vs never taking 2 1.161 0.981 1.374 Arava (Leflunomide) currently taking vs never taking 3 1.006 0.717 1.412 Cyclosporin 3 vs never taking 1 1.027 0.854 1.234 Cyclosporin 3 vs never taking 2 1.227 1.071 1.405 Cyclosporin 3 vs never taking 3 1.594 1.229 2.066 Cyclosporin 4 vs never taking 1 0.785 0.418 1.474 Cyclosporin 4 vs never taking 2 1.070 0.694 1.648 Cyclosporin 4 vs never taking 3 0.349 0.083 1.461 Cyclosporin currently taking vs never taking 1 1.697 0.876 3.287 Cyclosporin currently taking vs never taking 2 2.840 1.833 4.403 Cyclosporin currently taking vs never taking 3 2.778 1.173 6.577

Page 266 of 577

Odds Ratio Estimates 95% Wald Effect InfEent Point Estimate Confidence Limits Prednisolone 3 vs never taking 1 1.392 1.126 1.722 Prednisolone 3 vs never taking 2 1.291 1.095 1.521 Prednisolone 3 vs never taking 3 1.645 1.149 2.357 Prednisolone 4 vs never taking 1 2.190 0.729 6.576 Prednisolone 4 vs never taking 2 1.727 0.740 4.034 Prednisolone 4 vs never taking 3 2.047 0.267 15.680 Prednisolone currently taking vs never taking 1 1.182 0.955 1.462 Prednisolone currently taking vs never taking 2 1.140 0.967 1.343 Prednisolone currently taking vs never taking 3 1.479 1.032 2.118

Page 267 of 577

APPENDIX B: OUTPUT OF SAS FOR LUNG INFECTION

Table B.1- Complete statistics for Lung infection

Model Information Data Set WORK.IMPORT2 Response Variable InfLung InfLung Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table B.2- Observation status for Lung infection

Number of Observations Read 27711 Number of Observations Used 21506

Table B.3- response value for Lung infection

Response Profile Ordered Total Value InfLung Frequency 1 1 371 2 2 1379 3 3 624 4 4 19132

Logits modelled use InfLung='4' as the reference category.

Note: 6205 observations were deleted due to missing values for the response or explanatory variables.

Table B.4- Backward Elimination Procedure for Lung infection

Page 268 of 577

Backward Elimination Procedure Class Level Information

Class Value Design Variables

Etanercept 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Adalimumab 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Anakinra 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Infliximab 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Rituximab 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Abatacept 3 1 0 0 0

4 0 1 0 0

Page 269 of 577

currently taking 0 0 1 0

b never taking 0 0 0 1

Tocilizumab 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Golimumab 3 1 0 0

currently taking 0 1 0

b never taking 0 0 1

Methotrexate 1 1 0 0 0

2 0 1 0 0

3 0 0 1 0

4 0 0 0 1

Certolizumab 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Methotrexate (plus Folic acid) currently taking 1 0

b never taking 0 1

Hydroxychloroquine 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Sulphasalazine 3 1 0 0 0

Page 270 of 577

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Arava (Leflunomide) 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Azathioprine 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Cyclosporine 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Prednisolone 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

IM Gold 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Penicillamine 3 1 0 0 0

Page 271 of 577

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Step 0. The following effects were entered: Etanercept Adalimumab Anakinra Infliximab Rituximab Abatacept Tocilizumab Golimumab Methotrexate Certolizumab Methotrexate (plus Folic acid) Hydroxychloroquine Sulphasalazine Arava (Leflunomide) Azathioprine Cyclosporine Prednisolone IM Gold Penicillamine

Table B.5- Model Convergence status for Lung infection

Model Convergence Status Quasi-complete separation of data points detected.

Page 272 of 577

Table B.6- Model Fit statistics for Lung infection

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 19488.155 19401.988 SC 19512.083 20718.042 -2 Log L 19482.155 19071.988

Table B.7- Testing null hypothesis for Lung infection

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 410.1672 162 <.0001 Score 433.5687 162 <.0001 Wald 397.0385 162 <.0001

Table B.8- Model Fit statistics for removing covariant step 1

Step 1. Effect Certolizumab is removed:

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Page 273 of 577

Table B.9- Model Fit statistics for removing covariant step 1

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 19488.155 19396.072 SC 19512.083 20640.342 -2 Log L 19482.155 19084.072

Table B.10- Testing Null hypothesis after removing covariant step 1

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 398.0824 153 <.0001 Score 413.5994 153 <.0001 Wald 391.7407 153 <.0001

Table B.11- Residual removing covariant step 1

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 13.3285 9 0.1483

Page 274 of 577

Table B.12- Model Fit statistics for removing covariant step 2

Step 2. Effect Penicillamine is removed:

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table B.13- Model Fit statistics after removing covariant step 2

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 19488.155 19393.250 SC 19512.083 20565.735 -2 Log L 19482.155 19099.250

Table B.14- Testing Null hypothesis after removing covariant step 2

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 382.9044 144 <.0001 Score 400.4674 144 <.0001 Wald 382.3359 144 <.0001

Page 275 of 577

Table B.15- Residual removing covariant step 2

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 25.5437 18 0.1107

Table B.16- Model Fit statistics for removing covariant step 3

Step 3. Effect Methotrexate and Folic acid is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table B.17- Model Fit statistics after removing covariant step 3

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 19488.155 19390.300 SC 19512.083 20538.856 -2 Log L 19482.155 19102.300

Page 276 of 577

Table B.18- Testing Null hypothesis after removing covariant step 3

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 379.8552 141 <.0001 Score 397.4990 141 <.0001 Wald 379.3969 141 <.0001

Table B.19- Residual removing covariant step 3

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 28.2177 21 0.1341

Table B.20- Model Fit statistics for removing covariant step 4

Step 4. Effect Azathioprine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Page 277 of 577

Table B.21- Model Fit statistics after removing covariant step 4

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 19488.155 19383.751 SC 19512.083 20460.523 -2 Log L 19482.155 19113.751

Table B.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 368.4040 132 <.0001 Score 387.3219 132 <.0001 Wald 368.8331 132 <.0001

Table B.23- Residual removing covariant step 4

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 38.6702 30 0.1333

Page 278 of 577

Table B.24- Model Fit statistics for removing covariant step 5

Step 5. Effect Rituximab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table B.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 19488.155 19376.800 SC 19512.083 20381.787 -2 Log L 19482.155 19124.800

Table B.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 357.3544 123 <.0001 Score 375.9798 123 <.0001 Wald 357.3572 123 <.0001

Page 279 of 577

Table B.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 49.5870 39 0.1192

Table B.28- Model Fit statistics for removing covariant step 6

Step 6. Effect Infliximab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table B.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 19488.155 19371.000 SC 19512.083 20304.202 -2 Log L 19482.155 19137.000

Page 280 of 577

Table B.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 345.1549 114 <.0001 Score 362.5786 114 <.0001 Wald 344.7899 114 <.0001

Table B.31- Residual removing covariant step 7

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 62.0449 48 0.0838

Step 7. Effect Tocilizumab is removed:

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.

Page 281 of 577

Table B.32- Model Fit statistics after removing covariant step 6

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 19488.155 19367.574

SC 19512.083 20228.992

-2 Log L 19482.155 19151.574

Table B.33- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 330.5805 105 <.0001

Score 344.9075 105 <.0001

Wald 328.9232 105 <.0001

Table B.34- Residual removing covariant step 7

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

81.8928 57 0.0170

Page 282 of 577

Step 8. Effect Golimumab is removed:

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.

Table B.35- Model Fit statistics after removing covariant step 6

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 19488.155 19368.130

SC 19512.083 20181.691

-2 Log L 19482.155 19164.130

Table B.36- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 318.0249 99 <.0001

Score 331.5066 99 <.0001

Wald 316.0759 99 <.0001

Page 283 of 577

Table B.37- Residual removing covariant step 7

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

94.1491 63 0.0067

Step 9. Effect Arava (Leflunomide) is removed:

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.

Table B.38- Model Fit statistics after removing covariant step 6

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 19488.155 19366.491

SC 19512.083 20108.268

-2 Log L 19482.155 19180.491

Page 284 of 577

Table B.39- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 301.6634 90 <.0001

Score 316.5298 90 <.0001

Wald 300.9549 90 <.0001

Table B.40- Residual removing covariant step 7

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

111.8224 72 0.0018

Step 10. Effect Adalimumab is removed:

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.

Table B.41- Model Fit statistics after removing covariant step 6

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 19488.155 19366.037

SC 19512.083 20036.028

-2 Log L 19482.155 19198.037

Page 285 of 577

Table B.42- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 284.1184 81 <.0001

Score 299.1489 81 <.0001

Wald 283.8184 81 <.0001

Table B.43- Residual removing covariant step 7

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

128.1890 81 0.0007

Page 286 of 577

Table B.44- Summary of backward elimination in Lung

Summary of Backward Elimination

Wald Effect Number Chi- Variable Step Removed DF In Square Pr > ChiSq Label

1 Certolizumab 9 18 5.2822 0.8090 Certolizumab

2 Penicillamine 9 17 8.9767 0.4394 Penicillamine

3 Methotrexate (plus 3 16 3.0046 0.3909 Methotrexate (plus Folic Folic acid) acid)

4 Azathioprine 9 15 10.4127 0.3181 Azathioprine

5 Rituximab 9 14 10.6214 0.3026 Rituximab

6 Infliximab 9 13 12.4421 0.1895

7 Tocilizumab 9 12 14.5987 0.1026 Tocilizumab

8 Golimumab 6 11 11.7349 0.0682 Golimumab

9 Arava (Leflunomide) 9 10 16.1280 0.0643 Arava (Leflunomide)

10 Adalimumab 9 9 16.1807 0.0632

Page 287 of 577

Summary of Backward Elimination

Wald Effect Number Chi- Variable Step Removed DF In Square Pr > ChiSq Label

1 Certolizumab 9 18 5.2822 0.8090 Certolizumab

2 Penicillamine 9 17 8.9767 0.4394 Penicillamine

3 Methotrexate (plus 3 16 3.0046 0.3909 Methotrexate (plus Folic Folic acid) acid)

4 Azathioprine 9 15 10.4127 0.3181 Azathioprine

5 Rituximab 9 14 10.6214 0.3026 Rituximab

6 Infliximab 9 13 12.4421 0.1895

7 Tocilizumab 9 12 14.5987 0.1026 Tocilizumab

8 Golimumab 6 11 11.7349 0.0682 Golimumab

9 Arava (Leflunomide) 9 10 16.1280 0.0643 Arava (Leflunomide)

10 Adalimumab 9 9 16.1807 0.0632

Page 288 of 577

Table B.45- Type 3 analysis of effects in Lung

Type 3 Analysis of Effects

Wald Effect DF Chi-Square Pr > ChiSq

Etanercept 9 31.4874 0.0002

Anakinra 9 20.0990 0.0173

Abatacept 9 34.9246 <.0001

Methotrexate 9 20.5746 0.0147

Hydroxychloroquine 9 24.4648 0.0036

Sulphasalazine 9 20.8255 0.0134

Cyclosporine 9 20.6307 0.0144

Prednisolone 9 67.5034 <.0001

IM Gold 9 19.8810 0.0187

Table B.46- Analysis of maximum likelihood estimates in Lung

Analysis of Maximum Likelihood Estimates

Wald Standard Chi- Parameter InfLung DF Estimate Error Square Pr > ChiSq

Intercept Mild 1 - 132.2 0.0134 0.9079 15.2999

Intercept Mod 1 -4.3102 0.6251 47.5499 <.0001

Intercept Severe 1 -4.2292 0.5184 66.5431 <.0001

Abatacept 3 Mild 1 -0.0779 0.3188 0.0598 0.8068

Abatacept 3 Mod 1 0.3341 0.1584 4.4525 0.0349

Abatacept 3 Severe 1 0.5698 0.1984 8.2437 0.0041

Abatacept 4 Mild 1 1.0112 1.0300 0.9638 0.3262

Page 289 of 577

Analysis of Maximum Likelihood Estimates

Wald Standard Chi- Parameter InfLung DF Estimate Error Square Pr > ChiSq

Abatacept 4 Mod 1 -0.0585 0.7462 0.0062 0.9375

Abatacept 4 Severe 1 0.0807 0.9976 0.0065 0.9355

Abatacept currently Mild 1 0.3420 0.2178 2.4657 0.1164 taking

Abatacept currently Mod 1 0.5392 0.1180 20.8696 <.0001 taking

Abatacept currently Severe 1 -0.0929 0.2079 0.1996 0.6550 taking

Abatacept b never Mild 0 0 . . . taking

Abatacept b never Mod 0 0 . . . taking

Abatacept b never Severe 0 0 . . . taking

Anakinra 3 Mild 1 0.5311 0.3487 2.3189 0.1278

Anakinra 3 Mod 1 0.6433 0.1815 12.5635 0.0004

Anakinra 3 Severe 1 - 0.3302 0.0002 0.9882 0.00489

Anakinra 4 Mild 1 0.1512 0.6315 0.0573 0.8108

Anakinra 4 Mod 1 -0.0709 0.3858 0.0337 0.8543

Anakinra 4 Severe 1 -0.9695 0.6978 1.9308 0.1647

Anakinra currently Mild 1 1.1898 1.0395 1.3102 0.2524 taking

Anakinra currently Mod 1 1.0875 0.6364 2.9208 0.0874 taking

Page 290 of 577

Analysis of Maximum Likelihood Estimates

Wald Standard Chi- Parameter InfLung DF Estimate Error Square Pr > ChiSq

Anakinra currently Severe 1 - 348.5 0.0010 0.9745 taking 11.1525

Anakinra b never Mild 0 0 . . . taking

Anakinra b never Mod 0 0 . . . taking

Anakinra b never Severe 0 0 . . . taking

Cyclosporine 3 Mild 1 -0.1304 0.1673 0.6076 0.4357

Cyclosporine 3 Mod 1 0.0167 0.0825 0.0411 0.8393

Cyclosporine 3 Severe 1 -0.1421 0.1218 1.3629 0.2430

Cyclosporine 4 Mild 1 -0.6137 0.6260 0.9613 0.3269

Cyclosporine 4 Mod 1 -0.1896 0.2827 0.4496 0.5025

Cyclosporine 4 Severe 1 -0.1456 0.3622 0.1615 0.6878

Cyclosporine currently Mild 1 1.2314 0.3793 10.5374 0.0012 taking

Cyclosporine currently Mod 1 0.7209 0.2642 7.4466 0.0064 taking

Cyclosporine currently Severe 1 0.2533 0.4633 0.2990 0.5845 taking

Cyclosporine b never Mild 0 0 . . . taking

Cyclosporine b never Mod 0 0 . . . taking

Page 291 of 577

Analysis of Maximum Likelihood Estimates

Wald Standard Chi- Parameter InfLung DF Estimate Error Square Pr > ChiSq

Cyclosporine b never Severe 0 0 . . . taking

Etanercept 3 Mild 1 0.2222 0.1357 2.6825 0.1015

Etanercept 3 Mod 1 -0.1754 0.0777 5.1023 0.0239

Etanercept 3 Severe 1 0.00414 0.1087 0.0015 0.9696

Etanercept 4 Mild 1 0.9834 0.9443 1.0845 0.2977

Etanercept 4 Mod 1 1.1483 0.4726 5.9037 0.0151

Etanercept 4 Severe 1 1.7757 0.4978 12.7241 0.0004

Etanercept currently Mild 1 -0.1632 0.1350 1.4614 0.2267 taking

Etanercept currently Mod 1 -0.0438 0.0677 0.4200 0.5169 taking

Etanercept currently Severe 1 -0.1177 0.1018 1.3382 0.2473 taking

Etanercept b never Mild 0 0 . . . taking

Etanercept b never Mod 0 0 . . . taking

Etanercept b never Severe 0 0 . . . taking

Hydroxychloroquine 3 Mild 1 0.0478 0.1222 0.1529 0.6958

Hydroxychloroquine 3 Mod 1 0.0739 0.0647 1.3065 0.2530

Hydroxychloroquine 3 Severe 1 0.0384 0.0974 0.1554 0.6934

Hydroxychloroquine 4 Mild 1 -0.9431 1.0241 0.8481 0.3571

Page 292 of 577

Analysis of Maximum Likelihood Estimates

Wald Standard Chi- Parameter InfLung DF Estimate Error Square Pr > ChiSq

Hydroxychloroquine 4 Mod 1 0.1702 0.3488 0.2382 0.6255

Hydroxychloroquine 4 Severe 1 0.5359 0.3995 1.7995 0.1798

Hydroxychloroquine currently Mild 1 0.2042 0.1458 1.9617 0.1613 taking

Hydroxychloroquine currently Mod 1 0.1582 0.0804 3.8734 0.0491 taking

Hydroxychloroquine currently Severe 1 0.4302 0.1114 14.9202 0.0001 taking

Hydroxychloroquine b never Mild 0 0 . . . taking

Hydroxychloroquine b never Mod 0 0 . . . taking

Hydroxychloroquine b never Severe 0 0 . . . taking

IM Gold 3 Mild 1 -0.3051 0.1356 5.0582 0.0245

IM Gold 3 Mod 1 -0.0313 0.0676 0.2144 0.6433

IM Gold 3 Severe 1 0.1995 0.0945 4.4514 0.0349

IM Gold 4 Mild 1 0.7848 0.6264 1.5698 0.2102

IM Gold 4 Mod 1 0.6082 0.3463 3.0852 0.0790

IM Gold 4 Severe 1 0.7700 0.4271 3.2496 0.0714

IM Gold currently Mild 1 -1.3342 1.0067 1.7565 0.1851 taking

IM Gold currently Mod 1 0.2764 0.2666 1.0746 0.2999 taking

Page 293 of 577

Analysis of Maximum Likelihood Estimates

Wald Standard Chi- Parameter InfLung DF Estimate Error Square Pr > ChiSq

IM Gold currently Severe 1 -0.1061 0.4610 0.0529 0.8181 taking

IM Gold b never Mild 0 0 . . . taking

IM Gold b never Mod 0 0 . . . taking

IM Gold b never Severe 0 0 . . . taking

Methotrexate 1 Mild 1 10.9145 132.2 0.0068 0.9342

Methotrexate 1 Mod 1 1.2418 0.6189 4.0258 0.0448

Methotrexate 1 Severe 1 -0.0534 0.4945 0.0117 0.9140

Methotrexate Currently Mild 1 11.1219 132.2 0.0071 0.9329 taking

Methotrexate Currently Mod 1 1.4974 0.6251 5.7387 0.0166 taking

Methotrexate Currently Severe 1 -0.0981 0.5177 0.0359 0.8497 taking

Methotrexate 3 Mild 1 10.9202 132.2 0.0068 0.9342

Methotrexate 3 Mod 1 1.4460 0.6211 5.4198 0.0199

Methotrexate 3 Severe 1 0.1561 0.5001 0.0974 0.7549

Methotrexate 4 Mild 0 0 . . .

Methotrexate 4 Mod 0 0 . . .

Methotrexate 4 Severe 0 0 . . .

Prednisolone 3 Mild 1 0.3453 0.1830 3.5622 0.0591

Page 294 of 577

Analysis of Maximum Likelihood Estimates

Wald Standard Chi- Parameter InfLung DF Estimate Error Square Pr > ChiSq

Prednisolone 3 Mod 1 0.1782 0.0939 3.6004 0.0578

Prednisolone 3 Severe 1 0.4537 0.1637 7.6796 0.0056

Prednisolone 4 Mild 1 - 219.0 0.0023 0.9617 10.5110

Prednisolone 4 Mod 1 -1.1526 1.0341 1.2422 0.2650

Prednisolone 4 Severe 1 -0.6292 1.0771 0.3413 0.5591

Prednisolone currently Mild 1 0.4916 0.1786 7.5773 0.0059 taking

Prednisolone currently Mod 1 0.3192 0.0919 12.0656 0.0005 taking

Prednisolone currently Severe 1 0.8943 0.1574 32.2591 <.0001 taking

Prednisolone b never Mild 0 0 . . . taking

Prednisolone b never Mod 0 0 . . . taking

Prednisolone b never Severe 0 0 . . . taking

Sulphasalazine 3 Mild 1 0.00994 0.1188 0.0070 0.9333

Sulphasalazine 3 Mod 1 0.2041 0.0627 10.6112 0.0011

Sulphasalazine 3 Severe 1 0.1018 0.0918 1.2279 0.2678

Sulphasalazine 4 Mild 1 0.1048 0.4945 0.0449 0.8322

Sulphasalazine 4 Mod 1 -0.5335 0.3316 2.5882 0.1077

Sulphasalazine 4 Severe 1 0.1219 0.3492 0.1219 0.7270

Page 295 of 577

Analysis of Maximum Likelihood Estimates

Wald Standard Chi- Parameter InfLung DF Estimate Error Square Pr > ChiSq

Sulphasalazine currently Mild 1 0.3020 0.1660 3.3084 0.0689 taking

Sulphasalazine currently Mod 1 0.0113 0.1022 0.0122 0.9120 taking

Sulphasalazine currently Severe 1 0.00650 0.1447 0.0020 0.9641 taking

Sulphasalazine b never Mild 0 0 . . . taking

Sulphasalazine b never Mod 0 0 . . . taking

Sulphasalazine b never Severe 0 0 . . . taking Table B.47- Odds ratio estimates in Lung

Odds Ratio Estimates

Point 95% Wald Effect InfLung Estimate Confidence Limits

Etanercept 3 Versus never taking Mild 1.249 0.957 1.629

Etanercept 3 Versus never taking Mod 0.839 0.721 0.977

Etanercept 3 Versus never taking Severe 1.004 0.812 1.243

Etanercept 4 Versus never taking Mild 2.674 0.420 17.018

Etanercept 4 Versus never taking Mod 3.153 1.249 7.961

Etanercept 4 Versus never taking Severe 5.905 2.226 15.664

Etanercept currently taking Versus never taking Mild 0.849 0.652 1.107

Etanercept currently taking Versus never taking Mod 0.957 0.838 1.093

Page 296 of 577

Odds Ratio Estimates

Point 95% Wald Effect InfLung Estimate Confidence Limits

Etanercept currently taking Versus never taking Severe 0.889 0.728 1.085

Anakinra 3 Versus never taking Mild 1.701 0.859 3.369

Anakinra 3 Versus never taking Mod 1.903 1.333 2.716

Anakinra 3 Versus never taking Severe 0.995 0.521 1.901

Anakinra 4 Versus never taking Mild 1.163 0.337 4.010

Anakinra 4 Versus never taking Mod 0.932 0.437 1.985

Anakinra 4 Versus never taking Severe 0.379 0.097 1.489

Anakinra currently taking Versus never taking Mild 3.286 0.428 25.207

Anakinra currently taking Versus never taking Mod 2.967 0.852 10.327

Anakinra currently taking Versus never taking Severe <0.001 <0.001 >999.999

Abatacept 3 Versus never taking Mild 0.925 0.495 1.728

Abatacept 3 Versus never taking Mod 1.397 1.024 1.905

Abatacept 3 Versus never taking Severe 1.768 1.198 2.608

Abatacept 4 Versus never taking Mild 2.749 0.365 20.695

Abatacept 4 Versus never taking Mod 0.943 0.218 4.071

Abatacept 4 Versus never taking Severe 1.084 0.153 7.660

Abatacept currently taking Versus never taking Mild 1.408 0.919 2.157

Abatacept currently taking Versus never taking Mod 1.715 1.361 2.161

Abatacept currently taking Versus never taking Severe 0.911 0.606 1.370

Methotrexate 1 vs 4 Mild >999.999 <0.001 >999.999

Methotrexate 1 Versus never taking Mod 3.462 1.029 11.643

Methotrexate 1 Versus never taking Severe 0.948 0.360 2.499

Page 297 of 577

Odds Ratio Estimates

Point 95% Wald Effect InfLung Estimate Confidence Limits

Methotrexate currently taking Versus never Mild >999.999 <0.001 >999.999 taking

Methotrexate currently taking Versus never Mod 4.470 1.313 15.218 taking

Methotrexate currently taking Versus never Severe 0.907 0.329 2.501 taking

Methotrexate 3 vs 4 Mild >999.999 <0.001 >999.999

Methotrexate 3 vs 4 Mod 4.246 1.257 14.344

Methotrexate 3 vs 4 Severe 1.169 0.439 3.115

Hydroxychloroquine 3 Versus never Mild 1.049 0.826 1.333 taking

Hydroxychloroquine 3 Versus never Mod 1.077 0.949 1.222 taking

Hydroxychloroquine 3 Versus never Severe 1.039 0.858 1.258 taking

Hydroxychloroquine 4 Versus never Mild 0.389 0.052 2.898 taking

Hydroxychloroquine 4 Versus never Mod 1.186 0.598 2.349 taking

Hydroxychloroquine 4 Versus never Severe 1.709 0.781 3.739 taking

Hydroxychloroquine currently taking Versus Mild 1.227 0.922 1.632 never taking

Hydroxychloroquine currently taking Versus Mod 1.171 1.001 1.371 never taking

Hydroxychloroquine currently taking Versus Severe 1.538 1.236 1.913 never taking

Page 298 of 577

Odds Ratio Estimates

Point 95% Wald Effect InfLung Estimate Confidence Limits

Sulphasalazine 3 Versus never taking Mild 1.010 0.800 1.275

Sulphasalazine 3 Versus never taking Mod 1.226 1.085 1.387

Sulphasalazine 3 Versus never taking Severe 1.107 0.925 1.325

Sulphasalazine 4 Versus never taking Mild 1.110 0.421 2.927

Sulphasalazine 4 Versus never taking Mod 0.587 0.306 1.124

Sulphasalazine 4 Versus never taking Severe 1.130 0.570 2.240

Sulphasalazine currently taking Versus never Mild 1.353 0.977 1.873 taking

Sulphasalazine currently taking Versus never Mod 1.011 0.828 1.236 taking

Sulphasalazine currently taking Versus never Severe 1.007 0.758 1.337 taking

Cyclosporine 3 Versus never taking Mild 0.878 0.632 1.218

Cyclosporine 3 Versus never taking Mod 1.017 0.865 1.195

Cyclosporine 3 Versus never taking Severe 0.868 0.683 1.101

Cyclosporine 4 Versus never taking Mild 0.541 0.159 1.846

Cyclosporine 4 Versus never taking Mod 0.827 0.475 1.440

Cyclosporine 4 Versus never taking Severe 0.865 0.425 1.758

Cyclosporine currently taking Versus never Mild 3.426 1.629 7.206 taking

Cyclosporine currently taking Versus never Mod 2.056 1.225 3.451 taking

Cyclosporine currently taking Versus never Severe 1.288 0.520 3.194 taking

Prednisolone 3 Versus never taking Mild 1.412 0.987 2.022

Page 299 of 577

Odds Ratio Estimates

Point 95% Wald Effect InfLung Estimate Confidence Limits

Prednisolone 3 Versus never taking Mod 1.195 0.994 1.437

Prednisolone 3 Versus never taking Severe 1.574 1.142 2.170

Prednisolone 4 Versus never taking Mild <0.001 <0.001 >999.999

Prednisolone 4 Versus never taking Mod 0.316 0.042 2.397

Prednisolone 4 Versus never taking Severe 0.533 0.065 4.401

Prednisolone currently taking Versus never Mild 1.635 1.152 2.320 taking

Prednisolone currently taking Versus never Mod 1.376 1.149 1.648 taking

Prednisolone currently taking Versus never Severe 2.446 1.796 3.330 taking

IM Gold 3 Versus never taking Mild 0.737 0.565 0.962

IM Gold 3 Versus never taking Mod 0.969 0.849 1.106

IM Gold 3 Versus never taking Severe 1.221 1.014 1.469

IM Gold 4 Versus never taking Mild 2.192 0.642 7.482

IM Gold 4 Versus never taking Mod 1.837 0.932 3.621

IM Gold 4 Versus never taking Severe 2.160 0.935 4.989

IM Gold currently taking Versus never taking Mild 0.263 0.037 1.894

IM Gold currently taking Versus never taking Mod 1.318 0.782 2.223

IM Gold currently taking Versus never taking Severe 0.899 0.364 2.220

Page 300 of 577

APPENDIX C: OUTPUT OF SAS FOR NAIL AND SKIN INFECTION

Table C.1- Complete statistics for Nail and skin infection

Model Information Data Set WORK.IMPORT2 Response Variable Nail and skin infection InfSkin Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table C.2- Observation status for Nail and skin infection

Number of Observations Read 27711 Number of Observations Used 21506

Page 301 of 577

Table C.3- response value for Nail and skin infection

Response Profile Ordered Total Value Skin and nail infection Frequency 1 Mild 1253 2 Moderate 1039 3 Severe 361 4 No report 18853

Logits modelled use InfSkin='4' as the reference category.

Note: 6205 observations were deleted due to missing values for the response or explanatory variables.

3= Never taken 4= Don’t know

Page 302 of 577

Table C.4- Backward Elimination Procedure for Nail and skin infection

Backward Elimination Procedure Class Level Information Class Value Design Variables Etanercept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Adalimumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Anakinra 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Infliximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Rituximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Abatacept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Tocilizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Golimumab 3 1 0 0 currently taking 0 1 0

Page 303 of 577

never taking 0 0 1

Certolizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Folic Acid currently taking 1 0 never taking 0 1

Hydroxychloroquine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Sulphasalazine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Arava (Leflunomide) 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Azathioprine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Prednisolone 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

IM Gold injection 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0

Page 304 of 577

never taking 0 0 0 1

Penicillamine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Table C.5- Model Convergence status for Nail and skin infection

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table C.6- Model Fit statistics for Nail and skin infection

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21261.541 SC 21365.789 22505.811 -2 Log L 21335.861 20949.541

Table C.7- Testing null hypothesis for Nail and skin infection

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 386.3201 153 <.0001 Score 425.2533 153 <.0001 Wald 382.0017 153 <.0001

Page 305 of 577

Table C.8- Model Fit statistics for removing covariant step 1

Step 1. Effect Certolizumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table C.9- Model Fit statistics for removing covariant step 1

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21252.373 SC 21365.789 22424.858 -2 Log L 21335.861 20958.373

Table C.10- Testing Null hypothesis after removing covariant step 1

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 377.4876 144 <.0001 Score 417.3618 144 <.0001

Page 306 of 577

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Wald 374.6182 144 <.0001

Table C.11- Residual removing covariant step 1

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 7.5708 9 0.5779

Table C.12- Model Fit statistics for removing covariant step 2

Step 2. Effect Hydroxychloroquine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table C.13- Model Fit statistics after removing covariant step 2

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21241.218 SC 21365.789 22341.919 -2 Log L 21335.861 20965.218

Table C.14- Testing Null hypothesis after removing covariant step 2

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 370.6425 135 <.0001

Page 307 of 577

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Score 410.3001 135 <.0001 Wald 367.1567 135 <.0001

Table C.15- Residual removing covariant step 2

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 14.9868 18 0.6629

Table C.16- Model Fit statistics for removing covariant step 3

Step 3. Effect IM Gold injection is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table C.17- Model Fit statistics after removing covariant step 3

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21232.298 SC 21365.789 22261.213 -2 Log L 21335.861 20974.298

Page 308 of 577

Table C.18- Testing Null hypothesis after removing covariant step 3

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 361.5629 126 <.0001 Score 400.6553 126 <.0001 Wald 357.7151 126 <.0001

Table C.19- Residual removing covariant step 3

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 25.6149 27 0.5400

Page 309 of 577

Table C.20- Model Fit statistics for removing covariant step 4

Step 4. Effect Abatacept is removed:

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table C.21- Model Fit statistics after removing covariant step 4

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21231.336 SC 21365.789 22188.467 -2 Log L 21335.861 20991.336

Table C.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 344.5246 117 <.0001 Score 385.9767 117 <.0001 Wald 345.3642 117 <.0001

Page 310 of 577

Table C.23- Residual removing covariant step 4

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 40.5589 36 0.2763

Table C.24- Model Fit statistics for removing covariant step 5

Step 5. Effect Tocilizumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table C.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21224.235 SC 21365.789 22109.581 -2 Log L 21335.861 21002.235

Page 311 of 577

Table C.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 333.6261 108 <.0001 Score 374.4255 108 <.0001 Wald 334.8860 108 <.0001

Table C.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 52.5641 45 0.2044

Table C.28- Model Fit statistics for removing covariant step 6

Step 6. Effect Penicillamine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Page 312 of 577

Table C.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21226.325 SC 21365.789 22039.885 -2 Log L 21335.861 21022.325

Table C.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 313.5364 99 <.0001 Score 353.7825 99 <.0001 Wald 320.6062 99 <.0001

Table C.31- Residual removing covariant step 6

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 67.5116 54 0.1024

Page 313 of 577

Table C.32- Model Fit statistics for removing covariant step 7

Step 7. Effect Golimumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table C.33- Model Fit statistics after removing covariant step 7

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21223.598 SC 21365.789 21989.302 -2 Log L 21335.861 21031.598

Table C.34- Testing Null hypothesis after removing covariant step 7

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 304.2632 93 <.0001 Score 343.8880 93 <.0001 Wald 311.0294 93 <.0001

Page 314 of 577

Table C.35- Residual removing covariant step 8

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 77.4503 60 0.0643

Table C.36- Model Fit statistics for removing covariant step 8

Step 8. Effect Anakinra is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table C.37- Model Fit statistics after removing covariant step 8

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21220.330 SC 21365.789 21914.249 -2 Log L 21335.861 21046.330

Page 315 of 577

Table C.38- Testing Null hypothesis after removing covariant step 8

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 289.5314 84 <.0001 Score 325.8222 84 <.0001 Wald 295.3844 84 <.0001

Table C.39- Residual removing covariant step 8

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 94.3919 69 0.0229

Table C.40- Model Fit statistics for removing covariant step 9

Step 9. Effect Azathioprine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Page 316 of 577

Table C.41- Model Fit statistics after removing covariant step 9

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21224.705 SC 21365.789 21846.840 -2 Log L 21335.861 21068.705

Table C.42- Testing Null hypothesis after removing covariant step 9

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 267.1562 75 <.0001 Score 306.6692 75 <.0001 Wald 278.4445 75 <.0001

Table C.43- Residual removing covariant step 9

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 112.3163 78 0.0066

Table C.44- Model Fit statistics for removing covariant step 10

Step 10. Effect Azathioprine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied. Table C.45- Model Fit statistics after removing covariant step 10

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 21341.861 21220.298 SC 21365.789 21770.648

Page 317 of 577

Model Fit Statistics Criterion Intercept Only Intercept and Covariates -2 Log L 21335.861 21082.298

Table C.46- Testing Null hypothesis after removing covariant step 10

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 253.5629 66 <.0001 Score 289.6632 66 <.0001 Wald 262.6972 66 <.0001

Table C.47- Residual removing covariant step 10

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 129.4888 87 0.0021

Note: No (additional) effects met the 0.05 significance level for removal from the model.

Page 318 of 577

Table C.48- Summary of backward elimination in Nail and skin infection

Summary of Backward Elimination Effect Number Wald Variable Step Removed DF In Chi-Square Pr > ChiSq Label 1 Certolizumab 9 17 7.1370 0.6229 Certolizumab 2 Hydroxychloroquine 9 16 7.2627 0.6098 Hydroxychloroquine 3 IM Gold injection 9 15 9.6964 0.3756 IM Gold injection 4 Abatacept 9 14 11.5852 0.2377 Abatacept 5 Tocilizumab 9 13 10.3331 0.3242 Tocilizumab 6 Penicillamine 9 12 12.7175 0.1758 Penicillamine 7 Golimumab 6 11 9.9769 0.1256 Golimumab 8 Anakinra 9 10 14.7879 0.0969 9 Azathioprine 9 9 15.7929 0.0713 Azathioprine 10 Azathioprine 9 8 15.9375 0.0682 Azathioprine

Table C.49- Type 3 analysis of effects in Nail and skin infection

Type 3 Analysis of Effects Wald Effect DF Chi-Square Pr > ChiSq Etanercept 9 24.9423 0.0030

Page 319 of 577

Type 3 Analysis of Effects Wald Effect DF Chi-Square Pr > ChiSq Adalimumab 9 21.4151 0.0109 Infliximab 9 29.8970 0.0005 Rituximab 9 24.2231 0.0040 Folic Acid 3 25.6065 <.0001 Sulphasalazine 9 34.6779 <.0001 Arava (Leflunomide) 9 26.5839 0.0016 Prednisolone 9 38.0131 <.0001

Before studying the following tables please use the following codes: Skin and nail infection level (Infskin) are 1= mild, 2= moderate, 3= severe, 4= not reported Taking medication level just currently taking medication and never taken medication is important for us, but we have also a few reports for 3= stopped taking medication, 4= don’t know if patient took the medication or not.

Table C.50- Analysis of maximum likelihood estimates in Skin and nail infection

Analysis of Maximum Likelihood Estimates Taking Wald medication Skin and nail Standard Chi- Parameter status infection DF Estimate Error Square Pr > ChiSq Intercept Mild 1 -2.8266 0.0946 893.5678 <.0001 Intercept Mod 1 -2.9696 0.1016 853.9240 <.0001 Intercept Severe 1 -4.9152 0.2294 458.9436 <.0001 Adalimumab 3 Mild 1 0.1667 0.0834 3.9963 0.0456 Adalimumab 3 Mod 1 0.1512 0.0894 2.8596 0.0908 Adalimumab 3 Severe 1 0.3663 0.1402 6.8214 0.0090 Adalimumab 4 Mild 1 -0.3524 0.6067 0.3375 0.5613 Adalimumab 4 Mod 1 -0.2622 0.5666 0.2142 0.6435 Adalimumab 4 Severe 1 0.2124 0.6303 0.1136 0.7361 Adalimumab currently taking Mild 1 0.2213 0.0829 7.1270 0.0076 Adalimumab currently taking Mod 1 0.00778 0.0910 0.0073 0.9319 Adalimumab currently taking Severe 1 -0.0846 0.1565 0.2922 0.5888

Page 320 of 577

Analysis of Maximum Likelihood Estimates Taking Wald medication Skin and nail Standard Chi- Parameter status infection DF Estimate Error Square Pr > ChiSq Adalimumab never taking Mild 0 0 . . . Adalimumab never taking Mod 0 0 . . . Adalimumab never taking Severe 0 0 . . . Arava 3 Mild 1 0.0482 0.0837 0.3325 0.5642 (Leflunomide) Arava 3 Mod 1 0.1231 0.0913 1.8168 0.1777 (Leflunomide) Arava 3 Severe 1 -0.1156 0.1497 0.5969 0.4398 (Leflunomide) Arava 4 Mild 1 -0.9108 0.6082 2.2430 0.1342 (Leflunomide) Arava 4 Mod 1 0.2157 0.4031 0.2864 0.5926 (Leflunomide) Arava 4 Severe 1 -0.9201 0.7794 1.3938 0.2378 (Leflunomide) Arava currently taking Mild 1 0.3055 0.0932 10.7435 0.0010 (Leflunomide) Arava currently taking Mod 1 0.2191 0.1044 4.4011 0.0359 (Leflunomide) Arava currently taking Severe 1 0.1804 0.1689 1.1416 0.2853 (Leflunomide) Arava never taking Mild 0 0 . . . (Leflunomide) Arava never taking Mod 0 0 . . . (Leflunomide) Arava never taking Severe 0 0 . . . (Leflunomide) Etanercept 3 Mild 1 -0.0175 0.0818 0.0459 0.8304 Etanercept 3 Mod 1 -0.0827 0.0890 0.8645 0.3525 Etanercept 3 Severe 1 -0.1033 0.1422 0.5275 0.4677 Etanercept 4 Mild 1 0.4892 0.5332 0.8418 0.3589

Page 321 of 577

Analysis of Maximum Likelihood Estimates Taking Wald medication Skin and nail Standard Chi- Parameter status infection DF Estimate Error Square Pr > ChiSq Etanercept 4 Mod 1 1.1722 0.4277 7.5118 0.0061 Etanercept 4 Severe 1 1.4312 0.5565 6.6133 0.0101 Etanercept currently taking Mild 1 0.00155 0.0838 0.0003 0.9852 Etanercept currently taking Mod 1 -0.2081 0.0910 5.2237 0.0223 Etanercept currently taking Severe 1 -0.3476 0.1545 5.0621 0.0245 Etanercept never taking Mild 0 0 . . . Etanercept never taking Mod 0 0 . . . Etanercept never taking Severe 0 0 . . . Infliximab 3 Mild 1 -0.1862 0.1367 1.8550 0.1732 Infliximab 3 Mod 1 0.2063 0.1283 2.5871 0.1077 Infliximab 3 Severe 1 0.4813 0.1875 6.5890 0.0103 Infliximab 4 Mild 1 0.4947 0.3630 1.8573 0.1729 Infliximab 4 Mod 1 0.8282 0.3319 6.2268 0.0126 Infliximab 4 Severe 1 1.2836 0.4326 8.8057 0.0030 Infliximab currently taking Mild 1 0.2213 0.1753 1.5940 0.2067 Infliximab currently taking Mod 1 -0.1129 0.2084 0.2935 0.5880 Infliximab currently taking Severe 1 0.5461 0.2665 4.1992 0.0404 Infliximab never taking Mild 0 0 . . . Infliximab never taking Mod 0 0 . . . Infliximab never taking Severe 0 0 . . . Methotrexate and currently taking Mild 1 -0.3023 0.0732 17.0813 <.0001 Folic Acid Methotrexate and currently taking Mod 1 -0.1809 0.0780 5.3701 0.0205 Folic Acid Methotrexate and currently taking Severe 1 0.2144 0.1185 3.2733 0.0704 Folic Acid Methotrexate and never taking Mild 0 0 . . . Folic Acid Methotrexate and never taking Mod 0 0 . . . Folic Acid

Page 322 of 577

Analysis of Maximum Likelihood Estimates Taking Wald medication Skin and nail Standard Chi- Parameter status infection DF Estimate Error Square Pr > ChiSq Methotrexate and never taking Severe 0 0 . . . Folic Acid Prednisolone 3 Mild 1 -0.1068 0.0946 1.2757 0.2587 Prednisolone 3 Mod 1 -0.0786 0.1018 0.5955 0.4403 Prednisolone 3 Severe 1 0.6423 0.2296 7.8244 0.0052 Prednisolone 4 Mild 1 0.8885 0.4683 3.5995 0.0578 Prednisolone 4 Mod 1 -0.2050 0.6427 0.1017 0.7498 Prednisolone 4 Severe 1 -0.4415 1.1782 0.1404 0.7079 Prednisolone currently taking Mild 1 0.1044 0.0911 1.3132 0.2518 Prednisolone currently taking Mod 1 0.0129 0.0995 0.0169 0.8967 Prednisolone currently taking Severe 1 0.9617 0.2236 18.4899 <.0001 Prednisolone never taking Mild 0 0 . . . Prednisolone never taking Mod 0 0 . . . Prednisolone never taking Severe 0 0 . . . Rituximab 3 Mild 1 0.2088 0.1608 1.6855 0.1942 Rituximab 3 Mod 1 -0.2103 0.1883 1.2478 0.2640 Rituximab 3 Severe 1 -0.2355 0.2839 0.6882 0.4068 Rituximab 4 Mild 1 -0.5847 0.5650 1.0710 0.3007 Rituximab 4 Mod 1 -0.8432 0.5514 2.3389 0.1262 Rituximab 4 Severe 1 0.1096 0.5674 0.0373 0.8469 Rituximab currently taking Mild 1 -0.4992 0.1693 8.6962 0.0032 Rituximab currently taking Mod 1 -0.4075 0.1673 5.9319 0.0149 Rituximab currently taking Severe 1 -0.4535 0.2610 3.0184 0.0823 Rituximab never taking Mild 0 0 . . . Rituximab never taking Mod 0 0 . . . Rituximab never taking Severe 0 0 . . . Sulphasalazine 3 Mild 1 0.0853 0.0636 1.8015 0.1795 Sulphasalazine 3 Mod 1 0.1456 0.0706 4.2466 0.0393 Sulphasalazine 3 Severe 1 0.3094 0.1209 6.5541 0.0105 Sulphasalazine 4 Mild 1 -0.0553 0.2821 0.0385 0.8445 Sulphasalazine 4 Mod 1 0.6406 0.2353 7.4089 0.0065

Page 323 of 577

Analysis of Maximum Likelihood Estimates Taking Wald medication Skin and nail Standard Chi- Parameter status infection DF Estimate Error Square Pr > ChiSq Sulphasalazine 4 Severe 1 0.7439 0.3634 4.1916 0.0406 Sulphasalazine currently taking Mild 1 -0.3428 0.1158 8.7686 0.0031 Sulphasalazine currently taking Mod 1 -0.1098 0.1187 0.8565 0.3547 Sulphasalazine currently taking Severe 1 0.0885 0.1946 0.2068 0.6493 Sulphasalazine never taking Mild 0 0 . . . Sulphasalazine never taking Mod 0 0 . . . Sulphasalazine never taking Severe 0 0 . . .

Page 324 of 577

Table C.51- Odds ratio estimates in Skin and nail infection

Odds Ratio Estimates Point 95% Wald Effect InfSkin Estimate Confidence Limits Etanercept 3 vs never taking Mild 0.983 0.837 1.154 Etanercept 3 vs never taking Mod 0.921 0.773 1.096 Etanercept 3 vs never taking Severe 0.902 0.682 1.192 Etanercept 4 vs never taking Mild 1.631 0.574 4.638 Etanercept 4 vs never taking Mod 3.229 1.396 7.467 Etanercept 4 vs never taking Severe 4.184 1.406 12.454 Etanercept currently taking vs never taking Mild 1.002 0.850 1.180 Etanercept currently taking vs never taking Mod 0.812 0.679 0.971 Etanercept currently taking vs never taking Severe 0.706 0.522 0.956 Adalimumab 3 vs never taking Mild 1.181 1.003 1.391 Adalimumab 3 vs never taking Mod 1.163 0.976 1.386 Adalimumab 3 vs never taking Severe 1.442 1.096 1.899 Adalimumab 4 vs never taking Mild 0.703 0.214 2.309 Adalimumab 4 vs never taking Mod 0.769 0.253 2.336 Adalimumab 4 vs never taking Severe 1.237 0.360 4.253 Adalimumab currently taking vs never taking Mild 1.248 1.061 1.468 Adalimumab currently taking vs never taking Mod 1.008 0.843 1.205 Adalimumab currently taking vs never taking Severe 0.919 0.676 1.249 Infliximab 3 vs never taking Mild 0.830 0.635 1.085 Infliximab 3 vs never taking Mod 1.229 0.956 1.580 Infliximab 3 vs never taking Severe 1.618 1.121 2.337 Infliximab 4 vs never taking Mild 1.640 0.805 3.341 Infliximab 4 vs never taking Mod 2.289 1.194 4.387 Infliximab 4 vs never taking Severe 3.610 1.546 8.427 Infliximab currently taking vs never taking Mild 1.248 0.885 1.759 Infliximab currently taking vs never taking Mod 0.893 0.594 1.344 Infliximab currently taking vs never taking Severe 1.727 1.024 2.911 Rituximab 3 vs never taking Mild 1.232 0.899 1.689 Rituximab 3 vs never taking Mod 0.810 0.560 1.172 Rituximab 3 vs never taking Severe 0.790 0.453 1.378

Page 325 of 577

Odds Ratio Estimates Point 95% Wald Effect InfSkin Estimate Confidence Limits Rituximab 4 vs never taking Mild 0.557 0.184 1.687 Rituximab 4 vs never taking Mod 0.430 0.146 1.268 Rituximab 4 vs never taking Severe 1.116 0.367 3.393 Rituximab currently taking vs never taking Mild 0.607 0.436 0.846 Rituximab currently taking vs never taking Mod 0.665 0.479 0.924 Rituximab currently taking vs never taking Severe 0.635 0.381 1.060 Methotrexate and Folic Acid currently taking vs never Mild 0.739 0.640 0.853 taking Methotrexate and Folic Acid currently taking vs never Mod 0.835 0.716 0.972 taking Methotrexate and Folic Acid currently taking vs never Severe 1.239 0.982 1.563 taking Sulphasalazine 3 vs never taking Mild 1.089 0.961 1.234 Sulphasalazine 3 vs never taking Mod 1.157 1.007 1.328 Sulphasalazine 3 vs never taking Severe 1.363 1.075 1.727 Sulphasalazine 4 vs never taking Mild 0.946 0.544 1.645 Sulphasalazine 4 vs never taking Mod 1.898 1.196 3.010 Sulphasalazine 4 vs never taking Severe 2.104 1.032 4.289 Sulphasalazine currently taking vs never taking Mild 0.710 0.566 0.891 Sulphasalazine currently taking vs never taking Mod 0.896 0.710 1.131 Sulphasalazine currently taking vs never taking Severe 1.093 0.746 1.600 Arava (Leflunomide) 3 vs never taking Mild 1.049 0.891 1.236 Arava (Leflunomide) 3 vs never taking Mod 1.131 0.946 1.353 Arava (Leflunomide) 3 vs never taking Severe 0.891 0.664 1.194 Arava (Leflunomide) 4 vs never taking Mild 0.402 0.122 1.325 Arava (Leflunomide) 4 vs never taking Mod 1.241 0.563 2.734 Arava (Leflunomide) 4 vs never taking Severe 0.398 0.086 1.836 Arava (Leflunomide) currently taking vs never taking Mild 1.357 1.131 1.629 Arava (Leflunomide) currently taking vs never taking Mod 1.245 1.015 1.528 Arava (Leflunomide) currently taking vs never taking Severe 1.198 0.860 1.668 Prednisolone 3 vs never taking Mild 0.899 0.747 1.082 Prednisolone 3 vs never taking Mod 0.924 0.757 1.129

Page 326 of 577

Odds Ratio Estimates Point 95% Wald Effect InfSkin Estimate Confidence Limits Prednisolone 3 vs never taking Severe 1.901 1.212 2.981 Prednisolone 4 vs never taking Mild 2.432 0.971 6.089 Prednisolone 4 vs never taking Mod 0.815 0.231 2.871 Prednisolone 4 vs never taking Severe 0.643 0.064 6.474 Prednisolone currently taking vs never taking Mild 1.110 0.929 1.327 Prednisolone currently taking vs never taking Mod 1.013 0.833 1.231 Prednisolone currently taking vs never taking Severe 2.616 1.688 4.055

Page 327 of 577

APPENDIX D: OUTPUT OF SAS FOR ARTIFICIAL JOINT INFECTION

Table D.1- Complete statistics for Artificial Joint infection

Model Information Data Set WORK.IMPORT2 Response Variable TB Infection TB Infection Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table D.2- Observation status for Artificial Joint infection

Number of Observations Read 27711 Number of Observations Used 21506

Table D.3- response value for Artificial Joint infection

Response Profile Ordered Total Value TB Infection Frequency 1 1 1050 2 2 1829 3 3 406 4 4 18221

Page 328 of 577

Table D.4- Backward Elimination Procedure for ARTIFICIAL JOINT infection

Logits modelled use TB Infection='4' as the reference category. Note: 6205 observations were deleted due to missing values for the response or explanatory variables. Backward Elimination Procedure

Class Level Information Class Value Design Variables Etanercept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Adalimumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Anakinra 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Infliximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Rituximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Abatacept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Tocilizumab 3 1 0 0 0 4 0 1 0 0

Page 329 of 577

Class Level Information Class Value Design Variables currently taking 0 0 1 0 never taking 0 0 0 1

Golimumab 3 1 0 0 currently taking 0 1 0 never taking 0 0 1

Certolizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Folic Acid currently taking 1 0 never taking 0 1

Hydroxychloroquine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Sulphasalazine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Arava (Leflunomide) 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Azathioprine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Cyclosporin 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Prednisolone 3 1 0 0 0

Page 330 of 577

Class Level Information Class Value Design Variables 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

IM Gold injection 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Penicillamine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Page 331 of 577

Table D.5- Model Convergence status for ARTIFICIAL JOINT infection Step 0. The following effects were entered: Intercept Etanercept Adalimumab Anakinra Infliximab Rituximab Abatacept Tocilizumab Golimumab Certolizumab Folic Acid Hydroxychloroquine Sulphasalazine Arava (Leflunomide) Azathioprine Cyclosporin Prednisolone IM Gold injection Penicillamine

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table D.6- Model Fit statistics for ARTIFICIAL JOINT infection

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24501.128 SC 24650.284 25745.398 -2 Log L 24620.355 24189.128

Table D.7- Testing null hypothesis for ARTIFICIAL JOINT infection

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 431.2272 153 <.0001

Page 332 of 577

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Score 463.0664 153 <.0001 Wald 419.5882 153 <.0001

Table D.8- Model Fit statistics for removing covariant step 1 Step 1. Effect Azathioprine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table D.9- Model Fit statistics for removing covariant step 1

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24488.566 SC 24650.284 25661.051 -2 Log L 24620.355 24194.566

Table D.10- Testing Null hypothesis after removing covariant step 1

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 425.7897 144 <.0001 Score 457.8861 144 <.0001 Wald 415.1007 144 <.0001

Table D.11- Residual removing covariant step 1

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 5.0524 9 0.8297

Page 333 of 577

Table D.12- Model Fit statistics for removing covariant step 2

Step 2. Effect Certolizumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table D.13- Model Fit statistics after removing covariant step 2

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24476.658 SC 24650.284 25577.358 -2 Log L 24620.355 24200.658

Table D.14- Testing Null hypothesis after removing covariant step 2

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 419.6974 135 <.0001 Score 450.9468 135 <.0001 Wald 408.7712 135 <.0001

Page 334 of 577

Table D.15- Residual removing covariant step 2

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 10.9161 18 0.8979

Table D.16- Model Fit statistics for removing covariant step 3

Step 3. Effect Penicillamine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table D.17- Model Fit statistics after removing covariant step 3

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24473.673 SC 24650.284 25502.589 -2 Log L 24620.355 24215.673

Table D.18- Testing Null hypothesis after removing covariant step 3

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 404.6821 126 <.0001 Score 440.0301 126 <.0001 Wald 401.9656 126 <.0001

Page 335 of 577

Table D.19- Residual removing covariant step 3

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 22.0077 27 0.7370

Table D.20- Model Fit statistics for removing covariant step 4

Step 4. Effect IM Gold injection is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table D.21- Model Fit statistics after removing covariant step 4

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24465.880 SC 24650.284 25423.011 -2 Log L 24620.355 24225.880

Table D.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 394.4751 117 <.0001

Page 336 of 577

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Score 430.4313 117 <.0001 Wald 392.2553 117 <.0001

Table D.23- Residual removing covariant step 4

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 31.2787 36 0.6926

Table D.24- Model Fit statistics for removing covariant step 5

Step 5. Effect Rituximab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table D.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24461.305 SC 24650.284 25346.650 -2 Log L 24620.355 24239.305

Page 337 of 577

Table D.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 381.0509 108 <.0001 Score 415.9895 108 <.0001 Wald 378.4582 108 <.0001

Table D.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 44.9975 45 0.4721

Table D.28- Model Fit statistics for removing covariant step 6

Step 6. Effect Golimumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Page 338 of 577

Table D.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24462.511 SC 24650.284 25300.000 -2 Log L 24620.355 24252.511

Table D.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 367.8443 102 <.0001 Score 403.4935 102 <.0001 Wald 366.8141 102 <.0001

Table D.31- Residual removing covariant step 7

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 57.2672 51 0.2539

Note: No (additional) effects met the 0.05 significance level for removal from the model.

Page 339 of 577

Table D.32- Summary of backward elimination in ARTIFICIAL JOINT

Summary of Backward Elimination Effect Number Wald Variable Step Removed DF In Chi-Square Pr > ChiSq Label 1 Azathioprine 9 17 4.9893 0.8352 Azathioprine 2 Certolizumab 9 16 5.4537 0.7931 Certolizumab 3 Penicillamine 9 15 7.1956 0.6168 Penicillamine 4 IM Gold injection 9 14 9.1915 0.4198 IM Gold injection 5 Rituximab 9 13 13.6536 0.1352 Rituximab 6 Golimumab 6 12 11.2165 0.0819 Golimumab

Table D.33- Type 3 analysis of effects in ARTIFICIAL JOINT

Page 340 of 577

Table D.34- Analysis of maximum likelihood estimates in ARTIFICIAL JOINT

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Intercept 1 1 -3.4872 0.1190 859.2759 <.0001 Intercept 2 1 -2.9786 0.0928 1031.0501 <.0001 Intercept 3 1 -4.3609 0.1917 517.2695 <.0001 Etanercept 3 1 1 -0.0509 0.0911 0.3118 0.5766 Etanercept 3 2 1 -0.0713 0.0705 1.0220 0.3120 Etanercept 3 3 1 -0.3981 0.1457 7.4653 0.0063 Etanercept 4 1 1 1.3033 0.5444 5.7307 0.0167 Etanercept 4 2 1 1.9227 0.3431 31.3968 <.0001 Etanercept 4 3 1 1.3439 0.8633 2.4234 0.1195 Etanercept currently 1 1 0.1730 0.0941 3.3831 0.0659 taking Etanercept currently 2 1 0.0891 0.0722 1.5232 0.2171 taking Etanercept currently 3 1 -0.3383 0.1446 5.4736 0.0193 taking Etanercept never taking 1 0 0 . . . Etanercept never taking 2 0 0 . . . Etanercept never taking 3 0 0 . . . Adalimumab 3 1 1 0.0104 0.0914 0.0129 0.9094 Adalimumab 3 2 1 0.1823 0.0686 7.0504 0.0079 Adalimumab 3 3 1 0.1418 0.1403 1.0222 0.3120 Adalimumab 4 1 1 -0.5402 0.6756 0.6394 0.4239 Adalimumab 4 2 1 - 0.4440 0.0000 0.9984 0.00090 Adalimumab 4 3 1 - 147.6 0.0048 0.9447 10.2462 Adalimumab currently 1 1 0.2887 0.0941 9.4206 0.0021 taking Adalimumab currently 2 1 0.1847 0.0737 6.2813 0.0122 taking

Page 341 of 577

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Adalimumab currently 3 1 -0.0798 0.1470 0.2946 0.5873 taking Adalimumab never taking 1 0 0 . . . Adalimumab never taking 2 0 0 . . . Adalimumab never taking 3 0 0 . . . Anakinra 3 1 1 0.1448 0.2523 0.3295 0.5659 Anakinra 3 2 1 -0.0761 0.2191 0.1205 0.7285 Anakinra 3 3 1 0.4597 0.3413 1.8146 0.1780 Anakinra 4 1 1 -0.7187 0.6297 1.3026 0.2537 Anakinra 4 2 1 0.0275 0.4047 0.0046 0.9459 Anakinra 4 3 1 -0.4484 1.0513 0.1819 0.6697 Anakinra currently 1 1 - 512.1 0.0005 0.9814 taking 11.9697 Anakinra currently 2 1 1.7999 0.4745 14.3901 0.0001 taking Anakinra currently 3 1 - 809.5 0.0002 0.9879 taking 12.2422 Anakinra never taking 1 0 0 . . . Anakinra never taking 2 0 0 . . . Anakinra never taking 3 0 0 . . . Infliximab 3 1 1 0.0552 0.1337 0.1707 0.6795 Infliximab 3 2 1 -0.2055 0.1098 3.5047 0.0612 Infliximab 3 3 1 0.0478 0.1974 0.0585 0.8088 Infliximab 4 1 1 0.4422 0.4291 1.0621 0.3027 Infliximab 4 2 1 -0.1974 0.3745 0.2779 0.5981 Infliximab 4 3 1 -0.8062 0.8952 0.8110 0.3678 Infliximab currently 1 1 0.6440 0.1747 13.5909 0.0002 taking Infliximab currently 2 1 0.4727 0.1396 11.4614 0.0007 taking

Page 342 of 577

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Infliximab currently 3 1 -0.3706 0.3711 0.9971 0.3180 taking Infliximab never taking 1 0 0 . . . Infliximab never taking 2 0 0 . . . Infliximab never taking 3 0 0 . . . Abatacept 3 1 1 0.5166 0.1769 8.5260 0.0035 Abatacept 3 2 1 0.1147 0.1534 0.5587 0.4548 Abatacept 3 3 1 -0.4339 0.3573 1.4747 0.2246 Abatacept 4 1 1 0.2751 0.8455 0.1059 0.7449 Abatacept 4 2 1 -0.6022 0.6388 0.8887 0.3458 Abatacept 4 3 1 1.3251 1.0615 1.5582 0.2119 Abatacept currently 1 1 0.3362 0.1582 4.5176 0.0335 taking Abatacept currently 2 1 0.1491 0.1240 1.4460 0.2292 taking Abatacept currently 3 1 -0.2016 0.2673 0.5686 0.4508 taking Abatacept never taking 1 0 0 . . . Abatacept never taking 2 0 0 . . . Abatacept never taking 3 0 0 . . . Tocilizumab 3 1 1 0.1595 0.2454 0.4224 0.5157 Tocilizumab 3 2 1 0.1835 0.1951 0.8847 0.3469 Tocilizumab 3 3 1 0.7127 0.3269 4.7534 0.0292 Tocilizumab 4 1 1 - 529.0 0.0005 0.9827 11.4739 Tocilizumab 4 2 1 - 222.6 0.0027 0.9587 11.5154 Tocilizumab 4 3 1 - 820.5 0.0002 0.9894 10.9097 Tocilizumab currently 1 1 0.4933 0.1695 8.4682 0.0036 taking

Page 343 of 577

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Tocilizumab currently 2 1 0.3301 0.1348 5.9962 0.0143 taking Tocilizumab currently 3 1 0.1795 0.2814 0.4069 0.5236 taking Tocilizumab never taking 1 0 0 . . . Tocilizumab never taking 2 0 0 . . . Tocilizumab never taking 3 0 0 . . . Folic Acid currently 1 1 -0.1059 0.0761 1.9365 0.1641 taking Folic Acid currently 2 1 -0.1683 0.0598 7.9220 0.0049 taking Folic Acid currently 3 1 -0.0493 0.1190 0.1713 0.6789 taking Folic Acid never taking 1 0 0 . . . Folic Acid never taking 2 0 0 . . . Folic Acid never taking 3 0 0 . . . Hydroxychloroquine 3 1 1 0.1431 0.0736 3.7794 0.0519 Hydroxychloroquine 3 2 1 0.2299 0.0575 15.9873 <.0001 Hydroxychloroquine 3 3 1 0.1860 0.1175 2.5057 0.1134 Hydroxychloroquine 4 1 1 -0.0695 0.4165 0.0278 0.8676 Hydroxychloroquine 4 2 1 -0.0273 0.3338 0.0067 0.9348 Hydroxychloroquine 4 3 1 0.6305 0.5074 1.5444 0.2140 Hydroxychloroquine currently 1 1 0.0100 0.0960 0.0109 0.9168 taking Hydroxychloroquine currently 2 1 0.0789 0.0753 1.0991 0.2945 taking Hydroxychloroquine currently 3 1 0.0332 0.1546 0.0463 0.8297 taking Hydroxychloroquine never taking 1 0 0 . . . Hydroxychloroquine never taking 2 0 0 . . . Hydroxychloroquine never taking 3 0 0 . . .

Page 344 of 577

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Sulphasalazine 3 1 1 0.0933 0.0714 1.7093 0.1911 Sulphasalazine 3 2 1 0.2229 0.0554 16.1883 <.0001 Sulphasalazine 3 3 1 0.1577 0.1136 1.9284 0.1649 Sulphasalazine 4 1 1 0.1273 0.3002 0.1799 0.6714 Sulphasalazine 4 2 1 -0.2181 0.2582 0.7132 0.3984 Sulphasalazine 4 3 1 0.6855 0.3912 3.0697 0.0798 Sulphasalazine currently 1 1 0.1470 0.1112 1.7468 0.1863 taking Sulphasalazine currently 2 1 0.00403 0.0926 0.0019 0.9653 taking Sulphasalazine currently 3 1 -0.0445 0.1896 0.0551 0.8144 taking Sulphasalazine never taking 1 0 0 . . . Sulphasalazine never taking 2 0 0 . . . Sulphasalazine never taking 3 0 0 . . . Arava (Leflunomide) 3 1 1 0.1098 0.0935 1.3804 0.2400 Arava (Leflunomide) 3 2 1 0.1933 0.0729 7.0343 0.0080 Arava (Leflunomide) 3 3 1 0.1484 0.1434 1.0712 0.3007 Arava (Leflunomide) 4 1 1 -0.2856 0.5428 0.2768 0.5988 Arava (Leflunomide) 4 2 1 0.4582 0.3091 2.1967 0.1383 Arava (Leflunomide) 4 3 1 -0.7134 1.0581 0.4546 0.5002 Arava (Leflunomide) currently 1 1 0.2705 0.1060 6.5075 0.0107 taking Arava (Leflunomide) currently 2 1 0.1492 0.0858 3.0250 0.0820 taking Arava (Leflunomide) currently 3 1 0.00639 0.1726 0.0014 0.9705 taking Arava (Leflunomide) never taking 1 0 0 . . . Arava (Leflunomide) never taking 2 0 0 . . . Arava (Leflunomide) never taking 3 0 0 . . . Cyclosporin 3 1 1 0.0263 0.0937 0.0789 0.7788

Page 345 of 577

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Cyclosporin 3 2 1 0.2042 0.0692 8.7084 0.0032 Cyclosporin 3 3 1 0.4662 0.1325 12.3789 0.0004 Cyclosporin 4 1 1 -0.2418 0.3214 0.5662 0.4518 Cyclosporin 4 2 1 0.0673 0.2205 0.0931 0.7603 Cyclosporin 4 3 1 -1.0526 0.7303 2.0770 0.1495 Cyclosporin currently 1 1 0.5290 0.3373 2.4603 0.1168 taking Cyclosporin currently 2 1 1.0439 0.2236 21.7983 <.0001 taking Cyclosporin currently 3 1 1.0216 0.4398 5.3965 0.0202 taking Cyclosporin never taking 1 0 0 . . . Cyclosporin never taking 2 0 0 . . . Cyclosporin never taking 3 0 0 . . . Prednisolone 3 1 1 0.3310 0.1083 9.3359 0.0022 Prednisolone 3 2 1 0.2552 0.0838 9.2693 0.0023 Prednisolone 3 3 1 0.4980 0.1834 7.3738 0.0066 Prednisolone 4 1 1 0.7838 0.5610 1.9520 0.1624 Prednisolone 4 2 1 0.5466 0.4327 1.5961 0.2065 Prednisolone 4 3 1 0.7162 1.0389 0.4753 0.4906 Prednisolone currently 1 1 0.1671 0.1087 2.3642 0.1241 taking Prednisolone currently 2 1 0.1308 0.0838 2.4345 0.1187 taking Prednisolone currently 3 1 0.3911 0.1833 4.5509 0.0329 taking Prednisolone never taking 1 0 0 . . . Prednisolone never taking 2 0 0 . . . Prednisolone never taking 3 0 0 . . .

Page 346 of 577

Table D.35- Odds ratio estimates in ARTIFICIAL JOINT

Odds Ratio Estimates TB Point 95% Wald Effect Infection Estimate Confidence Limits Etanercept 3 vs never taking 1 0.950 0.795 1.136 Etanercept 3 vs never taking 2 0.931 0.811 1.069 Etanercept 3 vs never taking 3 0.672 0.505 0.894 Etanercept 4 vs never taking 1 3.682 1.266 10.702 Etanercept 4 vs never taking 2 6.840 3.491 13.400 Etanercept 4 vs never taking 3 3.834 0.706 20.819 Etanercept currently taking vs never taking 1 1.189 0.989 1.430 Etanercept currently taking vs never taking 2 1.093 0.949 1.259 Etanercept currently taking vs never taking 3 0.713 0.537 0.947 Adalimumab 3 vs never taking 1 1.010 0.845 1.209 Adalimumab 3 vs never taking 2 1.200 1.049 1.373 Adalimumab 3 vs never taking 3 1.152 0.875 1.517 Adalimumab 4 vs never taking 1 0.583 0.155 2.190 Adalimumab 4 vs never taking 2 0.999 0.419 2.385 Adalimumab 4 vs never taking 3 <0.001 <0.001 >999.999 Adalimumab currently taking vs never taking 1 1.335 1.110 1.605 Adalimumab currently taking vs never taking 2 1.203 1.041 1.390 Adalimumab currently taking vs never taking 3 0.923 0.692 1.232 Anakinra 3 vs never taking 1 1.156 0.705 1.895 Anakinra 3 vs never taking 2 0.927 0.603 1.424 Anakinra 3 vs never taking 3 1.584 0.811 3.091 Anakinra 4 vs never taking 1 0.487 0.142 1.675 Anakinra 4 vs never taking 2 1.028 0.465 2.272 Anakinra 4 vs never taking 3 0.639 0.081 5.013 Anakinra currently taking vs never taking 1 <0.001 <0.001 >999.999 Anakinra currently taking vs never taking 2 6.049 2.387 15.330 Anakinra currently taking vs never taking 3 <0.001 <0.001 >999.999 Infliximab 3 vs never taking 1 1.057 0.813 1.373 Infliximab 3 vs never taking 2 0.814 0.657 1.010 Infliximab 3 vs never taking 3 1.049 0.712 1.544

Page 347 of 577

Odds Ratio Estimates TB Point 95% Wald Effect Infection Estimate Confidence Limits Infliximab 4 vs never taking 1 1.556 0.671 3.608 Infliximab 4 vs never taking 2 0.821 0.394 1.710 Infliximab 4 vs never taking 3 0.447 0.077 2.582 Infliximab currently taking vs never taking 1 1.904 1.352 2.682 Infliximab currently taking vs never taking 2 1.604 1.220 2.109 Infliximab currently taking vs never taking 3 0.690 0.334 1.429 Abatacept 3 vs never taking 1 1.676 1.185 2.371 Abatacept 3 vs never taking 2 1.122 0.830 1.515 Abatacept 3 vs never taking 3 0.648 0.322 1.305 Abatacept 4 vs never taking 1 1.317 0.251 6.905 Abatacept 4 vs never taking 2 0.548 0.157 1.915 Abatacept 4 vs never taking 3 3.763 0.470 30.134 Abatacept currently taking vs never taking 1 1.400 1.027 1.908 Abatacept currently taking vs never taking 2 1.161 0.910 1.480 Abatacept currently taking vs never taking 3 0.817 0.484 1.380 Tocilizumab 3 vs never taking 1 1.173 0.725 1.897 Tocilizumab 3 vs never taking 2 1.201 0.820 1.761 Tocilizumab 3 vs never taking 3 2.039 1.075 3.870 Tocilizumab 4 vs never taking 1 <0.001 <0.001 >999.999 Tocilizumab 4 vs never taking 2 <0.001 <0.001 >999.999 Tocilizumab 4 vs never taking 3 <0.001 <0.001 >999.999 Tocilizumab currently taking vs never taking 1 1.638 1.175 2.283 Tocilizumab currently taking vs never taking 2 1.391 1.068 1.812 Tocilizumab currently taking vs never taking 3 1.197 0.689 2.077 Folic Acid currently taking vs never taking 1 0.899 0.775 1.044 Folic Acid currently taking vs never taking 2 0.845 0.752 0.950 Folic Acid currently taking vs never taking 3 0.952 0.754 1.202 Hydroxychloroquine 3 vs never taking 1 1.154 0.999 1.333 Hydroxychloroquine 3 vs never taking 2 1.259 1.124 1.409 Hydroxychloroquine 3 vs never taking 3 1.204 0.957 1.516 Hydroxychloroquine 4 vs never taking 1 0.933 0.412 2.110 Hydroxychloroquine 4 vs never taking 2 0.973 0.506 1.872

Page 348 of 577

Odds Ratio Estimates TB Point 95% Wald Effect Infection Estimate Confidence Limits Hydroxychloroquine 4 vs never taking 3 1.879 0.695 5.078 Hydroxychloroquine currently taking vs never 1 1.010 0.837 1.219 taking Hydroxychloroquine currently taking vs never 2 1.082 0.934 1.254 taking Hydroxychloroquine currently taking vs never 3 1.034 0.764 1.400 taking Sulphasalazine 3 vs never taking 1 1.098 0.955 1.263 Sulphasalazine 3 vs never taking 2 1.250 1.121 1.393 Sulphasalazine 3 vs never taking 3 1.171 0.937 1.463 Sulphasalazine 4 vs never taking 1 1.136 0.631 2.046 Sulphasalazine 4 vs never taking 2 0.804 0.485 1.334 Sulphasalazine 4 vs never taking 3 1.985 0.922 4.273 Sulphasalazine currently taking vs never taking 1 1.158 0.931 1.440 Sulphasalazine currently taking vs never taking 2 1.004 0.837 1.204 Sulphasalazine currently taking vs never taking 3 0.956 0.660 1.387 Arava (Leflunomide) 3 vs never taking 1 1.116 0.929 1.341 Arava (Leflunomide) 3 vs never taking 2 1.213 1.052 1.399 Arava (Leflunomide) 3 vs never taking 3 1.160 0.876 1.536 Arava (Leflunomide) 4 vs never taking 1 0.752 0.259 2.178 Arava (Leflunomide) 4 vs never taking 2 1.581 0.863 2.898 Arava (Leflunomide) 4 vs never taking 3 0.490 0.062 3.898 Arava (Leflunomide) currently taking vs never 1 1.311 1.065 1.613 taking Arava (Leflunomide) currently taking vs never 2 1.161 0.981 1.374 taking Arava (Leflunomide) currently taking vs never 3 1.006 0.717 1.412 taking Cyclosporin 3 vs never taking 1 1.027 0.854 1.234 Cyclosporin 3 vs never taking 2 1.227 1.071 1.405 Cyclosporin 3 vs never taking 3 1.594 1.229 2.066 Cyclosporin 4 vs never taking 1 0.785 0.418 1.474

Page 349 of 577

Odds Ratio Estimates TB Point 95% Wald Effect Infection Estimate Confidence Limits Cyclosporin 4 vs never taking 2 1.070 0.694 1.648 Cyclosporin 4 vs never taking 3 0.349 0.083 1.461 Cyclosporin currently taking vs never taking 1 1.697 0.876 3.287 Cyclosporin currently taking vs never taking 2 2.840 1.833 4.403 Cyclosporin currently taking vs never taking 3 2.778 1.173 6.577 Prednisolone 3 vs never taking 1 1.392 1.126 1.722 Prednisolone 3 vs never taking 2 1.291 1.095 1.521 Prednisolone 3 vs never taking 3 1.645 1.149 2.357 Prednisolone 4 vs never taking 1 2.190 0.729 6.576 Prednisolone 4 vs never taking 2 1.727 0.740 4.034 Prednisolone 4 vs never taking 3 2.047 0.267 15.680 Prednisolone currently taking vs never taking 1 1.182 0.955 1.462 Prednisolone currently taking vs never taking 2 1.140 0.967 1.343 Prednisolone currently taking vs never taking 3 1.479 1.032 2.118

Page 350 of 577

APPENDIX E: OUTPUT OF SAS FOR BONE MUSCLE JOINT INFECTION

Table E.1- Complete statistics for bone muscle joint infection.

Model Information Data Set WORK.IMPORT2 Response Variable InfBone, Joint and muscle InfBone, Joint and muscle Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table E.2- Observation status for BONE MUSCLE JOINT infection

Number of Observations Read 27711 Number of Observations Used 21506

Page 351 of 577

Table E.3- response value for BONE MUSCLE JOINT infection

Response Profile

Ordered Total Value Bone/Joint/Muscle infection Frequency

1 1 82

2 2 213

3 3 243

4 4 20968

Table E.4- Backward Elimination Procedure for bone muscle joint infection

Class Level Information

Class Value Design Variables

Etanercept 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Adalimumab 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Anakinra 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Page 352 of 577

Class Level Information

Class Value Design Variables

Infliximab 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Rituximab 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Abatacept 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Tocilizumab 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Golimumab 3 1 0 0

currently taking 0 1 0

b never taking 0 0 1

Methotrexate 1 1 0 0 0

2 0 1 0 0

3 0 0 1 0

4 0 0 0 1

Page 353 of 577

Class Level Information

Class Value Design Variables

Certolizumab 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Methotrexate (plus Folic acid) currently taking 1 0

b never taking 0 1

Hydroxychloroquine 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Sulphasalazine 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Arava (Leflunomide) 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Azathioprine 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Page 354 of 577

Class Level Information

Class Value Design Variables

Cyclosporin 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Prednisolone 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

IM Gold 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Penicillamine 3 1 0 0 0

4 0 1 0 0

currently taking 0 0 1 0

b never taking 0 0 0 1

Step 0. The following effects were entered:

Intercept Etanercept Adalimumab Anakinra Infliximab Rituximab Abatacept Tocilizumab Golimumab Methotrexate Certolizumab Methotrexate(plus Folic acid) Hydroxychloroquine Sulphasalazine Arava (Leflunomide) Azathioprine Cyclosporin Prednisolone IM Gold Penicillamine

Page 355 of 577

Table E.5- Model Convergence status for BONE MUSCLE JOINT infection

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table E.6- Model Fit statistics for BONE MUSCLE JOINT infection

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 6126.457 6158.903 SC 6150.385 7474.957 -2 Log L 6120.457 5828.903

Table E.7- Testing null hypothesis for BONE MUSCLE JOINT infection

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 291.5546 162 <.0001 Score 316.8699 162 <.0001 Wald 282.4093 162 <.0001

Page 356 of 577

Table E.8- Model Fit statistics for removing covariant step 1

Step 1. Effect Certolizumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table E.9- Model Fit statistics for removing covariant step 1

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 6126.457 6146.339 SC 6150.385 7390.609 -2 Log L 6120.457 5834.339

Table E.10- Testing Null hypothesis after removing covariant step 1

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 286.1178 153 <.0001 Score 313.7027 153 <.0001 Wald 282.2026 153 <.0001

Page 357 of 577

Table E.11- Residual removing covariant step 1

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 3.7370 9 0.9279

Table E.12- Model Fit statistics for removing covariant step 2

Step 2. Effect Azathioprine is removed:

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table E.13- Model Fit statistics after removing covariant step 2

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 6126.457 6132.503 SC 6150.385 7304.988 -2 Log L 6120.457 5838.503

Table E.14- Testing Null hypothesis after removing covariant step 2

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 281.9542 144 <.0001

Page 358 of 577

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Score 307.4486 144 <.0001 Wald 277.8842 144 <.0001

Table E.15- Residual removing covariant step 2

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 7.6575 18 0.9833

Table E.16- Model Fit statistics for removing covariant step 3

Step 3. Effect Anakinra is removed

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table E.17- Model Fit statistics after removing covariant step 3

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 6126.457 6119.824 SC 6150.385 7220.524 -2 Log L 6120.457 5843.824

Table E.18- Testing Null hypothesis after removing covariant step 3

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 276.6335 135 <.0001 Score 299.1248 135 <.0001 Wald 271.6162 135 <.0001

Page 359 of 577

Table E.19- Residual removing covariant step 3

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 12.9434 27 0.9896

Table E.20- Model Fit statistics for removing covariant step 4

Step 4. Effect IM Golimumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Page 360 of 577

Table E.21- Model Fit statistics after removing covariant step 4

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 6126.457 6111.659 SC 6150.385 7164.502 -2 Log L 6120.457 5847.659

Table E.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 272.7986 129 <.0001 Score 295.4929 129 <.0001 Wald 268.3543 129 <.0001

Table E.23- Residual removing covariant step 4

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 16.2276 33 0.9937

Page 361 of 577

Table E.24- Model Fit statistics for removing covariant step 5 Step 5. Effect Tocilizumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table E.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 6126.457 6103.307 SC 6150.385 7084.366 -2 Log L 6120.457 5857.307

Table E.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 263.1498 120 <.0001 Score 289.4733 120 <.0001 Wald 265.8322 120 <.0001

Page 362 of 577

Table E.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 24.2774 42 0.9870

Table E.28- Model Fit statistics for removing covariant step 6

Step 6. Effect Cyclosporin is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table E.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 6126.457 6090.488 SC 6150.385 6999.762 -2 Log L 6120.457 5862.488

Page 363 of 577

Table E.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 257.9689 111 <.0001 Score 281.3060 111 <.0001 Wald 258.4273 111 <.0001

Table E.31- Residual removing covariant step 7

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 31.0446 51 0.9877

Note: No (additional) effects met the 0.05 significance level for removal from the model.

Table E.32- Model Fit statistics for removing covariant step 7

Step 7. Effect Methotrexate is removed:

Model Convergence Status

Quasi-complete separation of data points detected.

Warning: The maximum likelihood estimate may not exist. Warning: The LOGISTIC procedure continues in spite of the above warning. Results shown are based on the last maximum likelihood iteration. Validity of the model fit is questionable.

Page 364 of 577

The LOGISTIC Procedure WARNING: The validity of the model fit is questionable. Saturday, 14 December 2019 07:51:16 PM 365 Table E.33- Model Fit statistics after removing covariant

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 6126.457 6081.126

SC 6150.385 6918.615

-2 Log L 6120.457 5871.126

Table E.34- Testing Null hypothesis after removing covariant

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 249.3310 102 <.0001

Score 269.9568 102 <.0001

Wald 249.6230 102 <.0001

Table E.35- Residual removing covariant

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

40.0344 60 0.9780

Table E.36- Model Fit statistics for removing covariant

The LOGISTIC Procedure WARNING: The validity of the model fit is questionable. Saturday, 14 December 2019 07:51:16 PM 366 Step 8. Effect Rituximab is removed:

Model Convergence Status

Quasi-complete separation of data points detected.

Table E.37- Model Fit statistics after removing covariant

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 6126.457 6074.207

SC 6150.385 6839.912

-2 Log L 6120.457 5882.207

Table E.38- Testing Null hypothesis after removing covariant

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 238.2500 93 <.0001

Score 258.7575 93 <.0001

Wald 237.4385 93 <.0001

Table E.39- Residual removing covariant

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

53.0612 69 0.9222

Table E.40- Model Fit statistics for removing covariant Step 9. Effect Abatacept is removed:

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.

367

Table E.41- Model Fit statistics after removing covariant

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 6126.457 6069.572

SC 6150.385 6763.492

-2 Log L 6120.457 5895.572

Table E.42- Testing Null hypothesis after removing covariant

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 224.8848 84 <.0001

Score 244.5347 84 <.0001

Wald 226.2461 84 <.0001

Table E.43- Residual removing covariant

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

65.8098 78 0.8359

Table E.44- Model Fit statistics for removing covariant

Step 10. Effect Sulphasalazine is removed:

368

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.

Table E.45- Model Fit statistics after removing covariant

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 6126.457 6067.034

SC 6150.385 6689.169

-2 Log L 6120.457 5911.034

Table E.46- Testing Null hypothesis after removing covariant

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 209.4231 75 <.0001

Score 229.3403 75 <.0001

Wald 210.2999 75 <.0001

Table E.47- Residual removing covariant

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

81.1596 87 0.6562

Table E.48- Model Fit statistics for removing covariant

369

Step 11. Effect Etanercept is removed:

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.

Table E.49- Model Fit statistics after removing covariant

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 6126.457 6065.927

SC 6150.385 6616.277

-2 Log L 6120.457 5927.927

Table E.50- Testing Null hypothesis after removing covariant

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 192.5302 66 <.0001

Score 213.8622 66 <.0001

Wald 195.0069 66 <.0001

Table E.51- Residual removing covariant

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

96.0334 96 0.4798

Table E.52- Model Fit statistics for removing covariant

370

Step 12. Effect Adalimumab is removed:

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.

Table E.53- Model Fit statistics after removing covariant

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 6126.457 6065.506

SC 6150.385 6544.071

-2 Log L 6120.457 5945.506

Table E.54- Testing Null hypothesis after removing covariant

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 174.9515 57 <.0001

Score 191.8185 57 <.0001

Wald 176.7190 57 <.0001

Table E.55- Residual removing covariant

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

110.9740 105 0.3262

371

Table E.56- Model Fit statistics for removing covariant

Step 13. Effect IM Gold is removed:

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.

Table E.57- Model Fit statistics after removing covariant

Model Fit Statistics

Criterion Intercept Only Intercept and Covariates

AIC 6126.457 6063.434

SC 6150.385 6470.215

-2 Log L 6120.457 5961.434

Table E.58- Testing Null hypothesis after removing covariant

Testing Global Null Hypothesis: BETA=0

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 159.0230 48 <.0001

Score 174.2918 48 <.0001

Wald 161.1437 48 <.0001

Table E.59- Residual removing covariant

Residual Chi-Square Test

Chi-Square DF Pr > ChiSq

129.3306 114 0.1546

372

Note: No (additional) effects met the 0.05 significance level for removal from the model.

Table E.60- Summary of backward elimination in bone muscle joint

373

Summary of Backward Elimination

Effect Number Wald Variable Step Removed DF In Chi-Square Pr > ChiSq Label

1 Certolizumab 9 18 1.6031 0.9963 Certolizumab

2 Azathioprine 9 17 3.8457 0.9213 Azathioprine

3 Anakinra 9 16 4.3211 0.8890

4 Golimumab 6 15 3.0321 0.8048 Golimumab

5 Tocilizumab 9 14 5.4188 0.7964 Tocilizumab

6 Cyclosporin 9 13 6.0931 0.7306 Cyclosporin

7 Methotrexate 9 12 8.0079 0.5334 Methotrexate

8 Rituximab 9 11 10.8068 0.2892 Rituximab

9 Abatacept 9 10 11.4004 0.2493 Abatacept

10 Sulphasalazine 9 9 14.0763 0.1196 Sulphasalazine

11 Etanercept 9 8 15.3543 0.0817

12 Adalimumab 9 7 15.0322 0.0901

13 IM Gold 9 6 16.1101 0.0646 IM Gold

Table E.61- Type 3 analysis of effects in BONE MUSCLE JOINT

Type 3 Analysis of Effects

Wald Effect DF Chi-Square Pr > ChiSq

Infliximab 9 24.8305 0.0032

Methotrexate (plus Folic acid) 3 8.3854 0.0387

Hydroxychloroquine 9 25.4841 0.0025

374

Type 3 Analysis of Effects

Wald Effect DF Chi-Square Pr > ChiSq

Arava (Leflunomide) 9 35.2574 <.0001

Prednisolone 9 33.2572 0.0001

Penicillamine 9 28.5823 0.0008

Table E.62- Analysis of maximum likelihood estimates in BONE MUSCLE JOINT

Analysis of Maximum Likelihood Estimates

Standar Wald Bone/Joint/Muscl D Estimat d Chi- Pr > ChiS Parameter e infection F e Error Square q

Intercept Mild 1 -5.1566 0.3087 278.964 <.0001 7

Intercept Mod 1 -4.9462 0.2433 413.256 <.0001 7

Intercept Severe 1 -5.3447 0.2712 388.337 <.0001 9

Arava 3 Mild 1 -0.1188 0.2888 0.1693 0.6808 (Leflunomide)

Arava 3 Mod 1 -0.1982 0.1832 1.1705 0.2793 (Leflunomide)

Arava 3 Severe 1 -0.0531 0.1837 0.0835 0.7726 (Leflunomide)

Arava 4 Mild 1 2.0608 0.8039 6.5709 0.0104 (Leflunomide)

Arava 4 Mod 1 0.1281 0.7933 0.0261 0.8718 (Leflunomide)

375

Analysis of Maximum Likelihood Estimates

Standar Wald Bone/Joint/Muscl D Estimat d Chi- Pr > ChiS Parameter e infection F e Error Square q

Arava 4 Severe 1 0.8231 0.6687 1.5153 0.2183 (Leflunomide)

Arava currentl Mild 1 0.1898 0.3336 0.3236 0.5694 (Leflunomide) y taking

Arava currentl Mod 1 0.1851 0.2077 0.7935 0.3730 (Leflunomide) y taking

Arava currentl Severe 1 0.6292 0.1968 10.2160 0.0014 (Leflunomide) y taking

Arava b never Mild 0 0 . . . (Leflunomide) taking

Arava b never Mod 0 0 . . . (Leflunomide) taking

Arava b never Severe 0 0 . . . (Leflunomide) taking

Hydroxychloroqui 3 Mild 1 -0.2910 0.2544 1.3078 0.2528 ne

Hydroxychloroqui 3 Mod 1 0.00465 0.1644 0.0008 0.9774 ne

Hydroxychloroqui 3 Severe 1 -0.5410 0.1560 12.0268 0.0005 ne

Hydroxychloroqui 4 Mild 1 - 412.2 0.0008 0.9775 ne 11.6261

Hydroxychloroqui 4 Mod 1 0.8595 0.5838 2.1676 0.1409 ne

Hydroxychloroqui 4 Severe 1 0.6666 0.5145 1.6791 0.1950 ne

376

Analysis of Maximum Likelihood Estimates

Standar Wald Bone/Joint/Muscl D Estimat d Chi- Pr > ChiS Parameter e infection F e Error Square q

Hydroxychloroqui currentl Mild 1 -0.2784 0.3205 0.7543 0.3851 ne y taking

Hydroxychloroqui currentl Mod 1 0.2671 0.1893 1.9918 0.1582 ne y taking

Hydroxychloroqui currentl Severe 1 0.1313 0.1708 0.5916 0.4418 ne y taking

Hydroxychloroqui b never Mild 0 0 . . . ne taking

Hydroxychloroqui b never Mod 0 0 . . . ne taking

Hydroxychloroqui b never Severe 0 0 . . . ne taking

Infliximab 3 Mild 1 0.2744 0.4349 0.3981 0.5281

Infliximab 3 Mod 1 0.0665 0.2839 0.0548 0.8149

Infliximab 3 Severe 1 0.8912 0.1977 20.3197 <.0001

Infliximab 4 Mild 1 - 435.3 0.0007 0.9792 11.3660

Infliximab 4 Mod 1 0.7831 0.5782 1.8342 0.1756

Infliximab 4 Severe 1 0.4180 0.6459 0.4189 0.5175

Infliximab currentl Mild 1 -0.2314 0.7289 0.1007 0.7510 y taking

Infliximab currentl Mod 1 -0.0882 0.4578 0.0372 0.8471 y taking

Infliximab currentl Severe 1 0.5992 0.3308 3.2819 0.0700 y taking

377

Analysis of Maximum Likelihood Estimates

Standar Wald Bone/Joint/Muscl D Estimat d Chi- Pr > ChiS Parameter e infection F e Error Square q

Infliximab b never Mild 0 0 . . . taking

Infliximab b never Mod 0 0 . . . taking

Infliximab b never Severe 0 0 . . . taking

Methotrexate (plus currentl Mild 1 -0.4576 0.2957 2.3950 0.1217 Folic acid) y taking

Methotrexate (plus currentl Mod 1 -0.4093 0.1788 5.2389 0.0221 Folic acid) y taking

Methotrexate (plus currentl Severe 1 0.1239 0.1458 0.7217 0.3956 Folic acid) y taking

Methotrexate (plus b never Mild 0 0 . . . Folic acid) taking

Methotrexate (plus b never Mod 0 0 . . . Folic acid) taking

Methotrexate (plus b never Severe 0 0 . . . Folic acid) taking

Penicillamine 3 Mild 1 0.0335 0.3849 0.0076 0.9306

Penicillamine 3 Mod 1 0.3480 0.2056 2.8650 0.0905

Penicillamine 3 Severe 1 0.7969 0.1692 22.1732 <.0001

Penicillamine 4 Mild 1 -0.2000 1.0526 0.0361 0.8493

Penicillamine 4 Mod 1 -0.2949 0.6090 0.2346 0.6281

Penicillamine 4 Severe 1 -0.3479 0.5814 0.3581 0.5496

378

Analysis of Maximum Likelihood Estimates

Standar Wald Bone/Joint/Muscl D Estimat d Chi- Pr > ChiS Parameter e infection F e Error Square q

Penicillamine currentl Mild 1 1.6260 1.0242 2.5204 0.1124 y taking

Penicillamine currentl Mod 1 - 913.3 0.0002 0.9886 y taking 13.0777

Penicillamine currentl Severe 1 - 847.3 0.0002 0.9878 y taking 12.9395

Penicillamine b never Mild 0 0 . . . taking

Penicillamine b never Mod 0 0 . . . taking

Penicillamine b never Severe 0 0 . . . taking

Prednisolone 3 Mild 1 -0.4588 0.3457 1.7618 0.1844

Prednisolone 3 Mod 1 0.2064 0.2531 0.6652 0.4147

Prednisolone 3 Severe 1 0.4961 0.2745 3.2653 0.0708

Prednisolone 4 Mild 1 - 537.3 0.0005 0.9826 11.7508

Prednisolone 4 Mod 1 0.7436 1.0554 0.4963 0.4811

Prednisolone 4 Severe 1 0.6734 1.0753 0.3921 0.5312

Prednisolone currentl Mild 1 0.0363 0.3124 0.0135 0.9075 y taking

Prednisolone currentl Mod 1 0.6298 0.2394 6.9233 0.0085 y taking

Prednisolone currentl Severe 1 0.9273 0.2630 12.4305 0.0004 y taking

379

Analysis of Maximum Likelihood Estimates

Standar Wald Bone/Joint/Muscl D Estimat d Chi- Pr > ChiS Parameter e infection F e Error Square q

Prednisolone b never Mild 0 0 . . . taking

Prednisolone b never Mod 0 0 . . . taking

Prednisolone b never Severe 0 0 . . . taking

Table E.63- Odds ratio estimates in BONE MUSCLE JOINT

Odds Ratio Estimates

Bone/Joint/Muscle Point 95% Wald Effect infection Estimate Confidence Limits

Infliximab 3 Versus 1 1.316 0.561 3.086 never taking

Infliximab 3 Versus 2 1.069 0.613 1.864 never taking

Infliximab 3 Versus 3 2.438 1.655 3.592 never taking

Infliximab 4 Versus 1 <0.001 <0.001 >999.999 never taking

Infliximab 4 Versus 2 2.188 0.705 6.796 never taking

Infliximab 4 Versus 3 1.519 0.428 5.387 never taking

Infliximab currently taking 1 0.793 0.190 3.311 Versus never taking

380

Odds Ratio Estimates

Bone/Joint/Muscle Point 95% Wald Effect infection Estimate Confidence Limits

Infliximab currently taking 2 0.916 0.373 2.246 Versus never taking

Infliximab currently taking 3 1.821 0.952 3.482 Versus never taking

Methotrexate (plus Folic acid) 1 0.633 0.354 1.130 currently taking Versus never taking

Methotrexate (plus Folic acid) 2 0.664 0.468 0.943 currently taking Versus never taking

Methotrexate (plus Folic acid) 3 1.132 0.851 1.506 currently taking Versus never taking

Hydroxychloroquine 3 1 0.748 0.454 1.231 Versus never taking

Hydroxychloroquine 3 2 1.005 0.728 1.387 Versus never taking

Hydroxychloroquine 3 3 0.582 0.429 0.790 Versus never taking

Hydroxychloroquine 4 1 <0.001 <0.001 >999.999 Versus never taking

Hydroxychloroquine 4 2 2.362 0.752 7.417 Versus never taking

Hydroxychloroquine 4 3 1.948 0.711 5.339 Versus never taking

Hydroxychloroquine currently 1 0.757 0.404 1.419 taking Versus never taking

381

Odds Ratio Estimates

Bone/Joint/Muscle Point 95% Wald Effect infection Estimate Confidence Limits

Hydroxychloroquine currently 2 1.306 0.901 1.893 taking Versus never taking

Hydroxychloroquine currently 3 1.140 0.816 1.594 taking Versus never taking

Arava (Leflunomide) 3 1 0.888 0.504 1.564 Versus never taking

Arava (Leflunomide) 3 2 0.820 0.573 1.175 Versus never taking

Arava (Leflunomide) 3 3 0.948 0.662 1.359 Versus never taking

Arava (Leflunomide) 4 1 7.852 1.624 37.956 Versus never taking

Arava (Leflunomide) 4 2 1.137 0.240 5.381 Versus never taking

Arava (Leflunomide) 4 3 2.278 0.614 8.446 Versus never taking

Arava (Leflunomide) 1 1.209 0.629 2.325 currently taking Versus never taking

Arava (Leflunomide) 2 1.203 0.801 1.808 currently taking Versus never taking

Arava (Leflunomide) 3 1.876 1.276 2.759 currently taking Versus never taking

Prednisolone 3 1 0.632 0.321 1.244 Versus never taking

382

Odds Ratio Estimates

Bone/Joint/Muscle Point 95% Wald Effect infection Estimate Confidence Limits

Prednisolone 3 2 1.229 0.749 2.019 Versus never taking

Prednisolone 3 3 1.642 0.959 2.813 Versus never taking

Prednisolone 4 1 <0.001 <0.001 >999.999 Versus never taking

Prednisolone 4 2 2.103 0.266 16.646 Versus never taking

Prednisolone 4 3 1.961 0.238 16.136 Versus never taking

Prednisolone currently taking 1 1.037 0.562 1.913 Versus never taking

Prednisolone currently taking 2 1.877 1.174 3.001 Versus never taking

Prednisolone currently taking 3 2.528 1.510 4.232 Versus never taking

Penicillamine 3 1 1.034 0.486 2.199 Versus never taking

Penicillamine 3 2 1.416 0.947 2.119 Versus never taking

Penicillamine 3 3 2.219 1.592 3.092 Versus never taking

Penicillamine 4 1 0.819 0.104 6.444 Versus never taking

Penicillamine 4 2 0.745 0.226 2.456 Versus never taking

383

Odds Ratio Estimates

Bone/Joint/Muscle Point 95% Wald Effect infection Estimate Confidence Limits

Penicillamine 4 3 0.706 0.226 2.207 Versus never taking

Penicillamine currently taking 1 5.084 0.683 37.844 Versus never taking

Penicillamine currently taking 2 <0.001 <0.001 >999.999 Versus never taking

Penicillamine currently taking 3 <0.001 <0.001 >999.999 Versus never taking

384

APPENDIX F: OUTPUT OF SAS FOR BLOOD INFECTION

Table F.1- Complete statistics for blood infection.

Model Information Data Set WORK.IMPORT2 Response Variable InfBlood InfBlood Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table F.2- Observation status for BLOOD infection

Number of Observations Read 27711 Number of Observations Used 21506

Table F.3- response value for BLOOD infection

Response Profile Ordered Total Value InfBlood Frequency 1 1 21 2 2 70 3 3 111 4 4 21304 0 .

Logits modelled use InfBlood='4' as the reference category. Note: 6205 observations were deleted due to missing values for the response or explanatory variables. 385

Note: 1 response level was deleted due to missing or invalid values for its explanatory, frequency, or weight variables.

Table F.4- Backward Elimination Procedure for BLOOD infection

Backward Elimination Procedure Class Level Information Class Value Design Variables Etanercept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Adalimumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Anakinra 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Infliximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Rituximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Abatacept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

386

Tocilizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Golimumab 3 1 0 0 currently taking 0 1 0 never taking 0 0 1

Certolizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Folic Acid currently taking 1 0 never taking 0 1

Hydroxychloroquine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Sulphasalazine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Arava (Leflunomide) 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Azathioprine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Cyclosporin 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

387

Prednisolone 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

IM Gold injection 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Penicillamine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Step 0. The following effects were entered:

Intercept Etanercept Adalimumab Anakinra Infliximab Rituximab Abatacept Tocilizumab Golimumab Certolizumab Folic Acid Hydroxychloroquine Sulphasalazine Arava (Leflunomide) Azathioprine Cyclosporin Prednisolone IM Gold injection Penicillamine

Table F.5- Model Convergence status for BLOOD infection

Model Convergence Status Quasi-complete separation of data points detected.

Table F.6- Model Fit statistics for BLOOD infection

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2718.376

388

Model Fit Statistics Criterion Intercept Only Intercept and Covariates SC 2694.189 3962.646 -2 Log L 2664.261 2406.376

Table F.7- Testing null hypothesis for BLOOD infection

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 257.8845 153 <.0001 Score 284.8579 153 <.0001 Wald 233.2076 153 <.0001

389

Table F.8- Model Fit statistics for removing covariant step 1

Step 1. Effect Anakinra is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table F.9- Model Fit statistics for removing covariant step 1

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2706.095 SC 2694.189 3878.580 -2 Log L 2664.261 2412.095

Table F.10- Testing Null hypothesis after removing covariant step 1

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 252.1657 144 <.0001 Score 282.2859 144 <.0001 Wald 231.6736 144 <.0001

390

Table F.11- Residual removing covariant step 1

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 3.1885 9 0.9563

Table F.12- Model Fit statistics for removing covariant step 2

Step 2. Effect Certolizumab is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table F.13- Model Fit statistics after removing covariant step 2

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2701.408 SC 2694.189 3802.108 -2 Log L 2664.261 2425.408

391

Table F.14- Testing Null hypothesis after removing covariant step 2

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 238.8526 135 <.0001 Score 268.2359 135 <.0001 Wald 222.0689 135 <.0001

Table F.15- Residual removing covariant step 2

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 11.6827 18 0.8632

Table F.16- Model Fit statistics for removing covariant step 3

Step 3. Effect Infliximab is removed: Model Convergence Status Quasi-complete separation of data points detected.

392

Table F.17- Model Fit statistics after removing covariant step 3

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2688.609 SC 2694.189 3717.524 -2 Log L 2664.261 2430.609

Table F.18- Testing Null hypothesis after removing covariant step 3

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 233.6524 126 <.0001 Score 261.5980 126 <.0001 Wald 217.2921 126 <.0001

Table F.19- Residual removing covariant step 3

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 16.0867 27 0.9513

393

Table F.20- Model Fit statistics for removing covariant step 4

Step 4. Effect Rituximab is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table F.21- Model Fit statistics after removing covariant step 4

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2678.948 SC 2694.189 3636.079 -2 Log L 2664.261 2438.948

Table F.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 225.3126 117 <.0001 Score 253.3066 117 <.0001 Wald 210.1667 117 <.0001

394

Table F.23- Residual removing covariant step 4

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 23.2706 36 0.9500

Table F.24- Model Fit statistics for removing covariant step 5 Step 5. Effect Arava (Leflunomide) is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table F.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2666.028 SC 2694.189 3551.373 -2 Log L 2664.261 2444.028

Table F.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 220.2334 108 <.0001 Score 246.5724 108 <.0001 Wald 203.9904 108 <.0001

395

Table F.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 28.7370 45 0.9717

Table F.28- Model Fit statistics for removing covariant step 6

Step 6. Effect Penicillamine is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table F.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2658.846 SC 2694.189 3472.407 -2 Log L 2664.261 2454.846

Table F.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 209.4150 99 <.0001 Score 235.7326 99 <.0001 Wald 196.5912 99 <.0001

396

Table F.31- Residual removing covariant step 6

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 37.4641 54 0.9577

Table F.32- Model Fit statistics for removing covariant step 7

Step 7. Effect Golimumab is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table F.33- Model Fit statistics after removing covariant step 7

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2655.632 SC 2694.189 3421.337 -2 Log L 2664.261 2463.632 Table F.34- Testing Null hypothesis after removing covariant step 7

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 200.6285 93 <.0001 Score 228.1270 93 <.0001 Wald 193.4477 93 <.0001

397

Table F.35- Residual removing covariant step 7

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 44.5830 60 0.9316

Table F.36- Model Fit statistics for removing covariant step 8 Step 8. Effect Cyclosporin is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table F.37- Model Fit statistics after removing covariant step 8

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2653.141 SC 2694.189 3347.061 -2 Log L 2664.261 2479.141

Table F.38- Testing Null hypothesis after removing covariant step 8

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 185.1200 84 <.0001 Score 211.1646 84 <.0001 Wald 178.6733 84 <.0001

398

Table F.39- Residual removing covariant step 8

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 59.2165 69 0.7934

Table F.40- Model Fit statistics for removing covariant step 9

Step 9. Effect Sulphasalazine is removed: Model Convergence Status Quasi-complete separation of data points detected.

399

Table F.41- Model Fit statistics after removing covariant step 9

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2642.193 SC 2694.189 3264.328 -2 Log L 2664.261 2486.193

Table F.42- Testing Null hypothesis after removing covariant step 9

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 178.0675 75 <.0001 Score 200.9988 75 <.0001 Wald 169.6897 75 <.0001

Table F.43- Residual removing covariant step 9

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 67.3223 78 0.8005

Table F.44- Model Fit statistics for removing covariant step 10 Step 10. Effect Azathioprine is removed: Model Convergence Status Quasi-complete separation of data points detected.

400

Table F.45- Model Fit statistics after removing covariant step 10

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2635.960 SC 2694.189 3186.310 -2 Log L 2664.261 2497.960

Table F.46- Testing Null hypothesis after removing covariant step 10

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 166.3009 66 <.0001 Score 184.7483 66 <.0001 Wald 158.3863 66 <.0001

Table F.47- Residual removing covariant step 10

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 81.5794 87 0.6439

Table F.48- Model Fit statistics for removing covariant step 11

Step 11. Effect Abatacept is removed: Model Convergence Status Quasi-complete separation of data points detected.

401

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2637.362 SC 2694.189 3115.928 -2 Log L 2664.261 2517.362

Table F.50- Testing Null hypothesis after removing covariant step 11

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 146.8987 57 <.0001 Score 164.0564 57 <.0001 Wald 139.9299 57 <.0001

Table F.51- Residual removing covariant step 11

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 103.8113 96 0.2753

Table F.52- Model Fit statistics for removing covariant step 12

Step 12. Effect Tocilizumab is removed: Model Convergence Status Quasi-complete separation of data points detected.

402

Table F.53- Model Fit statistics for removing covariant step 12

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2632.264 SC 2694.189 3039.045 -2 Log L 2664.261 2530.264

Table F.54- Testing Null hypothesis after removing covariant step 12

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 133.9966 48 <.0001 Score 144.2262 48 <.0001 Wald 125.1312 48 <.0001

Table F.55- Residual removing covariant step 12

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 127.3818 105 0.0679

403

Table F.56- Model Fit statistics for removing covariant step 13

Step 13. Effect Folic Acid is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table F.57- Model Fit statistics after removing covariant step 13

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2630.847 SC 2694.189 3013.699 -2 Log L 2664.261 2534.847

Table F.58- Testing Null hypothesis after removing covariant step 13

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 129.4141 45 <.0001 Score 138.7698 45 <.0001 Wald 120.0137 45 <.0001

Table F.59- Residual removing covariant step 13

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 133.2998 108 0.0497

404

Table F.60- Model Fit statistics for removing covariant step 14

Step 14. Effect IM Gold injection is removed:

Model Convergence Status Quasi-complete separation of data points detected.

Table F.61- Model Fit statistics after removing covariant step 14

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2627.554 SC 2694.189 2938.621 -2 Log L 2664.261 2549.554

Table F.62- Testing Null hypothesis after removing covariant step 14

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 114.7072 36 <.0001 Score 122.2303 36 <.0001 Wald 104.6456 36 <.0001

Table F.63- Residual removing covariant step 14

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 151.0167 117 0.0187

405

Table F.64- Model Fit statistics for removing covariant step 15

Step 15. Effect Adalimumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table F.65- Model Fit statistics after removing covariant step 15

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2629.777 SC 2694.189 2869.059 -2 Log L 2664.261 2569.777

Table F.66- Testing Null hypothesis after removing covariant step 15

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 94.4841 27 <.0001 Score 101.5949 27 <.0001 Wald 85.0867 27 <.0001

Table F.67- Residual removing covariant step 15

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 175.1574 126 0.0025

Table F.68- Model Fit statistics for removing covariant step 16

Step 16. Effect Etanercept is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

406

Table F.68- Model Fit statistics after removing covariant step 16

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 2670.261 2628.182 SC 2694.189 2795.680 -2 Log L 2664.261 2586.182

Table F.69- Testing Null hypothesis after removing covariant step 16

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 78.0792 18 <.0001 Score 84.6206 18 <.0001 Wald 69.6515 18 <.0001

Table F.70- Residual removing covariant step 16

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 191.7886 135 0.0010

Note: No (additional) effects met the 0.05 significance level for removal from the model.

Table F.71- Summary of backward elimination in BLOOD

Summary of Backward Elimination Effect Number Wald Variable Step Removed DF In Chi-Square Pr > ChiSq Label 1 Anakinra 9 17 0.2808 1.0000 2 Certolizumab 9 16 1.6755 0.9956 Certolizumab 3 Infliximab 9 15 3.5451 0.9387 4 Rituximab 9 14 4.8713 0.8454 Rituximab 407

Summary of Backward Elimination Effect Number Wald Variable Step Removed DF In Chi-Square Pr > ChiSq Label 5 Arava (Leflunomide) 9 13 5.3385 0.8039 Arava (Leflunomide) 6 Penicillamine 9 12 5.6548 0.7739 Penicillamine 7 Golimumab 6 11 3.5097 0.7427 Golimumab 8 Cyclosporin 9 10 7.0960 0.6271 Cyclosporin 9 Sulphasalazine 9 9 7.1260 0.6240 Sulphasalazine 10 Azathioprine 9 8 9.5235 0.3904 Azathioprine 11 Abatacept 9 7 12.2368 0.2003 Abatacept 12 Tocilizumab 9 6 12.2063 0.2019 Tocilizumab 13 Folic Acid 3 5 4.8032 0.1868 Folic Acid 14 IM Gold injection 9 4 15.5529 0.0768 IM Gold injection 15 Adalimumab 9 3 16.5298 0.0566 16 Etanercept 9 2 14.0456 0.1207

Table F.71- Type 3 analysis of effects in BLOOD

Type 3 Analysis of Effects Wald Effect DF Chi-Square Pr > ChiSq Hydroxychloroquine 9 18.1008 0.0340 Prednisolone 9 49.5445 <.0001

Table F.72- Analysis of maximum likelihood estimates in BLOOD

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfBlood DF Estimate Error Chi-Square Pr > ChiSq Intercept Mild 1 -5.9671 0.4591 168.9262 <.0001 Intercept Mod 1 -5.4061 0.3070 309.9955 <.0001 Intercept Severe 1 -6.2875 0.4552 190.7863 <.0001 Hydroxychloroquine 3 Mild 1 -1.1890 0.5196 5.2369 0.0221 Hydroxychloroquine 3 Mod 1 -0.4748 0.2724 3.0385 0.0813

408

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfBlood DF Estimate Error Chi-Square Pr > ChiSq Hydroxychloroquine 3 Severe 1 -0.1629 0.2022 0.6495 0.4203 Hydroxychloroquine 4 Mild 1 -11.8338 746.4 0.0003 0.9874 Hydroxychloroquine 4 Mod 1 0.1219 1.0413 0.0137 0.9068 Hydroxychloroquine 4 Severe 1 0.5880 0.7276 0.6530 0.4190 Hydroxychloroquine currently taking Mild 1 -1.9353 1.0331 3.5095 0.0610 Hydroxychloroquine currently taking Mod 1 -0.2859 0.3367 0.7211 0.3958 Hydroxychloroquine currently taking Severe 1 -0.8566 0.3478 6.0675 0.0138 Hydroxychloroquine never taking Mild 0 0 . . . Hydroxychloroquine never taking Mod 0 0 . . . Hydroxychloroquine never taking Severe 0 0 . . . Prednisolone 3 Mild 1 -0.4482 0.5889 0.5791 0.4467 Prednisolone 3 Mod 1 -0.8646 0.4105 4.4356 0.0352 Prednisolone 3 Severe 1 0.6384 0.4932 1.6756 0.1955 Prednisolone 4 Mild 1 -11.7277 1195.8 0.0001 0.9922 Prednisolone 4 Mod 1 2.1498 0.7901 7.4034 0.0065 Prednisolone 4 Severe 1 -10.3498 552.6 0.0004 0.9851 Prednisolone currently taking Mild 1 -0.3973 0.5607 0.5020 0.4786 Prednisolone currently taking Mod 1 0.2259 0.3281 0.4738 0.4912 Prednisolone currently taking Severe 1 1.6700 0.4622 13.0559 0.0003 Prednisolone never taking Mild 0 0 . . . Prednisolone never taking Mod 0 0 . . . Prednisolone never taking Severe 0 0 . . .

Table F.73- Odds ratio estimates in BLOOD

Odds Ratio Estimates 95% Wald Effect InfBlood Point Estimate Confidence Limits Hydroxychloroquine 3 vs never taking Mild 0.305 0.110 0.843 Hydroxychloroquine 3 vs never taking Mod 0.622 0.365 1.061 Hydroxychloroquine 3 vs never taking Severe 0.850 0.572 1.263 Hydroxychloroquine 4 vs never taking Mild <0.001 <0.001 >999.999 409

Odds Ratio Estimates 95% Wald Effect InfBlood Point Estimate Confidence Limits Hydroxychloroquine 4 vs never taking Mod 1.130 0.147 8.695 Hydroxychloroquine 4 vs never taking Severe 1.800 0.433 7.493 Hydroxychloroquine currently taking vs never taking Mild 0.144 0.019 1.094 Hydroxychloroquine currently taking vs never taking Mod 0.751 0.388 1.453 Hydroxychloroquine currently taking vs never taking Severe 0.425 0.215 0.839 Prednisolone 3 vs never taking Mild 0.639 0.201 2.026 Prednisolone 3 vs never taking Mod 0.421 0.188 0.942 Prednisolone 3 vs never taking Severe 1.893 0.720 4.977 Prednisolone 4 vs never taking Mild <0.001 <0.001 >999.999 Prednisolone 4 vs never taking Mod 8.583 1.824 40.380 Prednisolone 4 vs never taking Severe <0.001 <0.001 >999.999 Prednisolone currently taking vs never taking Mild 0.672 0.224 2.017 Prednisolone currently taking vs never taking Mod 1.253 0.659 2.385 Prednisolone currently taking vs never taking Severe 5.312 2.147 13.142

410

APPENDIX G: OUTPUT OF SAS FOR GIT INFECTION

Table G.1- complete statistics for GIT infection

Model Information Data Set WORK.IMPORT2 Response Variable InfGit InfGit Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table G.2- Observation status for GIT infection

Number of Observations Read 27711 Number of Observations Used 21506

Table G.3- response value for GIT infection

Response Profile Ordered Total Value InfGit Frequency 1 1 118 2 2 241 3 3 155 4 4 20992

411

Logits modelled use InfGit='4' as the reference category.

Note: 6205 observations were deleted due to missing values for the response or explanatory variables.

Table G.4- Backward Elimination Procedure for GIT infection

Backward Elimination Procedure Class Level Information Class Value Design Variables Etanercept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Adalimumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Anakinra 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Infliximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Rituximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Abatacept 3 1 0 0 0 4 0 1 0 0

412

currently taking 0 0 1 0 never taking 0 0 0 1

Tocilizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Golimumab 3 1 0 0 currently taking 0 1 0 never taking 0 0 1

Certolizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Folic Acid currently taking 1 0 never taking 0 1

Hydroxychloroquine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Sulphasalazine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Arava (Leflunomide) 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Azathioprine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Cyclosporin 3 1 0 0 0 4 0 1 0 0

413

currently taking 0 0 1 0 never taking 0 0 0 1

Prednisolone 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

IM Gold injection 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Penicillamine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Step 0. The following effects were entered:

Table G.5- Model Convergence status for GIT infection

Model Convergence Status Quasi-complete separation of data points detected.

414

Table G.6- Model Fit statistics for GIT infection

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 6016.696 SC 5967.947 7260.965 -2 Log L 5938.018 5704.696

Table G.7- Testing null hypothesis for GIT infection

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 233.3227 153 <.0001 Score 267.8657 153 <.0001 Wald 231.4222 153 <.0001

Table G.8- Model Fit statistics for removing covariant step 1

Step 1. Effect Certolizumab is removed: Model Convergence Status Quasi-complete separation of data points detected. Warning: The maximum likelihood estimate may not exist. Warning: The LOGISTIC procedure continues in spite of the above warning. Results shown are based on the last maximum likelihood iteration. Validity of the model fit is questionable

Table G.10- Testing Null hypothesis after removing covariant step 1 Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 225.2887 144 <.0001 Score 262.5650 144 <.0001 Wald 229.5599 144 <.0001

415

Table G.11- Residual removing covariant step 1

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 4.6014 9 0.8676

Table G.12- Model Fit statistics for removing covariant step 2

Step 2. Effect Azathioprine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table G.13- Model Fit statistics after removing covariant step 2 Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5995.805 SC 5967.947 7096.505 -2 Log L 5938.018 5719.805

Table G.14- Testing Null hypothesis after removing covariant step 2

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 218.2134 135 <.0001 Score 256.4742 135 <.0001 Wald 225.2282 135 <.0001

Table G.15- Residual removing covariant step 2

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 10.5472 18 0.9126

416

Table G.16- Model Fit statistics for removing covariant step 3 Step 3. Effect IM Gold injection is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table G.17- Model Fit statistics after removing covariant step 3

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5987.499 SC 5967.947 7016.414 -2 Log L 5938.018 5729.499

Table G.18- Testing Null hypothesis after removing covariant step 3

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 208.5195 126 <.0001 Score 249.5871 126 <.0001 Wald 218.5522 126 <.0001

Table G.19- Residual removing covariant step 3 Residual Chi-Square Test Chi-Square DF Pr > ChiSq 18.1288 27 0.8995

Table G.20- Model Fit statistics for removing covariant step 4

Step 4. Effect Golimumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

417

Table G.21- Model Fit statistics after removing covariant step 4

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5981.489 SC 5967.947 6962.548 -2 Log L 5938.018 5735.489

Table G.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 202.5292 120 <.0001 Score 244.1877 120 <.0001 Wald 215.0861 120 <.0001

Table G.23- Residual removing covariant step 4

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 22.1197 33 0.9249

418

Table G.24- Model Fit statistics for removing covariant step 5

Step 5. Effect Tocilizumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table G.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5968.320 SC 5967.947 6877.594 -2 Log L 5938.018 5740.320

Table G.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 197.6985 111 <.0001 Score 238.1638 111 <.0001 Wald 209.0709 111 <.0001

Table G.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 26.6451 42 0.9688

Table G.28- Model Fit statistics for removing covariant step 6

Step 6. Effect Etanercept is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

419

Table G.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5956.458 SC 5967.947 6793.947 -2 Log L 5938.018 5746.458

Table G.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 191.5606 102 <.0001 Score 232.3845 102 <.0001 Wald 204.2680 102 <.0001

Table G.31- Residual removing covariant step 6

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 32.5729 51 0.9792

Table G.32- Model Fit statistics for removing covariant step 7 Step 7. Effect Arava (Leflunomide) is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table G.33- Model Fit statistics after removing covariant step 7

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5951.321 SC 5967.947 6717.025 -2 Log L 5938.018 5759.321 420

Table G.34- Testing Null hypothesis after removing covariant step 7 Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 178.6973 93 <.0001 Score 222.8700 93 <.0001 Wald 195.3658 93 <.0001

Table G.35- Residual removing covariant step 7

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 42.0866 60 0.9618

Table G.36- Model Fit statistics for removing covariant step 8 Step 8. Effect Anakinra is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table G.37- Model Fit statistics after removing covariant step 8

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5941.440 SC 5967.947 6635.359 -2 Log L 5938.018 5767.440

Table G.38- Testing Null hypothesis after removing covariant step 8

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 170.5786 84 <.0001

421

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Score 212.2333 84 <.0001 Wald 186.4767 84 <.0001

Table G.39- Residual removing covariant step 8

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 51.4198 69 0.9439

Table G.40- Model Fit statistics for removing covariant step 9

Step 9. Effect Penicillamine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table G.41- Model Fit statistics after removing covariant step 9

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5938.817 SC 5967.947 6560.952 -2 Log L 5938.018 5782.817

Table G.42- Testing Null hypothesis after removing covariant step 9 Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 155.2011 75 <.0001 Score 199.4953 75 <.0001 Wald 176.5609 75 <.0001

422

Table G.43- Residual removing covariant step 9

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 63.9498 78 0.8742

423

Table G.44- Model Fit statistics for removing covariant step 10 Step 10. Effect Abatacept is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table G.45- Model Fit statistics after removing covariant step 10

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5932.444 SC 5967.947 6482.794 -2 Log L 5938.018 5794.444

Table G.46- Testing Null hypothesis after removing covariant step 10

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 143.5746 66 <.0001 Score 181.8225 66 <.0001 Wald 161.2515 66 <.0001

Table G.47- Residual removing covariant step 10

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 74.9017 87 0.8192

Table G.48- Model Fit statistics for removing covariant step 11

Step 11. Effect Folic Acid is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

424

Table G.49- Model Fit statistics after removing covariant step 11

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5929.937 SC 5967.947 6456.359 -2 Log L 5938.018 5797.937

Table G.50- Testing Null hypothesis after removing covariant step 11

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 140.0810 63 <.0001 Score 178.1497 63 <.0001 Wald 157.6714 63 <.0001

Table G.51- Residual removing covariant step 11

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 77.7585 90 0.8178

425

Table G.52- Model Fit statistics for removing covariant step 12

Step 12. Effect Rituximab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table G.53- Model Fit statistics for removing covariant step 12

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5923.849 SC 5967.947 6378.486 -2 Log L 5938.018 5809.849

Table G.54- Testing Null hypothesis after removing covariant step 12

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 128.1694 54 <.0001 Score 163.3382 54 <.0001 Wald 144.8243 54 <.0001

Table G.55- Residual removing covariant step 12

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 93.3813 99 0.6404

Table G.56- Model Fit statistics for removing covariant step 13

426

Step 13. Effect Hydroxychloroquine is removed:

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table G.57- Model Fit statistics after removing covariant step 13

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5917.151 SC 5967.947 6300.003 -2 Log L 5938.018 5821.151

Table G.58- Testing Null hypothesis after removing covariant step 13

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 116.8672 45 <.0001 Score 151.1024 45 <.0001 Wald 132.5918 45 <.0001

Table G.59- Residual removing covariant step 13

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 105.9946 108 0.5366

Table G.60- Model Fit statistics for removing covariant step 14

Step 14. Effect Sulphasalazine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

427

Table G.61- Model Fit statistics after removing covariant step 14

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 5944.018 5913.025 SC 5967.947 6224.093 -2 Log L 5938.018 5835.025

Table G.62- Testing Null hypothesis after removing covariant step 14

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 102.9929 36 <.0001 Score 137.5680 36 <.0001 Wald 118.9900 36 <.0001

Table G.63- Residual removing covariant step 14

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 118.3833 117 0.4468

Table G.64- Summary of backward elimination in GIT

Note: No (additional) effects met the 0.05 significance level for removal from the model.

Summary of Backward Elimination Effect Number Wald Variable Step Removed DF In Chi-Square Pr > ChiSq Label 1 Certolizumab 9 17 1.0448 0.9993 Certolizumab 2 Azathioprine 9 16 3.6049 0.9354 Azathioprine 3 IM Gold injection 9 15 4.1551 0.9009 IM Gold injection 4 Golimumab 6 14 2.5967 0.8575 Golimumab

428

Summary of Backward Elimination Effect Number Wald Variable Step Removed DF In Chi-Square Pr > ChiSq Label 5 Tocilizumab 9 13 4.0299 0.9094 Tocilizumab 6 Etanercept 9 12 4.7392 0.8564 7 Arava (Leflunomide) 9 11 7.0144 0.6356 Arava (Leflunomide) 8 Anakinra 9 10 8.1882 0.5153 9 Penicillamine 9 9 9.0251 0.4350 Penicillamine 10 Abatacept 9 8 9.7122 0.3743 Abatacept 11 Folic Acid 3 7 3.3746 0.3374 Folic Acid 12 Rituximab 9 6 11.2027 0.2621 Rituximab 13 Hydroxychloroquine 9 5 12.0724 0.2093 Hydroxychloroquine 14 Sulphasalazine 9 4 13.3164 0.1488 Sulphasalazine

Table G.65- Type 3 analysis of effects in GIT

Type 3 Analysis of Effects Wald Effect DF Chi-Square Pr > ChiSq Adalimumab 9 17.6510 0.0394 Infliximab 9 20.5203 0.0150 Cyclosporin 9 45.7794 <.0001 Prednisolone 9 21.3027 0.0114

Table G.66- Analysis of maximum likelihood estimates in GIT

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfGit DF Estimate Error Chi-Square Pr > ChiSq Intercept Mild 1 -5.2849 0.2624 405.6943 <.0001 Intercept Mod 1 -5.0911 0.2188 541.1765 <.0001 Intercept Severe 1 -5.3511 0.2559 437.3119 <.0001 Adalimumab 3 Mild 1 -0.2671 0.2430 1.2084 0.2716 Adalimumab 3 Mod 1 0.3088 0.1609 3.6832 0.0550

429

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfGit DF Estimate Error Chi-Square Pr > ChiSq Adalimumab 3 Severe 1 0.4587 0.1904 5.8070 0.0160 Adalimumab 4 Mild 1 -10.7504 379.2 0.0008 0.9774 Adalimumab 4 Mod 1 -0.0400 0.8145 0.0024 0.9609 Adalimumab 4 Severe 1 0.2794 1.1205 0.0622 0.8031 Adalimumab currently taking Mild 1 -0.6383 0.2648 5.8087 0.0159 Adalimumab currently taking Mod 1 0.1312 0.1664 0.6217 0.4304 Adalimumab currently taking Severe 1 -0.1035 0.2231 0.2151 0.6428 Adalimumab never taking Mild 0 0 . . . Adalimumab never taking Mod 0 0 . . . Adalimumab never taking Severe 0 0 . . . Cyclosporin 3 Mild 1 0.2056 0.2607 0.6218 0.4304 Cyclosporin 3 Mod 1 0.1697 0.1743 0.9487 0.3300 Cyclosporin 3 Severe 1 -0.0189 0.2203 0.0073 0.9318 Cyclosporin 4 Mild 1 -0.1041 1.0088 0.0106 0.9178 Cyclosporin 4 Mod 1 -1.6749 1.0458 2.5652 0.1092 Cyclosporin 4 Severe 1 -0.8419 1.0380 0.6579 0.4173 Cyclosporin currently taking Mild 1 1.8937 0.4688 16.3187 <.0001 Cyclosporin currently taking Mod 1 1.8260 0.3563 26.2594 <.0001 Cyclosporin currently taking Severe 1 0.7529 0.7200 1.0933 0.2957 Cyclosporin never taking Mild 0 0 . . . Cyclosporin never taking Mod 0 0 . . . Cyclosporin never taking Severe 0 0 . . . Infliximab 3 Mild 1 0.1003 0.3783 0.0703 0.7909 Infliximab 3 Mod 1 0.2989 0.2397 1.5553 0.2124 Infliximab 3 Severe 1 0.6058 0.2596 5.4445 0.0196 Infliximab 4 Mild 1 -11.1291 340.5 0.0011 0.9739 Infliximab 4 Mod 1 1.3895 0.4631 9.0037 0.0027 Infliximab 4 Severe 1 -0.0667 1.0786 0.0038 0.9507 Infliximab currently taking Mild 1 -0.6060 0.7169 0.7147 0.3979 Infliximab currently taking Mod 1 0.6928 0.3058 5.1333 0.0235 Infliximab currently taking Severe 1 0.2397 0.4640 0.2669 0.6054 Infliximab never taking Mild 0 0 . . .

430

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfGit DF Estimate Error Chi-Square Pr > ChiSq Infliximab never taking Mod 0 0 . . . Infliximab never taking Severe 0 0 . . . Prednisolone 3 Mild 1 0.3716 0.2979 1.5564 0.2122 Prednisolone 3 Mod 1 0.4629 0.2400 3.7208 0.0537 Prednisolone 3 Severe 1 -0.0677 0.2991 0.0513 0.8209 Prednisolone 4 Mild 1 -11.1279 597.6 0.0003 0.9851 Prednisolone 4 Mod 1 1.6744 0.7943 4.4439 0.0350 Prednisolone 4 Severe 1 1.2636 1.0893 1.3457 0.2460 Prednisolone currently taking Mild 1 0.1868 0.3008 0.3856 0.5346 Prednisolone currently taking Mod 1 0.4391 0.2384 3.3921 0.0655 Prednisolone currently taking Severe 1 0.5437 0.2780 3.8254 0.0505 Prednisolone never taking Mild 0 0 . . . Prednisolone never taking Mod 0 0 . . . Prednisolone never taking Severe 0 0 . . .

Table G.67- Odds ratio estimates in GIT

Odds Ratio Estimates 95% Wald Effect InfGit Point Estimate Confidence Limits Adalimumab 3 vs never taking Mild 0.766 0.476 1.233 Adalimumab 3 vs never taking Mod 1.362 0.993 1.867 Adalimumab 3 vs never taking Severe 1.582 1.089 2.297 Adalimumab 4 vs never taking Mild <0.001 <0.001 >999.999 Adalimumab 4 vs never taking Mod 0.961 0.195 4.742 Adalimumab 4 vs never taking Severe 1.322 0.147 11.889 Adalimumab currently taking vs never taking Mild 0.528 0.314 0.888 Adalimumab currently taking vs never taking Mod 1.140 0.823 1.580 Adalimumab currently taking vs never taking Severe 0.902 0.582 1.396 Infliximab 3 vs never taking Mild 1.105 0.527 2.320 Infliximab 3 vs never taking Mod 1.348 0.843 2.157 Infliximab 3 vs never taking Severe 1.833 1.102 3.048 431

Odds Ratio Estimates 95% Wald Effect InfGit Point Estimate Confidence Limits Infliximab 4 vs never taking Mild <0.001 <0.001 >999.999 Infliximab 4 vs never taking Mod 4.013 1.619 9.946 Infliximab 4 vs never taking Severe 0.935 0.113 7.748 Infliximab currently taking vs never taking Mild 0.546 0.134 2.223 Infliximab currently taking vs never taking Mod 1.999 1.098 3.640 Infliximab currently taking vs never taking Severe 1.271 0.512 3.155 Cyclosporin 3 vs never taking Mild 1.228 0.737 2.047 Cyclosporin 3 vs never taking Mod 1.185 0.842 1.667 Cyclosporin 3 vs never taking Severe 0.981 0.637 1.511 Cyclosporin 4 vs never taking Mild 0.901 0.125 6.508 Cyclosporin 4 vs never taking Mod 0.187 0.024 1.455 Cyclosporin 4 vs never taking Severe 0.431 0.056 3.295 Cyclosporin currently taking vs never taking Mild 6.644 2.651 16.651 Cyclosporin currently taking vs never taking Mod 6.209 3.088 12.484 Cyclosporin currently taking vs never taking Severe 2.123 0.518 8.707 Prednisolone 3 vs never taking Mild 1.450 0.809 2.600 Prednisolone 3 vs never taking Mod 1.589 0.993 2.543 Prednisolone 3 vs never taking Severe 0.935 0.520 1.679 Prednisolone 4 vs never taking Mild <0.001 <0.001 >999.999 Prednisolone 4 vs never taking Mod 5.335 1.125 25.308 Prednisolone 4 vs never taking Severe 3.538 0.418 29.923 Prednisolone currently taking vs never taking Mild 1.205 0.668 2.174 Prednisolone currently taking vs never taking Mod 1.551 0.972 2.475 Prednisolone currently taking vs never taking Severe 1.722 0.999 2.970

432

APPENDIX H: OUTPUT OF SAS FOR NERVOUS SYSTEM INFECTION

Table H.1- Complete statistics for Nervous system infection

Model Information Data Set WORK.IMPORT2 Response Variable InfNeuro InfNeuro Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table H.2- Observation status for Nervous system infection

Number of Observations Read 27711 Number of Observations Used 21506

433

Table H.3- response value for Nervous system infection

Response Profile Ordered Total Value InfNeuro Frequency 1 1 9 2 2 9 3 3 12 4 4 21476

Logits modelled use InfNeuro='4' as the reference category.

Note: 6205 observations were deleted due to missing values for the response or explanatory variables.

Table H.4- Backward Elimination Procedure for Nervous system infection

Backward Elimination Procedure

Class Level Information

Class Value Design Variables Etanercept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Adalimumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Anakinra 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0

434

never taking 0 0 0 1

Infliximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Rituximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Abatacept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Tocilizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Golimumab 3 1 0 0 currently taking 0 1 0 never taking 0 0 1

Certolizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Folic Acid currently taking 1 0 never taking 0 1

Hydroxychloroquine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Sulphasalazine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0

435

never taking 0 0 0 1

Arava (Leflunomide) 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Azathioprine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Cyclosporin 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Prednisolone 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

IM Gold injection 3 1 0 0 0

4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Penicillamine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

436

Table H.6- Model Fit statistics for Nervous system infection

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 719.397 SC 549.714 1963.667 -2 Log L 519.786 407.397

Table H.7- Testing null hypothesis for Nervous system infection

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 112.3885 153 0.9943 Score 170.5176 153 0.1578 Wald 83.3711 153 1.0000

Table H.8- Model Fit statistics for removing covariant step 1

Step 1. Effect Certolizumab is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table H.11- Residual removing covariant step 1

437

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 0.9204 9 0.9996

Table H.12- Model Fit statistics for removing covariant step 2

Step 2. Effect Anakinra is removed: Model Convergence Status Quasi-complete separation of data points detected.

438

Table H.13- Model Fit statistics after removing covariant step 2

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 686.901 SC 549.714 1787.601 -2 Log L 519.786 410.901

Table H.14- Testing Null hypothesis after removing covariant step 2

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 108.8846 135 0.9519 Score 158.9265 135 0.0781 Wald 81.8673 135 0.9999

Table H.15- Residual removing covariant step 2

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 1.9099 18 1.0000

439

Table H.16- Model Fit statistics for removing covariant step 3

Model Convergence Status Quasi-complete separation of data points detected.

Table H.17- Model Fit statistics after removing covariant step 3 Step 3. Effect Golimumab is removed:

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 677.704 SC 549.714 1730.547 -2 Log L 519.786 413.704

Table H.18- Testing Null hypothesis after removing covariant step 3

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 106.0818 129 0.9305 Score 156.3178 129 0.0511 Wald 80.1217 129 0.9998

440

Table H.19- Residual removing covariant step 3

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 3.7095 24 1.0000

Table H.20- Model Fit statistics for removing covariant step 4

Step 4. Effect Penicillamine is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table H.21- Model Fit statistics after removing covariant step 4

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 663.254 SC 549.714 1644.313 -2 Log L 519.786 417.254

441

Table H.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 102.5317 120 0.8737 Score 149.3559 120 0.0358 Wald 78.8583 120 0.9986

Table H.23- Residual removing covariant step 4

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 5.9028 33 1.0000

Table H.24- Model Fit statistics for removing covariant step 5 Step 5. Effect Tocilizumab is removed: Model Convergence Status Quasi-complete separation of data points detected.

442

Table H.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 649.106 SC 549.714 1558.380 -2 Log L 519.786 421.106

Table H.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 98.6796 111 0.7923 Score 143.4857 111 0.0206 Wald 77.1681 111 0.9939

Table H.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 9.2419 42 1.0000

443

Table H.28- Model Fit statistics for removing covariant step 6

Step 6. Effect Rituximab is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table H.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 634.968 SC 549.714 1472.457 -2 Log L 519.786 424.968

Table H.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 94.8175 102 0.6802 Score 132.3092 102 0.0234 Wald 74.4728 102 0.9815

Table H.31- Residual removing covariant step 6

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 12.5221 51 1.0000

444

Table H.32- Model Fit statistics for removing covariant step 7

Step 7. Effect Arava (Leflunomide) is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table H.33- Model Fit statistics after removing covariant step 7

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 622.390 SC 549.714 1388.095 -2 Log L 519.786 430.390

445

Table H.34- Testing Null hypothesis after removing covariant step 7

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 89.3954 93 0.5866 Score 127.7230 93 0.0099 Wald 75.4615 93 0.9078

Table H.35- Residual removing covariant step 7

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 15.8844 60 1.0000

Table H.36- Model Fit statistics for removing covariant step 8 Step 8. Effect Abatacept is removed: Model Convergence Status Quasi-complete separation of data points detected.

446

Table H.37- Model Fit statistics after removing covariant step 8

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 606.681 SC 549.714 1300.601 -2 Log L 519.786 432.681

Table H.38- Testing Null hypothesis after removing covariant step 8

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 87.1046 84 0.3867 Score 124.9842 84 0.0025 Wald 75.0946 84 0.7457

Table H.39- Residual removing covariant step 8

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 18.2897 69 1.0000

Table H.40- Model Fit statistics for removing covariant step 9

Step 9. Effect Hydroxychloroquine is removed: Model Convergence Status Quasi-complete separation of data points detected.

447

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 592.141 SC 549.714 1214.276 -2 Log L 519.786 436.141

Table H.42- Testing Null hypothesis after removing covariant step 9

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 83.6442 75 0.2314 Score 120.9163 75 0.0006 Wald 72.3119 75 0.5665

Table H.43- Residual removing covariant step 9

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 21.4107 78 1.0000

448

Table H.44- Model Fit statistics for removing covariant step 10 Step 10. Effect Adalimumab is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table H.45- Model Fit statistics after removing covariant step 10

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 585.500 SC 549.714 1135.851 -2 Log L 519.786 447.500

Table H.46- Testing Null hypothesis after removing covariant step 10

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 72.2850 66 0.2782 Score 103.0295 66 0.0024 Wald 68.9775 66 0.3771

449

Table H.47- Residual removing covariant step 10

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 31.5451 87 1.0000

Table H.48- Model Fit statistics for removing covariant step 11

Step 11. Effect Infliximab is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table H.49- Model Fit statistics after removing covariant step 11

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 576.716 SC 549.714 1055.281 -2 Log L 519.786 456.716

450

Table H.50- Testing Null hypothesis after removing covariant step 11

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 63.0694 57 0.2705 Score 84.0113 57 0.0115 Wald 60.0232 57 0.3667

Table H.51- Residual removing covariant step 11

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 44.0268 96 1.0000

Table H.52- Model Fit statistics for removing covariant step 12 Step 12. Effect Prednisolone is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table H.53- Model Fit statistics for removing covariant step 12

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 566.295 SC 549.714 973.076 -2 Log L 519.786 464.295

451

Table H.54- Testing Null hypothesis after removing covariant step 12

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 55.4901 48 0.2132 Score 78.2457 48 0.0038 Wald 58.2370 48 0.1479

Table H.55- Residual removing covariant step 12

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 50.5258 105 1.0000

Table H.56- Model Fit statistics for removing covariant step 13 Step 13. Effect Etanercept is removed: Model Convergence Status Quasi-complete separation of data points detected.

452

Table H.57- Model Fit statistics after removing covariant step 13

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 555.825 SC 549.714 890.820 -2 Log L 519.786 471.825

Table H.58- Testing Null hypothesis after removing covariant step 13

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 47.9610 39 0.1538 Score 69.5512 39 0.0019 Wald 51.9824 39 0.0798

Table H.59- Residual removing covariant step 13

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 60.1603 114 1.0000

Table H.60- Model Fit statistics for removing covariant step 14

Step 14. Effect Folic Acid is removed: Model Convergence Status Quasi-complete separation of data points detected. Warning: The maximum likelihood estimate may not exist. Warning: The LOGISTIC procedure continues in spite of the above warning. Results shown are based on the last maximum likelihood iteration. Validity of the model fit is questionable Table H.61- Model Fit statistics after removing covariant step 14

453

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 552.316 SC 549.714 863.383 -2 Log L 519.786 474.316

Table H.62- Testing Null hypothesis after removing covariant step 14

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 45.4695 36 0.1339 Score 67.3854 36 0.0012 Wald 49.9133 36 0.0615

Table H.63- Residual removing covariant step 14

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 62.5527 117 1.0000

454

Table H.64- Model Fit statistics for removing covariant step 15

Step 15. Effect IM Gold injection is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table H.65- Model Fit statistics after removing covariant step 15

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 551.109 SC 549.714 790.391 -2 Log L 519.786 491.109

Table H.66- Testing Null hypothesis after removing covariant step 15

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 28.6770 27 0.3767 Score 52.8187 27 0.0021 Wald 37.7840 27 0.0813

455

Table H.67- Residual removing covariant step 15

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 79.9812 126 0.9995

Table H.68(1)- Model Fit statistics for removing covariant step 16

Step 16. Effect Sulphasalazine is removed:

Model Convergence Status Quasi-complete separation of data points detected.

Table H.68 (2)- Model Fit statistics after removing covariant step 16

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 541.646 SC 549.714 709.144 -2 Log L 519.786 499.646

456

Table H.69- Testing Null hypothesis after removing covariant step 16

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 20.1392 18 0.3250 Score 41.3631 18 0.0014 Wald 28.2599 18 0.0582

Table H.70- Residual removing covariant step 16

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 99.5067 135 0.9904

Table H.71- Model Fit statistics for removing covariant step 17 Step 17. Effect Cyclosporin is removed: Model Convergence Status Quasi-complete separation of data points detected.

Table H.72- Model Fit statistics after removing covariant step 17

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 525.786 534.280 SC 549.714 629.993 -2 Log L 519.786 510.280

457

Table H.73- Testing Null hypothesis after removing covariant step 17

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 9.5056 9 0.3920 Score 16.0170 9 0.0665 Wald 10.8344 9 0.2872

Table H.74- Residual removing covariant step 17

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 123.0393 144 0.8963

Table H.75- Model Fit statistics for removing covariant step 18

Step 18. Effect Azathioprine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

458

Table H.76- Model Fit statistics after removing covariant step 18

-2 Log L = 519.786

Table H.77- Testing Null hypothesis after removing covariant step 18

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 170.5176 153 0.1578

Table H.78- Summary of backward elimination in Nervous system infection Note: All effects have been removed from the model.

Summary of Backward Elimination Wald Effect Number Chi- Variable Step Removed DF In Square Pr > ChiSq Label 1 Certolizumab 9 17 0.0029 1.0000 Certolizumab 2 Anakinra 9 16 0.0051 1.0000 3 Golimumab 6 15 0.0054 1.0000 Golimumab 4 Penicillamine 9 14 0.3675 1.0000 Penicillamine 5 Tocilizumab 9 13 0.5269 1.0000 Tocilizumab 6 Rituximab 9 12 0.5838 0.9999 Rituximab 7 Arava (Leflunomide) 9 11 1.0212 0.9994 Arava (Leflunomide) 8 Abatacept 9 10 1.5025 0.9971 Abatacept 9 Hydroxychloroquine 9 9 2.5989 0.9781 Hydroxychloroquine 10 Adalimumab 9 8 4.0119 0.9106 11 Infliximab 9 7 4.8230 0.8495 12 Prednisolone 9 6 4.6980 0.8598 Prednisolone 13 Etanercept 9 5 6.7886 0.6591 14 Folic Acid 3 4 2.0592 0.5602 Folic Acid 15 IM Gold injection 9 3 8.1059 0.5235 IM Gold injection

16 Sulphasalazine 9 2 8.6399 0.4712 Sulphasalazine

459

17 Cyclosporin 9 1 11.0409 0.2729 Cyclosporin 18 Azathioprine 9 0 10.8344 0.2872 Azathioprine

Table H.79- Analysis of maximum likelihood estimates in Nervous system infection

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfNeuro DF Estimate Error Chi-Square Pr > ChiSq Intercept 1 1 -7.7775 0.3334 544.1729 <.0001 Intercept 2 1 -7.7775 0.3334 544.1729 <.0001 Intercept 3 1 -7.4898 0.2888 672.7866 <.0001

460

APPENDIX I: OUTPUT OF SAS FOR TB INFECTION

Table I.1- Complete statistics for TB infection

Model Information Data Set WORK.IMPORT2 Response Variable TB Infection TB Infection Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table I.2- Observation status for TB infection

Number of Observations Read 27711 Number of Observations Used 21506

Table I.3- response value for TB infection

Response Profile Ordered Total Value TB Infection Frequency 1 1 1050 2 2 1829 3 3 406 4 4 18221

Logits modelled use TB Infection='4' as the reference category.

461

Note: 6205 observations were deleted due to missing values for the response or explanatory variables.

Table I.4- Backward Elimination Procedure for TB infection

Backward Elimination Procedure Class Level Information Class Value Design Variables Etanercept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Adalimumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Anakinra 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Infliximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Rituximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Abatacept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Tocilizumab 3 1 0 0 0

462

4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Golimumab 3 1 0 0 currently taking 0 1 0 never taking 0 0 1

Certolizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Folic Acid currently taking 1 0 never taking 0 1

Hydroxychloroquine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Sulphasalazine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Arava (Leflunomide) 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Azathioprine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Cyclosporin 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Prednisolone 3 1 0 0 0

463

4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

IM Gold injection 3 1 0 0 0

4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Penicillamine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Table I.5- Model Convergence status for TB infection

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table I.6- Model Fit statistics for TB infection

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24501.128 SC 24650.284 25745.398 -2 Log L 24620.355 24189.128

464

Table I.7- Testing null hypothesis for TB infection

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 431.2272 153 <.0001 Score 463.0664 153 <.0001 Wald 419.5882 153 <.0001

Table I.8- Model Fit statistics for removing covariant step 1

Step 1. Effect Azathioprine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table I.9- Model Fit statistics for removing covariant step 1

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24488.566 SC 24650.284 25661.051 -2 Log L 24620.355 24194.566

Table I.10- Testing Null hypothesis after removing covariant step 1

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 425.7897 144 <.0001 Score 457.8861 144 <.0001 Wald 415.1007 144 <.0001

465

Table I.11- Residual removing covariant step 1

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 5.0524 9 0.8297

466

Table I.12- Model Fit statistics for removing covariant step 2 Step 2. Effect Certolizumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table I.13- Model Fit statistics after removing covariant step 2

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24476.658 SC 24650.284 25577.358 -2 Log L 24620.355 24200.658

Table I.14- Testing Null hypothesis after removing covariant step 2

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 419.6974 135 <.0001 Score 450.9468 135 <.0001 Wald 408.7712 135 <.0001

467

Table I.15- Residual removing covariant step 2

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 10.9161 18 0.8979

Table I.16- Model Fit statistics for removing covariant step 3

Step 3. Effect Penicillamine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table I.17- Model Fit statistics after removing covariant step 3

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24473.673 SC 24650.284 25502.589 -2 Log L 24620.355 24215.673

468

Table I.18- Testing Null hypothesis after removing covariant step 3

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 404.6821 126 <.0001 Score 440.0301 126 <.0001 Wald 401.9656 126 <.0001

Table I.19- Residual removing covariant step 3

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 22.0077 27 0.7370

Table I.20- Model Fit statistics for removing covariant step 4

Step 4. Effect IM Gold injection is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

469

Table I.21- Model Fit statistics after removing covariant step 4

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24465.880 SC 24650.284 25423.011 -2 Log L 24620.355 24225.880

Table I.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 394.4751 117 <.0001 Score 430.4313 117 <.0001 Wald 392.2553 117 <.0001

Table I.23- Residual removing covariant step 4

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 31.2787 36 0.6926

470

Table I.24- Model Fit statistics for removing covariant step 5 Step 5. Effect Rituximab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table I.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24461.305 SC 24650.284 25346.650 -2 Log L 24620.355 24239.305

Table I.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 381.0509 108 <.0001 Score 415.9895 108 <.0001 Wald 378.4582 108 <.0001

471

Table I.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 44.9975 45 0.4721

Table I.28- Model Fit statistics for removing covariant step 6

Step 6. Effect Golimumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table I.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 24626.355 24462.511 SC 24650.284 25300.000 -2 Log L 24620.355 24252.511

Table I.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 367.8443 102 <.0001 Score 403.4935 102 <.0001 Wald 366.8141 102 <.0001

Table I.31- Residual removing covariant step 6

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 57.2672 51 0.2539

472

Note: No (additional) effects met the 0.05 significance level for removal from the model.

Table I.32- Summary of backward elimination in TB infection

473

Table I.33- Type 3 analysis of effects in TB infection

Arava (Leflunomide) 9 17.5339 0.0410 Cyclosporin 9 47.3358 <.0001 Prednisolone 9 29.4764 0.0005

Table I.34- Analysis of maximum likelihood estimates in TB infection

474

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Etanercept currently 1 1 0.1730 0.0941 3.3831 0.0659 taking Etanercept currently 2 1 0.0891 0.0722 1.5232 0.2171 taking Etanercept currently 3 1 -0.3383 0.1446 5.4736 0.0193 taking Etanercept never taking 1 0 0 . . . Etanercept never taking 2 0 0 . . . Etanercept never taking 3 0 0 . . . Adalimumab 3 1 1 0.0104 0.0914 0.0129 0.9094 Adalimumab 3 2 1 0.1823 0.0686 7.0504 0.0079 Adalimumab 3 3 1 0.1418 0.1403 1.0222 0.3120 Adalimumab 4 1 1 -0.5402 0.6756 0.6394 0.4239 Adalimumab 4 2 1 - 0.4440 0.0000 0.9984 0.00090 Adalimumab 4 3 1 - 147.6 0.0048 0.9447 10.2462 Adalimumab currently 1 1 0.2887 0.0941 9.4206 0.0021 taking Adalimumab currently 2 1 0.1847 0.0737 6.2813 0.0122 taking Adalimumab currently 3 1 -0.0798 0.1470 0.2946 0.5873 taking Adalimumab never taking 1 0 0 . . . Adalimumab never taking 2 0 0 . . . Adalimumab never taking 3 0 0 . . . Anakinra 3 1 1 0.1448 0.2523 0.3295 0.5659 Anakinra 3 2 1 -0.0761 0.2191 0.1205 0.7285 Anakinra 3 3 1 0.4597 0.3413 1.8146 0.1780 Anakinra 4 1 1 -0.7187 0.6297 1.3026 0.2537 Anakinra 4 2 1 0.0275 0.4047 0.0046 0.9459

475

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Anakinra 4 3 1 -0.4484 1.0513 0.1819 0.6697 Anakinra currently 1 1 - 512.1 0.0005 0.9814 taking 11.9697 Anakinra currently 2 1 1.7999 0.4745 14.3901 0.0001 taking Anakinra currently 3 1 - 809.5 0.0002 0.9879 taking 12.2422 Anakinra never taking 1 0 0 . . . Anakinra never taking 2 0 0 . . . Anakinra never taking 3 0 0 . . . Infliximab 3 1 1 0.0552 0.1337 0.1707 0.6795 Infliximab 3 2 1 -0.2055 0.1098 3.5047 0.0612 Infliximab 3 3 1 0.0478 0.1974 0.0585 0.8088 Infliximab 4 1 1 0.4422 0.4291 1.0621 0.3027 Infliximab 4 2 1 -0.1974 0.3745 0.2779 0.5981 Infliximab 4 3 1 -0.8062 0.8952 0.8110 0.3678 Infliximab currently 1 1 0.6440 0.1747 13.5909 0.0002 taking Infliximab currently 2 1 0.4727 0.1396 11.4614 0.0007 taking Infliximab currently 3 1 -0.3706 0.3711 0.9971 0.3180 taking Infliximab never taking 1 0 0 . . . Infliximab never taking 2 0 0 . . . Infliximab never taking 3 0 0 . . . Abatacept 3 1 1 0.5166 0.1769 8.5260 0.0035 Abatacept 3 2 1 0.1147 0.1534 0.5587 0.4548 Abatacept 3 3 1 -0.4339 0.3573 1.4747 0.2246 Abatacept 4 1 1 0.2751 0.8455 0.1059 0.7449 Abatacept 4 2 1 -0.6022 0.6388 0.8887 0.3458 Abatacept 4 3 1 1.3251 1.0615 1.5582 0.2119

476

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Abatacept currently 1 1 0.3362 0.1582 4.5176 0.0335 taking Abatacept currently 2 1 0.1491 0.1240 1.4460 0.2292 taking Abatacept currently 3 1 -0.2016 0.2673 0.5686 0.4508 taking Abatacept never taking 1 0 0 . . . Abatacept never taking 2 0 0 . . . Abatacept never taking 3 0 0 . . . Tocilizumab 3 1 1 0.1595 0.2454 0.4224 0.5157 Tocilizumab 3 2 1 0.1835 0.1951 0.8847 0.3469 Tocilizumab 3 3 1 0.7127 0.3269 4.7534 0.0292 Tocilizumab 4 1 1 - 529.0 0.0005 0.9827 11.4739 Tocilizumab 4 2 1 - 222.6 0.0027 0.9587 11.5154 Tocilizumab 4 3 1 - 820.5 0.0002 0.9894 10.9097 Tocilizumab currently 1 1 0.4933 0.1695 8.4682 0.0036 taking Tocilizumab currently 2 1 0.3301 0.1348 5.9962 0.0143 taking Tocilizumab currently 3 1 0.1795 0.2814 0.4069 0.5236 taking Tocilizumab never taking 1 0 0 . . . Tocilizumab never taking 2 0 0 . . . Tocilizumab never taking 3 0 0 . . . Folic Acid currently 1 1 -0.1059 0.0761 1.9365 0.1641 taking Folic Acid currently 2 1 -0.1683 0.0598 7.9220 0.0049 taking

477

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Folic Acid currently 3 1 -0.0493 0.1190 0.1713 0.6789 taking Folic Acid never taking 1 0 0 . . . Folic Acid never taking 2 0 0 . . . Folic Acid never taking 3 0 0 . . . Hydroxychloroquine 3 1 1 0.1431 0.0736 3.7794 0.0519 Hydroxychloroquine 3 2 1 0.2299 0.0575 15.9873 <.0001 Hydroxychloroquine 3 3 1 0.1860 0.1175 2.5057 0.1134 Hydroxychloroquine 4 1 1 -0.0695 0.4165 0.0278 0.8676 Hydroxychloroquine 4 2 1 -0.0273 0.3338 0.0067 0.9348 Hydroxychloroquine 4 3 1 0.6305 0.5074 1.5444 0.2140 Hydroxychloroquine currently 1 1 0.0100 0.0960 0.0109 0.9168 taking Hydroxychloroquine currently 2 1 0.0789 0.0753 1.0991 0.2945 taking Hydroxychloroquine currently 3 1 0.0332 0.1546 0.0463 0.8297 taking Hydroxychloroquine never taking 1 0 0 . . . Hydroxychloroquine never taking 2 0 0 . . . Hydroxychloroquine never taking 3 0 0 . . . Sulphasalazine 3 1 1 0.0933 0.0714 1.7093 0.1911 Sulphasalazine 3 2 1 0.2229 0.0554 16.1883 <.0001 Sulphasalazine 3 3 1 0.1577 0.1136 1.9284 0.1649 Sulphasalazine 4 1 1 0.1273 0.3002 0.1799 0.6714 Sulphasalazine 4 2 1 -0.2181 0.2582 0.7132 0.3984 Sulphasalazine 4 3 1 0.6855 0.3912 3.0697 0.0798 Sulphasalazine currently 1 1 0.1470 0.1112 1.7468 0.1863 taking Sulphasalazine currently 2 1 0.00403 0.0926 0.0019 0.9653 taking

478

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Sulphasalazine currently 3 1 -0.0445 0.1896 0.0551 0.8144 taking Sulphasalazine never taking 1 0 0 . . . Sulphasalazine never taking 2 0 0 . . . Sulphasalazine never taking 3 0 0 . . . Arava (Leflunomide) 3 1 1 0.1098 0.0935 1.3804 0.2400 Arava (Leflunomide) 3 2 1 0.1933 0.0729 7.0343 0.0080 Arava (Leflunomide) 3 3 1 0.1484 0.1434 1.0712 0.3007 Arava (Leflunomide) 4 1 1 -0.2856 0.5428 0.2768 0.5988 Arava (Leflunomide) 4 2 1 0.4582 0.3091 2.1967 0.1383 Arava (Leflunomide) 4 3 1 -0.7134 1.0581 0.4546 0.5002 Arava (Leflunomide) currently 1 1 0.2705 0.1060 6.5075 0.0107 taking Arava (Leflunomide) currently 2 1 0.1492 0.0858 3.0250 0.0820 taking Arava (Leflunomide) currently 3 1 0.00639 0.1726 0.0014 0.9705 taking Arava (Leflunomide) never taking 1 0 0 . . . Arava (Leflunomide) never taking 2 0 0 . . . Arava (Leflunomide) never taking 3 0 0 . . . Cyclosporin 3 1 1 0.0263 0.0937 0.0789 0.7788 Cyclosporin 3 2 1 0.2042 0.0692 8.7084 0.0032 Cyclosporin 3 3 1 0.4662 0.1325 12.3789 0.0004 Cyclosporin 4 1 1 -0.2418 0.3214 0.5662 0.4518 Cyclosporin 4 2 1 0.0673 0.2205 0.0931 0.7603 Cyclosporin 4 3 1 -1.0526 0.7303 2.0770 0.1495 Cyclosporin currently 1 1 0.5290 0.3373 2.4603 0.1168 taking Cyclosporin currently 2 1 1.0439 0.2236 21.7983 <.0001 taking

479

Analysis of Maximum Likelihood Estimates Wald TB Standard Chi- Parameter Infection DF Estimate Error Square Pr > ChiSq Cyclosporin currently 3 1 1.0216 0.4398 5.3965 0.0202 taking Cyclosporin never taking 1 0 0 . . . Cyclosporin never taking 2 0 0 . . . Cyclosporin never taking 3 0 0 . . . Prednisolone 3 1 1 0.3310 0.1083 9.3359 0.0022 Prednisolone 3 2 1 0.2552 0.0838 9.2693 0.0023 Prednisolone 3 3 1 0.4980 0.1834 7.3738 0.0066 Prednisolone 4 1 1 0.7838 0.5610 1.9520 0.1624 Prednisolone 4 2 1 0.5466 0.4327 1.5961 0.2065 Prednisolone 4 3 1 0.7162 1.0389 0.4753 0.4906 Prednisolone currently 1 1 0.1671 0.1087 2.3642 0.1241 taking Prednisolone currently 2 1 0.1308 0.0838 2.4345 0.1187 taking Prednisolone currently 3 1 0.3911 0.1833 4.5509 0.0329 taking Prednisolone never taking 1 0 0 . . . Prednisolone never taking 2 0 0 . . . Prednisolone never taking 3 0 0 . . .

Table I.35- Odds ratio estimates in TB infection

Odds Ratio Estimates TB Point 95% Wald Effect Infection Estimate Confidence Limits Etanercept 4 vs never taking 3 3.834 0.706 20.819 Etanercept currently taking vs never taking 1 1.189 0.989 1.430 Etanercept currently taking vs never taking 2 1.093 0.949 1.259 Etanercept currently taking vs never taking 3 0.713 0.537 0.947 Adalimumab 3 vs never taking 1 1.010 0.845 1.209 Adalimumab 3 vs never taking 2 1.200 1.049 1.373 Adalimumab 3 vs never taking 3 1.152 0.875 1.517 Adalimumab 4 vs never taking 1 0.583 0.155 2.190 Adalimumab 4 vs never taking 2 0.999 0.419 2.385 Adalimumab 4 vs never taking 3 <0.001 <0.001 >999.999 Adalimumab currently taking vs never taking 1 1.335 1.110 1.605 Adalimumab currently taking vs never taking 2 1.203 1.041 1.390 Adalimumab currently taking vs never taking 3 0.923 0.692 1.232 Anakinra 3 vs never taking 1 1.156 0.705 1.895 Anakinra 3 vs never taking 2 0.927 0.603 1.424 Anakinra 3 vs never taking 3 1.584 0.811 3.091 Anakinra 4 vs never taking 1 0.487 0.142 1.675 Anakinra 4 vs never taking 2 1.028 0.465 2.272 Anakinra 4 vs never taking 3 0.639 0.081 5.013 Anakinra currently taking vs never taking 1 <0.001 <0.001 >999.999 Anakinra currently taking vs never taking 2 6.049 2.387 15.330 Anakinra currently taking vs never taking 3 <0.001 <0.001 >999.999 Infliximab 3 vs never taking 1 1.057 0.813 1.373 Infliximab 3 vs never taking 2 0.814 0.657 1.010 Infliximab 3 vs never taking 3 1.049 0.712 1.544 Infliximab 4 vs never taking 1 1.556 0.671 3.608 Infliximab 4 vs never taking 2 0.821 0.394 1.710 Infliximab 4 vs never taking 3 0.447 0.077 2.582 Infliximab currently taking vs never taking 1 1.904 1.352 2.682 Infliximab currently taking vs never taking 2 1.604 1.220 2.109 Infliximab currently taking vs never taking 3 0.690 0.334 1.429 Abatacept 3 vs never taking 1 1.676 1.185 2.371

481

Odds Ratio Estimates TB Point 95% Wald Effect Infection Estimate Confidence Limits Abatacept 3 vs never taking 2 1.122 0.830 1.515 Abatacept 3 vs never taking 3 0.648 0.322 1.305 Abatacept 4 vs never taking 1 1.317 0.251 6.905 Abatacept 4 vs never taking 2 0.548 0.157 1.915 Abatacept 4 vs never taking 3 3.763 0.470 30.134 Abatacept currently taking vs never taking 1 1.400 1.027 1.908 Abatacept currently taking vs never taking 2 1.161 0.910 1.480 Abatacept currently taking vs never taking 3 0.817 0.484 1.380 Tocilizumab 3 vs never taking 1 1.173 0.725 1.897 Tocilizumab 3 vs never taking 2 1.201 0.820 1.761 Tocilizumab 3 vs never taking 3 2.039 1.075 3.870 Tocilizumab 4 vs never taking 1 <0.001 <0.001 >999.999 Tocilizumab 4 vs never taking 2 <0.001 <0.001 >999.999 Tocilizumab 4 vs never taking 3 <0.001 <0.001 >999.999 Tocilizumab currently taking vs never taking 1 1.638 1.175 2.283 Tocilizumab currently taking vs never taking 2 1.391 1.068 1.812 Tocilizumab currently taking vs never taking 3 1.197 0.689 2.077 Folic Acid currently taking vs never taking 1 0.899 0.775 1.044 Folic Acid currently taking vs never taking 2 0.845 0.752 0.950 Folic Acid currently taking vs never taking 3 0.952 0.754 1.202 Hydroxychloroquine 3 vs never taking 1 1.154 0.999 1.333 Hydroxychloroquine 3 vs never taking 2 1.259 1.124 1.409 Hydroxychloroquine 3 vs never taking 3 1.204 0.957 1.516 Hydroxychloroquine 4 vs never taking 1 0.933 0.412 2.110 Hydroxychloroquine 4 vs never taking 2 0.973 0.506 1.872 Hydroxychloroquine 4 vs never taking 3 1.879 0.695 5.078 Hydroxychloroquine currently taking vs never 1 1.010 0.837 1.219 taking Hydroxychloroquine currently taking vs never 2 1.082 0.934 1.254 taking Hydroxychloroquine currently taking vs never 3 1.034 0.764 1.400 taking

482

Odds Ratio Estimates TB Point 95% Wald Effect Infection Estimate Confidence Limits Sulphasalazine 3 vs never taking 1 1.098 0.955 1.263 Sulphasalazine 3 vs never taking 2 1.250 1.121 1.393 Sulphasalazine 3 vs never taking 3 1.171 0.937 1.463 Sulphasalazine 4 vs never taking 1 1.136 0.631 2.046 Sulphasalazine 4 vs never taking 2 0.804 0.485 1.334 Sulphasalazine 4 vs never taking 3 1.985 0.922 4.273 Sulphasalazine currently taking vs never taking 1 1.158 0.931 1.440 Sulphasalazine currently taking vs never taking 2 1.004 0.837 1.204 Sulphasalazine currently taking vs never taking 3 0.956 0.660 1.387 Arava (Leflunomide) 3 vs never taking 1 1.116 0.929 1.341 Arava (Leflunomide) 3 vs never taking 2 1.213 1.052 1.399 Arava (Leflunomide) 3 vs never taking 3 1.160 0.876 1.536 Arava (Leflunomide) 4 vs never taking 1 0.752 0.259 2.178 Arava (Leflunomide) 4 vs never taking 2 1.581 0.863 2.898 Arava (Leflunomide) 4 vs never taking 3 0.490 0.062 3.898 Arava (Leflunomide) currently taking vs never 1 1.311 1.065 1.613 taking Arava (Leflunomide) currently taking vs never 2 1.161 0.981 1.374 taking Arava (Leflunomide) currently taking vs never 3 1.006 0.717 1.412 taking Cyclosporin 3 vs never taking 1 1.027 0.854 1.234 Cyclosporin 3 vs never taking 2 1.227 1.071 1.405 Cyclosporin 3 vs never taking 3 1.594 1.229 2.066 Cyclosporin 4 vs never taking 1 0.785 0.418 1.474 Cyclosporin 4 vs never taking 2 1.070 0.694 1.648 Cyclosporin 4 vs never taking 3 0.349 0.083 1.461 Cyclosporin currently taking vs never taking 1 1.697 0.876 3.287 Cyclosporin currently taking vs never taking 2 2.840 1.833 4.403 Cyclosporin currently taking vs never taking 3 2.778 1.173 6.577 Prednisolone 3 vs never taking 1 1.392 1.126 1.722 Prednisolone 3 vs never taking 2 1.291 1.095 1.521

483

Odds Ratio Estimates TB Point 95% Wald Effect Infection Estimate Confidence Limits Prednisolone 3 vs never taking 3 1.645 1.149 2.357 Prednisolone 4 vs never taking 1 2.190 0.729 6.576 Prednisolone 4 vs never taking 2 1.727 0.740 4.034 Prednisolone 4 vs never taking 3 2.047 0.267 15.680 Prednisolone currently taking vs never taking 1 1.182 0.955 1.462 Prednisolone currently taking vs never taking 2 1.140 0.967 1.343 Prednisolone currently taking vs never taking 3 1.479 1.032 2.118

484

APPENDIX J: OUTPUT OF SAS FOR URINARY TRACT INFECTION

Table J.1- Complete statistics for UTI

Model Information Data Set WORK.IMPORT2 Response Variable InfKidUri InfKidUri Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table J.2- Observation status for UTI

Number of Observations Read 27711 Number of Observations Used 21506

Table J.3- response value for UTI

Response Profile Ordered Total Value InfKidUri Frequency 1 1 290 2 2 833 3 3 256 4 4 20127

Logits modelled use InfKidUri='4' as the reference category.

485

Note: 6205 observations were deleted due to missing values for the response or explanatory variables.

Table J.4- Backward Elimination Procedure for UTI

Backward Elimination Procedure

Class Level Information

Class Value Design Variables Etanercept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Adalimumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Anakinra 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Infliximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Rituximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Abatacept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

486

Tocilizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Golimumab 3 1 0 0 currently taking 0 1 0 never taking 0 0 1

Certolizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Folic Acid currently taking 1 0 never taking 0 1

Hydroxychloroquine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Sulphasalazine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Arava (Leflunomide) 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Azathioprine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Cyclosporin 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

487

Prednisolone 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

IM Gold injection 3 1 0 0 0

4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Penicillamine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Table J.5- Model Convergence status for UTI

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

488

Table J.6- Model Fit statistics for UTI

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 12856.104 12720.097 SC 12880.032 13964.367 -2 Log L 12850.104 12408.097

Table J.7- Testing null hypothesis for UTI

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 442.0070 153 <.0001 Score 494.1448 153 <.0001 Wald 442.5757 153 <.0001

489

Table J.8- Model Fit statistics status for removing covariant step 1

Step 1. Effect Abatacept is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table J.9- Model Fit statistics for removing covariant step 1

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 12856.104 12708.936 SC 12880.032 13881.421 -2 Log L 12850.104 12414.936

Table J.10- Testing Null hypothesis after removing covariant step 1

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 435.1680 144 <.0001 Score 486.6474 144 <.0001 Wald 438.5445 144 <.0001

490

Table J.11- Residual removing covariant step 1

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 5.2162 9 0.8151

Table J.12- Model Fit statistics for removing covariant step 2

Step 2. Effect Anakinra is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table J.13- Model Fit statistics after removing covariant step 2

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 12856.104 12698.658 SC 12880.032 13799.358 -2 Log L 12850.104 12422.658

491

Table J.14- Testing Null hypothesis after removing covariant step 2

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 427.4457 135 <.0001 Score 479.8413 135 <.0001 Wald 433.5090 135 <.0001

Table J.15- Residual removing covariant step 2

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 10.7595 18 0.9043

Table J.16- Model Fit statistics for removing covariant step 3 Step 3. Effect Certolizumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table J.17- Model Fit statistics after removing covariant step 3

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 12856.104 12695.973 SC 12880.032 13724.888 -2 Log L 12850.104 12437.973

Table J.18- Testing Null hypothesis after removing covariant step 3

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 412.1314 126 <.0001 Score 467.3671 126 <.0001 Wald 425.1267 126 <.0001 492

Table J.19- Residual removing covariant step 3

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 21.4657 27 0.7640

Table J.20- Model Fit statistics for removing covariant step 4 Step 4. Effect Golimumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

493

Table J.21- Model Fit statistics after removing covariant step 4

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 12856.104 12687.586 SC 12880.032 13668.645 -2 Log L 12850.104 12441.586

Table J.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 408.5177 120 <.0001 Score 463.4235 120 <.0001 Wald 421.3840 120 <.0001

Table J.23- Residual removing covariant step 4

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 25.7755 33 0.8106

494

Table J.24- Model Fit statistics for removing covariant step 5 Step 5. Effect Tocilizumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table J.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 12856.104 12680.185 SC 12880.032 13589.459 -2 Log L 12850.104 12452.185

Table J.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 397.9187 111 <.0001 Score 449.3858 111 <.0001 Wald 408.9965 111 <.0001

495

Table J.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 37.0911 42 0.6860

Table J.28- Model Fit statistics for removing covariant step 6 Step 6. Effect Sulphasalazine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table J.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 12856.104 12673.942 SC 12880.032 13511.431 -2 Log L 12850.104 12463.942

496

Table J.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 386.1623 102 <.0001 Score 438.4015 102 <.0001 Wald 398.7265 102 <.0001

Table J.31- Residual removing covariant step 6

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 47.9903 51 0.5939

Table J.32- Model Fit statistics for removing covariant step 7

Step 7. Effect Adalimumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

497

Table J.33- Model Fit statistics after removing covariant step 7

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 12856.104 12669.937 SC 12880.032 13435.642 -2 Log L 12850.104 12477.937

Table J.34- Testing Null hypothesis after removing covariant step 7

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 372.1669 93 <.0001 Score 422.5308 93 <.0001 Wald 384.7193 93 <.0001

Table J.35- Residual removing covariant step 7

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 62.5971 60 0.3842

498

Table J.36- Model Fit statistics for removing covariant step 8 Step 8. Effect Rituximab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table J.37- Model Fit statistics after removing covariant step 8

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 12856.104 12667.611 SC 12880.032 13361.531 -2 Log L 12850.104 12493.611

Table J.38- Testing Null hypothesis after removing covariant step 8

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 356.4931 84 <.0001 Score 401.5030 84 <.0001 Wald 367.1355 84 <.0001

499

Table J.39- Residual removing covariant step 8

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 80.3612 69 0.1649

Table J.40- Model Fit statistics for removing covariant step 9

Step 9. Effect Folic Acid is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table J.41- Model Fit statistics after removing covariant step 9

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 12856.104 12669.548 SC 12880.032 13339.540 -2 Log L 12850.104 12501.548

500

Table J.42- Testing Null hypothesis after removing covariant step 9

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 348.5557 81 <.0001 Score 394.3441 81 <.0001 Wald 359.6161 81 <.0001

Table J.43- Residual removing covariant step 9

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 87.7330 72 0.1001

Note: No (additional) effects met the 0.05 significance level for removal from the model.

Table J.44- Summary of backward elimination in UTI

Summary of Backward Elimination Effect Number Wald Variable Step Removed DF In Chi-Square Pr > ChiSq Label 1 Abatacept 9 17 2.9651 0.9657 Abatacept 2 Anakinra 9 16 3.6765 0.9314 3 Certolizumab 9 15 6.0565 0.7343 Certolizumab 4 Golimumab 6 14 4.1063 0.6623 Golimumab 5 Tocilizumab 9 13 9.3430 0.4062 Tocilizumab 6 Sulphasalazine 9 12 10.7076 0.2963 Sulphasalazine 7 Adalimumab 9 11 13.7461 0.1316 8 Rituximab 9 10 15.7336 0.0727 Rituximab 9 Folic Acid 3 9 7.4688 0.0584 Folic Acid

501

Table J.45- Type 3 analysis of effects in UTI

Type 3 Analysis of Effects Wald Effect DF Chi-Square Pr > ChiSq Etanercept 9 27.3183 0.0012 Infliximab 9 24.4209 0.0037 Hydroxychloroquine 9 20.7884 0.0136 Arava (Leflunomide) 9 22.1605 0.0084 Azathioprine 9 34.4145 <.0001 Cyclosporin 9 61.9727 <.0001 Prednisolone 9 56.1144 <.0001 IM Gold injection 9 26.8635 0.0015 Penicillamine 9 46.2679 <.0001

Table J.46- Analysis of maximum likelihood estimates in UTI

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfKidUri DF Estimate Error Chi-Square Pr > ChiSq Intercept Mild 1 -4.3801 0.1853 558.8895 <.0001 Intercept Mod 1 -3.6145 0.1223 872.9303 <.0001 Intercept Severe 1 -4.8991 0.2400 416.8122 <.0001 Arava (Leflunomide) 3 Mild 1 -0.2240 0.1564 2.0524 0.1520 Arava (Leflunomide) 3 Mod 1 0.1646 0.0988 2.7759 0.0957 Arava (Leflunomide) 3 Severe 1 0.0506 0.1776 0.0812 0.7757 Arava (Leflunomide) 4 Mild 1 -0.0668 0.7969 0.0070 0.9332 Arava (Leflunomide) 4 Mod 1 0.0573 0.4921 0.0135 0.9074 Arava (Leflunomide) 4 Severe 1 0.8420 0.6011 1.9621 0.1613 Arava (Leflunomide) currently taking Mild 1 -0.2083 0.1928 1.1666 0.2801 Arava (Leflunomide) currently taking Mod 1 -0.1359 0.1248 1.1862 0.2761 Arava (Leflunomide) currently taking Severe 1 -0.5347 0.2388 5.0129 0.0252 Arava (Leflunomide) never taking Mild 0 0 . . . Arava (Leflunomide) never taking Mod 0 0 . . .

502

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfKidUri DF Estimate Error Chi-Square Pr > ChiSq Arava (Leflunomide) never taking Severe 0 0 . . . Azathioprine 3 Mild 1 -0.5205 0.2460 4.4756 0.0344 Azathioprine 3 Mod 1 -0.7160 0.1596 20.1196 <.0001 Azathioprine 3 Severe 1 0.4816 0.2146 5.0389 0.0248 Azathioprine 4 Mild 1 -0.6989 0.6277 1.2400 0.2655 Azathioprine 4 Mod 1 0.1760 0.3073 0.3283 0.5667 Azathioprine 4 Severe 1 0.6065 0.4734 1.6415 0.2001 Azathioprine currently taking Mild 1 -11.8768 269.9 0.0019 0.9649 Azathioprine currently taking Mod 1 -0.8026 0.5911 1.8433 0.1746 Azathioprine currently taking Severe 1 -11.9304 256.6 0.0022 0.9629 Azathioprine never taking Mild 0 0 . . . Azathioprine never taking Mod 0 0 . . . Azathioprine never taking Severe 0 0 . . . Cyclosporin 3 Mild 1 0.0286 0.1771 0.0261 0.8717 Cyclosporin 3 Mod 1 0.1203 0.1042 1.3324 0.2484 Cyclosporin 3 Severe 1 -1.1810 0.2454 23.1502 <.0001 Cyclosporin 4 Mild 1 1.1876 0.4920 5.8263 0.0158 Cyclosporin 4 Mod 1 -0.5268 0.3731 1.9935 0.1580 Cyclosporin 4 Severe 1 -0.7341 0.5515 1.7717 0.1832 Cyclosporin currently taking Mild 1 1.5236 0.3766 16.3680 <.0001 Cyclosporin currently taking Mod 1 1.0570 0.2906 13.2255 0.0003 Cyclosporin currently taking Severe 1 0.6898 0.5322 1.6798 0.1950 Cyclosporin never taking Mild 0 0 . . . Cyclosporin never taking Mod 0 0 . . . Cyclosporin never taking Severe 0 0 . . . Etanercept 3 Mild 1 0.1829 0.1636 1.2501 0.2635 Etanercept 3 Mod 1 -0.2009 0.0961 4.3684 0.0366 Etanercept 3 Severe 1 0.3480 0.1590 4.7881 0.0287 Etanercept 4 Mild 1 -12.6298 554.8 0.0005 0.9818 Etanercept 4 Mod 1 -0.3715 0.8011 0.2151 0.6428 Etanercept 4 Severe 1 2.0503 0.6520 9.8884 0.0017 Etanercept currently taking Mild 1 0.3258 0.1452 5.0337 0.0249

503

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfKidUri DF Estimate Error Chi-Square Pr > ChiSq Etanercept currently taking Mod 1 -0.1108 0.0873 1.6099 0.2045 Etanercept currently taking Severe 1 -0.0992 0.1693 0.3431 0.5581 Etanercept never taking Mild 0 0 . . . Etanercept never taking Mod 0 0 . . . Etanercept never taking Severe 0 0 . . . Hydroxychloroquine 3 Mild 1 0.00780 0.1381 0.0032 0.9550 Hydroxychloroquine 3 Mod 1 0.3203 0.0836 14.6913 0.0001 Hydroxychloroquine 3 Severe 1 0.000272 0.1453 0.0000 0.9985 Hydroxychloroquine 4 Mild 1 -0.7822 1.0278 0.5792 0.4466 Hydroxychloroquine 4 Mod 1 -0.4848 0.4789 1.0250 0.3113 Hydroxychloroquine 4 Severe 1 -1.2246 1.0441 1.3755 0.2409 Hydroxychloroquine currently taking Mild 1 0.0696 0.1733 0.1611 0.6882 Hydroxychloroquine currently taking Mod 1 0.0797 0.1096 0.5288 0.4671 Hydroxychloroquine currently taking Severe 1 -0.1762 0.1963 0.8059 0.3693 Hydroxychloroquine never taking Mild 0 0 . . . Hydroxychloroquine never taking Mod 0 0 . . . Hydroxychloroquine never taking Severe 0 0 . . . IM Gold injection 3 Mild 1 0.4411 0.1444 9.3249 0.0023 IM Gold injection 3 Mod 1 0.1758 0.0890 3.9017 0.0482 IM Gold injection 3 Severe 1 0.1856 0.1582 1.3767 0.2407 IM Gold injection 4 Mild 1 0.2851 0.8555 0.1111 0.7389 IM Gold injection 4 Mod 1 0.7416 0.3618 4.2028 0.0404 IM Gold injection 4 Severe 1 -1.1430 0.9031 1.6021 0.2056 IM Gold injection currently taking Mild 1 -0.0266 0.7187 0.0014 0.9705 IM Gold injection currently taking Mod 1 0.4818 0.3327 2.0963 0.1477 IM Gold injection currently taking Severe 1 1.1145 0.4351 6.5630 0.0104 IM Gold injection never taking Mild 0 0 . . . IM Gold injection never taking Mod 0 0 . . . IM Gold injection never taking Severe 0 0 . . . Infliximab 3 Mild 1 -0.4458 0.2930 2.3149 0.1281 Infliximab 3 Mod 1 -0.3481 0.1641 4.5005 0.0339 Infliximab 3 Severe 1 0.6390 0.2074 9.4950 0.0021

504

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfKidUri DF Estimate Error Chi-Square Pr > ChiSq Infliximab 4 Mild 1 -0.9060 1.0372 0.7631 0.3824 Infliximab 4 Mod 1 -0.5810 0.5126 1.2847 0.2570 Infliximab 4 Severe 1 -0.1676 0.6351 0.0696 0.7919 Infliximab currently taking Mild 1 0.3314 0.3353 0.9770 0.3229 Infliximab currently taking Mod 1 0.0242 0.2202 0.0121 0.9123 Infliximab currently taking Severe 1 -1.4910 0.7397 4.0632 0.0438 Infliximab never taking Mild 0 0 . . . Infliximab never taking Mod 0 0 . . . Infliximab never taking Severe 0 0 . . . Penicillamine 3 Mild 1 0.6192 0.1785 12.0339 0.0005 Penicillamine 3 Mod 1 0.4545 0.1140 15.8882 <.0001 Penicillamine 3 Severe 1 0.0475 0.2175 0.0477 0.8271 Penicillamine 4 Mild 1 -0.4054 0.6593 0.3781 0.5386 Penicillamine 4 Mod 1 1.0189 0.2755 13.6737 0.0002 Penicillamine 4 Severe 1 1.3537 0.4485 9.1120 0.0025 Penicillamine currently taking Mild 1 -12.0460 451.7 0.0007 0.9787 Penicillamine currently taking Mod 1 -12.1211 270.8 0.0020 0.9643 Penicillamine currently taking Severe 1 -12.0786 478.6 0.0006 0.9799 Penicillamine never taking Mild 0 0 . . . Penicillamine never taking Mod 0 0 . . . Penicillamine never taking Severe 0 0 . . . Prednisolone 3 Mild 1 -0.0749 0.1901 0.1553 0.6935 Prednisolone 3 Mod 1 -0.00739 0.1231 0.0036 0.9521 Prednisolone 3 Severe 1 0.1364 0.2528 0.2909 0.5897 Prednisolone 4 Mild 1 1.6536 0.6799 5.9144 0.0150 Prednisolone 4 Mod 1 0.6808 0.5292 1.6549 0.1983 Prednisolone 4 Severe 1 1.2628 0.8600 2.1559 0.1420 Prednisolone currently taking Mild 1 -0.0322 0.1887 0.0291 0.8645 Prednisolone currently taking Mod 1 0.3588 0.1186 9.1561 0.0025 Prednisolone currently taking Severe 1 0.7760 0.2393 10.5125 0.0012 Prednisolone never taking Mild 0 0 . . . Prednisolone never taking Mod 0 0 . . .

505

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfKidUri DF Estimate Error Chi-Square Pr > ChiSq Prednisolone never taking Severe 0 0 . . .

Table J.47- Odds ratio estimates in UTI

Odds Ratio Estimates 95% Wald Effect InfKidUri Point Estimate Confidence Limits Etanercept 3 vs never taking Mild 1.201 0.871 1.655 Etanercept 3 vs never taking Mod 0.818 0.678 0.988 Etanercept 3 vs never taking Severe 1.416 1.037 1.934 Etanercept 4 vs never taking Mild <0.001 <0.001 >999.999 Etanercept 4 vs never taking Mod 0.690 0.143 3.316 Etanercept 4 vs never taking Severe 7.770 2.165 27.887 Etanercept currently taking vs never taking Mild 1.385 1.042 1.841 Etanercept currently taking vs never taking Mod 0.895 0.754 1.062 Etanercept currently taking vs never taking Severe 0.906 0.650 1.262 Infliximab 3 vs never taking Mild 0.640 0.361 1.137 Infliximab 3 vs never taking Mod 0.706 0.512 0.974 Infliximab 3 vs never taking Severe 1.895 1.262 2.845 Infliximab 4 vs never taking Mild 0.404 0.053 3.086 Infliximab 4 vs never taking Mod 0.559 0.205 1.528 Infliximab 4 vs never taking Severe 0.846 0.244 2.937 Infliximab currently taking vs never taking Mild 1.393 0.722 2.687 Infliximab currently taking vs never taking Mod 1.025 0.665 1.578 Infliximab currently taking vs never taking Severe 0.225 0.053 0.960 Hydroxychloroquine 3 vs never taking Mild 1.008 0.769 1.321 Hydroxychloroquine 3 vs never taking Mod 1.377 1.169 1.623 Hydroxychloroquine 3 vs never taking Severe 1.000 0.752 1.330 Hydroxychloroquine 4 vs never taking Mild 0.457 0.061 3.429 Hydroxychloroquine 4 vs never taking Mod 0.616 0.241 1.574 Hydroxychloroquine 4 vs never taking Severe 0.294 0.038 2.275

506

Odds Ratio Estimates 95% Wald Effect InfKidUri Point Estimate Confidence Limits Hydroxychloroquine currently taking vs never taking Mild 1.072 0.763 1.506 Hydroxychloroquine currently taking vs never taking Mod 1.083 0.874 1.343 Hydroxychloroquine currently taking vs never taking Severe 0.838 0.571 1.232 Arava (Leflunomide) 3 vs never taking Mild 0.799 0.588 1.086 Arava (Leflunomide) 3 vs never taking Mod 1.179 0.971 1.431 Arava (Leflunomide) 3 vs never taking Severe 1.052 0.743 1.490 Arava (Leflunomide) 4 vs never taking Mild 0.935 0.196 4.460 Arava (Leflunomide) 4 vs never taking Mod 1.059 0.404 2.778 Arava (Leflunomide) 4 vs never taking Severe 2.321 0.715 7.540 Arava (Leflunomide) currently taking vs never taking Mild 0.812 0.556 1.185 Arava (Leflunomide) currently taking vs never taking Mod 0.873 0.683 1.115 Arava (Leflunomide) currently taking vs never taking Severe 0.586 0.367 0.936 Azathioprine 3 vs never taking Mild 0.594 0.367 0.962 Azathioprine 3 vs never taking Mod 0.489 0.357 0.668 Azathioprine 3 vs never taking Severe 1.619 1.063 2.465 Azathioprine 4 vs never taking Mild 0.497 0.145 1.701 Azathioprine 4 vs never taking Mod 1.192 0.653 2.178 Azathioprine 4 vs never taking Severe 1.834 0.725 4.638 Azathioprine currently taking vs never taking Mild <0.001 <0.001 >999.999 Azathioprine currently taking vs never taking Mod 0.448 0.141 1.428 Azathioprine currently taking vs never taking Severe <0.001 <0.001 >999.999 Cyclosporin 3 vs never taking Mild 1.029 0.727 1.456 Cyclosporin 3 vs never taking Mod 1.128 0.919 1.383 Cyclosporin 3 vs never taking Severe 0.307 0.190 0.497 Cyclosporin 4 vs never taking Mild 3.279 1.250 8.602 Cyclosporin 4 vs never taking Mod 0.590 0.284 1.227 Cyclosporin 4 vs never taking Severe 0.480 0.163 1.415 Cyclosporin currently taking vs never taking Mild 4.589 2.193 9.599 Cyclosporin currently taking vs never taking Mod 2.878 1.628 5.087 Cyclosporin currently taking vs never taking Severe 1.993 0.702 5.658 Prednisolone 3 vs never taking Mild 0.928 0.639 1.347 Prednisolone 3 vs never taking Mod 0.993 0.780 1.263

507

Odds Ratio Estimates 95% Wald Effect InfKidUri Point Estimate Confidence Limits Prednisolone 3 vs never taking Severe 1.146 0.698 1.881 Prednisolone 4 vs never taking Mild 5.226 1.378 19.811 Prednisolone 4 vs never taking Mod 1.975 0.700 5.573 Prednisolone 4 vs never taking Severe 3.535 0.655 19.076 Prednisolone currently taking vs never taking Mild 0.968 0.669 1.402 Prednisolone currently taking vs never taking Mod 1.432 1.135 1.806 Prednisolone currently taking vs never taking Severe 2.173 1.359 3.473 IM Gold injection 3 vs never taking Mild 1.554 1.171 2.063 IM Gold injection 3 vs never taking Mod 1.192 1.001 1.419 IM Gold injection 3 vs never taking Severe 1.204 0.883 1.642 IM Gold injection 4 vs never taking Mild 1.330 0.249 7.113 IM Gold injection 4 vs never taking Mod 2.099 1.033 4.266 IM Gold injection 4 vs never taking Severe 0.319 0.054 1.872 IM Gold injection currently taking vs never taking Mild 0.974 0.238 3.983 IM Gold injection currently taking vs never taking Mod 1.619 0.843 3.108 IM Gold injection currently taking vs never taking Severe 3.048 1.299 7.151 Penicillamine 3 vs never taking Mild 1.857 1.309 2.636 Penicillamine 3 vs never taking Mod 1.575 1.260 1.970 Penicillamine 3 vs never taking Severe 1.049 0.685 1.606 Penicillamine 4 vs never taking Mild 0.667 0.183 2.427 Penicillamine 4 vs never taking Mod 2.770 1.614 4.754 Penicillamine 4 vs never taking Severe 3.872 1.608 9.325 Penicillamine currently taking vs never taking Mild <0.001 <0.001 >999.999 Penicillamine currently taking vs never taking Mod <0.001 <0.001 >999.999 Penicillamine currently taking vs never taking Severe <0.001 <0.001 >999.999

508

APPENDIX K: OUTPUT OF SAS FOR VIRAL INFECTION

Table K.1- Complete statistics for viral infection

Model Information Data Set WORK.IMPORT2 Response Variable InfVir InfVir Number of Response Levels 4 Model generalized logit Optimization Technique Newton-Raphson

Table K.2- Observation status for VIRAL INFECTION

Number of Observations Read 27711 Number of Observations Used 21506

Table K.3- response value for VIRAL INFECTION

Response Profile Ordered Total Value InfVir Frequency 1 1 435 2 2 837 3 3 305 4 4 19929 0 .

509

Logits modelled use InfVir='4' as the reference category.

Note: 6205 observations were deleted due to missing values for the response or explanatory variables.

Note1 response level was deleted due to missing or invalid values for its explanatory, frequency, or weight variables

Table K.4- Backward Elimination Procedure for VIRAL INFECTION

Backward Elimination Procedure

Class Level Information

Class Value Design Variables Etanercept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Adalimumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Anakinra 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Infliximab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Rituximab 3 1 0 0 0 4 0 1 0 0

510

currently taking 0 0 1 0 never taking 0 0 0 1

Abatacept 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Tocilizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Golimumab 3 1 0 0 currently taking 0 1 0 never taking 0 0 1

Certolizumab 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Folic Acid currently taking 1 0 never taking 0 1

Hydroxychloroquine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Sulphasalazine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Arava (Leflunomide) 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Azathioprine 3 1 0 0 0 4 0 1 0 0

511

currently taking 0 0 1 0 never taking 0 0 0 1

Cyclosporin 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Prednisolone 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

IM Gold injection 3 1 0 0 0

4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

Penicillamine 3 1 0 0 0 4 0 1 0 0 currently taking 0 0 1 0 never taking 0 0 0 1

512

Table K.5- Model Convergence status for VIRAL INFECTION

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.6- Model Fit statistics for VIRAL INFECTION

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14470.560 SC 14489.267 15714.830 -2 Log L 14459.339 14158.560

Table K.7- Testing null hypothesis for VIRAL INFECTION

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 300.7784 153 <.0001 Score 331.3978 153 <.0001 Wald 311.3192 153 <.0001

513

Table K.8- Model Fit statistics for removing covariant step 1

Step 1. Effect Penicillamine is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.9- Model Fit statistics for removing covariant step 1

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14462.278 SC 14489.267 15634.763 -2 Log L 14459.339 14168.278

Table K.10- Testing Null hypothesis after removing covariant step 1

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 291.0609 144 <.0001 Score 325.8514 144 <.0001 Wald 306.2753 144 <.0001

514

Table K.11- Residual removing covariant step 1

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 7.0603 9 0.6308

Table K.12- Model Fit statistics for removing covariant step 2

Step 2. Effect Certolizumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.13- Model Fit statistics after removing covariant step 2

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14452.290 SC 14489.267 15552.990 -2 Log L 14459.339 14176.290

515

Table K.14- Testing Null hypothesis after removing covariant step 2

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 283.0486 135 <.0001 Score 316.2107 135 <.0001 Wald 297.3478 135 <.0001

Table K.15- Residual removing covariant step 2

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 15.7431 18 0.6105

Table K.16- Model Fit statistics for removing covariant step 3

Step 3. Effect Golimumab is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

516

Table K.17- Model Fit statistics after removing covariant step 3

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14445.753 SC 14489.267 15498.596 -2 Log L 14459.339 14181.753

Table K.18- Testing Null hypothesis after removing covariant step 3

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 277.5858 129 <.0001 Score 311.0451 129 <.0001 Wald 292.3479 129 <.0001

Table K.19- Residual removing covariant step 3

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 21.2000 24 0.6269

517

Table K.20- Model Fit statistics for removing covariant step 4

Step 4. Effect Arava (Leflunomide) is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.21- Model Fit statistics after removing covariant step 4

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14443.433 SC 14489.267 15424.492 -2 Log L 14459.339 14197.433

Table K.22- Testing Null hypothesis after removing covariant step 4

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 261.9054 120 <.0001 Score 299.6898 120 <.0001 Wald 274.5846 120 <.0001

518

Table K.23- Residual removing covariant step 4

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 33.6212 33 0.4372

Table K.24- Model Fit statistics for removing covariant step 5 Step 5. Effect Abatacept is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.25- Model Fit statistics after removing covariant step 5

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14435.111 SC 14489.267 15344.385 -2 Log L 14459.339 14207.111

Table K.26- Testing Null hypothesis after removing covariant step 5

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 252.2274 111 <.0001 Score 291.0613 111 <.0001 Wald 266.0126 111 <.0001

Table K.27- Residual removing covariant step 5

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 44.0788 42 0.3837

519

Table K.28- Model Fit statistics for removing covariant step 6

Step 6. Effect IM Gold injection is removed: Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.29- Model Fit statistics after removing covariant step 6

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14427.297 SC 14489.267 15264.786 -2 Log L 14459.339 14217.297

520

Table K.30- Testing Null hypothesis after removing covariant step 6

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 242.0415 102 <.0001 Score 278.2277 102 <.0001 Wald 255.0262 102 <.0001

Table K.31- Residual removing covariant step 6

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 55.5679 51 0.3068

Step 7. Effect Azathioprine is removed:

Table K.32- Model Fit statistics for removing covariant step 7

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

521

Table K.33- Model Fit statistics after removing covariant step 7

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14422.480 SC 14489.267 15188.185 -2 Log L 14459.339 14230.480

Table K.34- Testing Null hypothesis after removing covariant step 7

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 228.8582 93 <.0001 Score 264.9909 93 <.0001 Wald 242.7767 93 <.0001

Table K.35- Residual removing covariant step 7

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 67.4931 60 0.2365

Step 8. Effect Sulphasalazine is removed: Table K.36- Model Fit statistics for removing covariant step 8 Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.37- Model Fit statistics after removing covariant step 8

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14413.609

522

Model Fit Statistics Criterion Intercept Only Intercept and Covariates SC 14489.267 15107.528 -2 Log L 14459.339 14239.609

Table K.38- Testing Null hypothesis after removing covariant step 8

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 219.7299 84 <.0001 Score 255.3106 84 <.0001 Wald 232.3624 84 <.0001

Table K.39- Residual removing covariant step 8

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 78.7472 69 0.1977

523

Step 9. Effect Anakinra is removed:

Table K.40- Model Fit statistics for removing covariant step 9

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.41- Model Fit statistics after removing covariant step 9

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14406.926 SC 14489.267 15029.060 -2 Log L 14459.339 14250.926

Table K.42- Testing Null hypothesis after removing covariant step 9

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 208.4131 75 <.0001 Score 242.5271 75 <.0001 Wald 221.0994 75 <.0001

524

Table K.43- Residual removing covariant step 9

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 88.8401 78 0.1885

Step 10. Effect Tocilizumab is removed:

Table K.44- Model Fit statistics for removing covariant step 10 Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.45- Model Fit statistics after removing covariant step 10

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14402.102 SC 14489.267 14952.452 -2 Log L 14459.339 14264.102

525

Table K.46- Testing Null hypothesis after removing covariant step 10

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 195.2369 66 <.0001 Score 227.9013 66 <.0001 Wald 207.5631 66 <.0001

Table K.47- Residual removing covariant step 10

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 101.7314 87 0.1337

Step 11. Effect Adalimumab is removed:

Table K.48- Model Fit statistics for removing covariant step 11

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.49- Model Fit statistics after removing covariant step 11

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14399.250 SC 14489.267 14877.815 -2 Log L 14459.339 14279.250 Table K.50- Testing Null hypothesis after removing covariant step 11

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 180.0886 57 <.0001

526

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Score 211.6551 57 <.0001 Wald 191.8318 57 <.0001

Table K.51- Residual removing covariant step 11

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 116.7690 96 0.0735

Step 12. Effect Rituximab is removed:

Table K.52- Model Fit statistics for removing covariant step 12

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

Table K.53- Model Fit statistics for removing covariant step 12

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14402.145 SC 14489.267 14808.925 -2 Log L 14459.339 14300.145

Table K.54- Testing Null hypothesis after removing covariant step 12

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 159.1937 48 <.0001 Score 188.7845 48 <.0001 Wald 170.0883 48 <.0001

527

Table K.55- Residual removing covariant step 12

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 134.6372 105 0.0272

Step 13. Effect Infliximab is removed:

Table K.56- Model Fit statistics for removing covariant step 13

Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied.

528

Table K.57- Model Fit statistics after removing covariant step 13

Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 14465.339 14396.196 SC 14489.267 14731.191 -2 Log L 14459.339 14312.196

Table K.58- Testing Null hypothesis after removing covariant step 13

Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 147.1431 39 <.0001 Score 173.8189 39 <.0001 Wald 156.0719 39 <.0001

Table K.59- Residual removing covariant step 13

Residual Chi-Square Test Chi-Square DF Pr > ChiSq 150.7370 114 0.0121

Note: No (additional) effects met the 0.05 significance level for removal from the model.

529

Table K.60 - Summary of backward elimination in VIRAL INFECTION

Summary of Backward Elimination Effect Number Wald Variable Step Removed DF In Chi-Square Pr > ChiSq Label 1 Penicillamine 9 17 4.9471 0.8389 Penicillamine 2 Certolizumab 9 16 8.1929 0.5148 Certolizumab 3 Golimumab 6 15 5.4697 0.4851 Golimumab 4 Arava (Leflunomide) 9 14 7.7065 0.5640 Arava (Leflunomide) 5 Abatacept 9 13 10.0331 0.3478 Abatacept 6 IM Gold injection 9 12 10.9493 0.2792 IM Gold injection 7 Azathioprine 9 11 11.2061 0.2618 Azathioprine 8 Sulphasalazine 9 10 9.9071 0.3581 Sulphasalazine 9 Anakinra 9 9 10.1434 0.3390 10 Tocilizumab 9 8 11.9686 0.2151 Tocilizumab 11 Adalimumab 9 7 15.0634 0.0892 12 Rituximab 9 6 16.7000 0.0536 Rituximab 13 Infliximab 9 5 13.1407 0.1563

Table K.61- Type 3 analysis of effects in VIRAL INFECTION

Type 3 Analysis of Effects Wald Effect DF Chi-Square Pr > ChiSq Etanercept 9 43.9767 <.0001 Folic Acid 3 13.7765 0.0032 Hydroxychloroquine 9 29.2958 0.0006 Cyclosporin 9 29.3135 0.0006 Prednisolone 9 25.6360 0.0023

530

Table K.62- Analysis of maximum likelihood estimates in VIRAL INFECTION

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfVir DF Estimate Error Chi-Square Pr > ChiSq Intercept 1 1 -4.0046 0.1449 763.8821 <.0001 Intercept 2 1 -3.7221 0.1215 938.7083 <.0001 Intercept 3 1 -4.9163 0.2170 513.2153 <.0001 Etanercept 3 1 1 0.0300 0.1317 0.0520 0.8196 Etanercept 3 2 1 -0.0552 0.0924 0.3574 0.5499 Etanercept 3 3 1 -0.0276 0.1511 0.0334 0.8550 Etanercept 4 1 1 0.8042 0.7503 1.1486 0.2838 Etanercept 4 2 1 1.4693 0.3785 15.0703 0.0001 Etanercept 4 3 1 2.4269 0.4611 27.6982 <.0001 Etanercept currently taking 1 1 0.2158 0.1135 3.6129 0.0573 Etanercept currently taking 2 1 -0.0696 0.0858 0.6574 0.4175 Etanercept currently taking 3 1 0.0166 0.1397 0.0142 0.9053 Etanercept never taking 1 0 0 . . . Etanercept never taking 2 0 0 . . . Etanercept never taking 3 0 0 . . . Folic Acid currently taking 1 1 -0.2782 0.1215 5.2471 0.0220 Folic Acid currently taking 2 1 -0.2498 0.0874 8.1597 0.0043 Folic Acid currently taking 3 1 -0.1302 0.1385 0.8845 0.3470 Folic Acid never taking 1 0 0 . . . Folic Acid never taking 2 0 0 . . . Folic Acid never taking 3 0 0 . . . Hydroxychloroquine 3 1 1 0.2452 0.1114 4.8438 0.0277 Hydroxychloroquine 3 2 1 0.3634 0.0826 19.3661 <.0001 Hydroxychloroquine 3 3 1 0.2016 0.1322 2.3252 0.1273 Hydroxychloroquine 4 1 1 -0.2717 0.7210 0.1420 0.7063 Hydroxychloroquine 4 2 1 0.7413 0.3267 5.1494 0.0233 Hydroxychloroquine 4 3 1 0.4258 0.5950 0.5121 0.4742 Hydroxychloroquine currently taking 1 1 0.2225 0.1394 2.5473 0.1105 Hydroxychloroquine currently taking 2 1 0.2648 0.1050 6.3532 0.0117 Hydroxychloroquine currently taking 3 1 0.1380 0.1700 0.6594 0.4168 Hydroxychloroquine never taking 1 0 0 . . . 531

Analysis of Maximum Likelihood Estimates Standard Wald Parameter InfVir DF Estimate Error Chi-Square Pr > ChiSq Hydroxychloroquine never taking 2 0 0 . . . Hydroxychloroquine never taking 3 0 0 . . . Cyclosporin 3 1 1 -0.1466 0.1488 0.9698 0.3247 Cyclosporin 3 2 1 0.3263 0.0938 12.0966 0.0005 Cyclosporin 3 3 1 0.2517 0.1518 2.7469 0.0974 Cyclosporin 4 1 1 -0.4712 0.5218 0.8156 0.3665 Cyclosporin 4 2 1 0.3039 0.2642 1.3229 0.2501 Cyclosporin 4 3 1 -1.0400 0.7326 2.0151 0.1557 Cyclosporin currently taking 1 1 0.000418 0.5873 0.0000 0.9994 Cyclosporin currently taking 2 1 0.9855 0.2950 11.1579 0.0008 Cyclosporin currently taking 3 1 0.0567 0.7165 0.0063 0.9369 Cyclosporin never taking 1 0 0 . . . Cyclosporin never taking 2 0 0 . . . Cyclosporin never taking 3 0 0 . . . Prednisolone 3 1 1 0.1061 0.1497 0.5023 0.4785 Prednisolone 3 2 1 0.4422 0.1240 12.7089 0.0004 Prednisolone 3 3 1 0.6023 0.2242 7.2156 0.0072 Prednisolone 4 1 1 0.6125 0.7489 0.6690 0.4134 Prednisolone 4 2 1 0.0692 0.6535 0.0112 0.9156 Prednisolone 4 3 1 -9.4512 139.2 0.0046 0.9459 Prednisolone currently taking 1 1 -0.00427 0.1498 0.0008 0.9772 Prednisolone currently taking 2 1 0.3461 0.1239 7.8094 0.0052 Prednisolone currently taking 3 1 0.7363 0.2202 11.1834 0.0008 Prednisolone never taking 1 0 0 . . . Prednisolone never taking 2 0 0 . . . Prednisolone never taking 3 0 0 . . .

532

Table K.63- Odds ratio estimates in VIRAL INFECTION

Odds Ratio Estimates 95% Wald Effect InfVir Point Estimate Confidence Limits Etanercept 3 vs never taking 1 1.030 0.796 1.334 Etanercept 3 vs never taking 2 0.946 0.790 1.134 Etanercept 3 vs never taking 3 0.973 0.723 1.308 Etanercept 4 vs never taking 1 2.235 0.514 9.726 Etanercept 4 vs never taking 2 4.346 2.070 9.126 Etanercept 4 vs never taking 3 11.323 4.586 27.957 Etanercept currently taking vs never taking 1 1.241 0.993 1.550 Etanercept currently taking vs never taking 2 0.933 0.788 1.104 Etanercept currently taking vs never taking 3 1.017 0.773 1.337 Folic Acid currently taking vs never taking 1 0.757 0.597 0.961 Folic Acid currently taking vs never taking 2 0.779 0.656 0.925 Folic Acid currently taking vs never taking 3 0.878 0.669 1.152 Hydroxychloroquine 3 vs never taking 1 1.278 1.027 1.590 Hydroxychloroquine 3 vs never taking 2 1.438 1.223 1.691 Hydroxychloroquine 3 vs never taking 3 1.223 0.944 1.585 Hydroxychloroquine 4 vs never taking 1 0.762 0.185 3.131 Hydroxychloroquine 4 vs never taking 2 2.099 1.106 3.982 Hydroxychloroquine 4 vs never taking 3 1.531 0.477 4.914 Hydroxychloroquine currently taking vs never taking 1 1.249 0.951 1.642 Hydroxychloroquine currently taking vs never taking 2 1.303 1.061 1.601 Hydroxychloroquine currently taking vs never taking 3 1.148 0.823 1.602 Cyclosporin 3 vs never taking 1 0.864 0.645 1.156 Cyclosporin 3 vs never taking 2 1.386 1.153 1.665 Cyclosporin 3 vs never taking 3 1.286 0.955 1.732 Cyclosporin 4 vs never taking 1 0.624 0.225 1.736 Cyclosporin 4 vs never taking 2 1.355 0.807 2.275 Cyclosporin 4 vs never taking 3 0.353 0.084 1.486 Cyclosporin currently taking vs never taking 1 1.000 0.316 3.163 Cyclosporin currently taking vs never taking 2 2.679 1.503 4.777 Cyclosporin currently taking vs never taking 3 1.058 0.260 4.310

533

Odds Ratio Estimates 95% Wald Effect InfVir Point Estimate Confidence Limits Prednisolone 3 vs never taking 1 1.112 0.829 1.491 Prednisolone 3 vs never taking 2 1.556 1.220 1.984 Prednisolone 3 vs never taking 3 1.826 1.177 2.834 Prednisolone 4 vs never taking 1 1.845 0.425 8.007 Prednisolone 4 vs never taking 2 1.072 0.298 3.858 Prednisolone 4 vs never taking 3 <0.001 <0.001 >999.999 Prednisolone currently taking vs never taking 1 0.996 0.742 1.336 Prednisolone currently taking vs never taking 2 1.414 1.109 1.802 Prednisolone currently taking vs never taking 3 2.088 1.356 3.215

534

APPENDIX L: ETHICAL APPROVAL FOR THE THESIS

535

APPENDIX M: SAMPLE OF ARAD QUESTIONNAIRE

536

537

538

539

540

541

542

543

544

545

546

547

548

549

550

551

552

553

554

555

556

557

558

559

560