Identifying and Quantifying Dynamic Risk Factors for Coronary Artery Disease in Systemic Lupus Erythematosus

Transcription

1 Identifying and Quantifying Dynamic Risk Factors for Coronary Artery Disease in Systemic Lupus Erythematosus by Mandana Nikpour A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Institute of Medical Science University of Toronto Copyright by Mandana Nikpour 2011

2 ii Identifying and Quantifying Dynamic Risk Factors for Coronary Artery Disease in Systemic Lupus Erythematosus Abstract Mandana Nikpour Doctor of Philosophy Institute of Medical Science University of Toronto 2011 Systemic Lupus Erythematosus (SLE), a prototypic multi-organ autoimmune disease, is associated with a dramatically increased risk of coronary artery disease (CAD) manifesting as angina, myocardial infarction and sudden cardiac death. Traditional cardiac risk factors such as hypertension and hypercholesterolemia, measured at baseline in accordance with the Framingham model, only partially account for the increased risk of CAD in SLE. In this thesis, I have shown that blood pressure (BP), lipids and novel risk factors such as the inflammatory marker high-sensitivity C-reactive protein (hscrp), take a dynamic course in SLE, with more than half of the variance in serial measurements over time occurring within rather than between individuals. This variability is due to changes in disease activity, treatment, accrual of other cardiac risk factors, and complications such as infection. I have demonstrated that by capturing cumulative ii

3 exposure over time, summary measures such as arithmetic mean and time-adjusted mean (AM) are better able to quantify CAD risk in patients with SLE than single-pointin-time measurements of risk factors. By incorporating summary measures such as mean and AM into time-dependent covariate survival analysis models, I was able to quantify the magnitude of increase in CAD risk associated with increments in systolic and diastolic BP, and to demonstrate and quantify the association between several lipids / lipoproteins and CAD risk in SLE. Using this methodology, I was also able to demonstrate that despite marked variability over time, summary measures of hscrp are independently predictive of CAD risk among patients with SLE, highlighting the pivotal role of inflammation in atherosclerosis. Furthermore, I was able to determine lipid and hscrp cut-points that will aid clinicians in identifying a subgroup of patients with SLE who are at significantly increased cardiac risk. iii iii

4 iv Acknowledgments Foremost, I wish to thank my supervisors Dr Murray Urowitz and Dr Dafna Gladman for their boundless support and guidance. They are wonderful teachers, mentors and role models, and I hope to make them proud. Sincere thanks also to Dr Paula Harvey for supervision from a cardiology perspective, and for providing a unique and comforting Australian-Canadian point of view. Special thanks to Dominique Ibanez for her assistance with statistical analyses, for being a great source of inspiration, and for her friendship. I wish to also thank Dr Richard Cook for his help with planning the statistical analyses for my project and Anne Mackinnon for her kindness, generosity and invaluable help with the administrative aspects of my graduate studies. I am enormously grateful to the Centre for Prognosis Studies in The Rheumatic Diseases, The Smythe Foundation, Ontario Lupus Association, Lupus Flare Foundation and The Lupus Society of Alberta for providing financial support for my work. I was very fortunate to hold Arthritis Centre of Excellence and Geoff Carr Lupus Fellowships during my candidacy. I dedicate this work to Marc, Silas, Esau, and the other members of my wonderful family, with love. iv

5 v Thesis Publications Published papers: Mandana Nikpour, Dafna D Gladman, Dominique Ibanez, Paula J Harvey and Murray B Urowitz. Variability over time and correlates of cholesterol and blood pressure in systemic lupus erythematosus: a longitudinal cohort study. Arthritis Research & Therapy. 2010;12(3):R125. Epub 2010 Jun 30. Mandana Nikpour, Dafna D. Gladman, Dominique Ibañez, Murray B. Urowitz. Variability and Correlates of High Sensitivity C-reactive Protein in Systemic Lupus Erythematosus. Lupus. 2009;18(11): Mandana Nikpour, Murray B. Urowitz, Dafna D. Gladman. Epidemiology of Atherosclerosis in Systemic Lupus Erythematosus. Current Rheumatology Reports. 2009;11(4): Mandana Nikpour, Murray B. Urowitz, Dafna D. Gladman. Premature Atherosclerosis in Systemic Lupus Erythematosus. Rheumatic Disease Clinics of North America 2005;31(2): Manuscripts submitted for publication: Mandana Nikpour, Murray B Urowitz, Dominique Ibanez, Paula J Harvey and Dafna D Gladman. Optimal Frequency of Cardiovascular Risk Assessment for Modelling of Coronary Events in Systemic Lupus Erythematosus. Mandana Nikpour, Murray B Urowitz, Dominique Ibanez, Paula J Harvey and Dafna D Gladman. Importance of Remote, Recent and Cumulative Exposure to Risk Factors for Atherosclerotic Coronary Artery Disease in Systemic Lupus Erythematosus. Mandana Nikpour, Dafna D Gladman, Dominique Ibanez, Paula J Harvey and Murray B Urowitz. Low-density Lipoprotein Cholesterol, Total cholesterol to High-density Lipoprotein Cholesterol Ratio and Triglycerides as Risk Factors for Atherosclerotic Coronary Artery Disease in Systemic Lupus Erythematosus. Mandana Nikpour, Dafna D Gladman, Dominique Ibanez, Paula J Harvey and Murray B Urowitz. High-sensitivity C-reactive Protein as an Independent Risk Factor for Coronary Artery Disease in Systemic Autoimmune Disease: the Systemic Lupus Erythematosus Cardiac Risk Evaluation Study. v

6 vi Table of Contents Acknowledgments...iv Thesis Publications...v Table of Contents...vi List of Tables...xviii List of Figures...xxii List of Abbreviations...xxii Chapter Introduction The association of SLE with premature coronary artery disease (CAD) Case series Cohort studies Community based studies The link between coronary and non-coronary atherosclerosis in SLE Cardiac risk factors in SLE Longitudinal cohort studies of cardiac risk factors in SLE Traditional cardiac risk factors in the general population Sex and age Family history Diabetes mellitus Hyperlipidemia Total cholesterol (TC) Lipoprotein measurement Low-density lipoprotein cholesterol (LDL-C)...12 vi

7 vii High density lipoprotein cholesterol (HDL-C) Total cholesterol to HDL-C ratio (TC:HDL-C) Triglycerides (TG) Hypertension Goal blood pressure Prehypertension and borderline hypertension White coat hypertension Interaction between SBP and TC Role of traditional cardiac risk factors in SLE Assessment of lipid risk factors in SLE Hypertension and CAD in SLE Methodological considerations in studies of coronary risk factors in SLE Time-to-event analysis using time-dependent covariates Summarising dynamic cardiac risk factors over time Use of summary measures of cardiac risk factors in a time-dependent survival model Non-traditional risk factors for CAD in SLE High sensitivity C-reactive protein (hscrp) SLE disease activity measures SLE disease and treatment related factors Markers of subclinical CAD in SLE Myocardial perfusion imaging The relationship between subclinical and clinical CAD Summary of background points of key relevance to thesis...37 vii

8 viii Table Summary of risk factors associated with the development of clinical coronary artery disease in selected studies...42 Figure Model for estimating the ten-year risk of coronary artery disease using data from the Framingham Heart Study...43 Table NCEP-ATP III classification of total cholesterol in relation to CAD risk...44 Table NCEP-ATP III recommended goals for LDL-C...45 Figure SLE Disease Activity Index 2000 (SLEDAI-2K)...46 Chapter Hypothesis, Objectives and Overall Framework of Thesis Hypothesis Objectives Objective Objective Objective Objective Objective Objective Summary of thesis objectives Overall framework of thesis...49 Figure A depiction of the overall framework of the thesis...50 Chapter Setting The University of Toronto lupus clinic and the lupus database Patients and visits Collection of demographic and disease-related data The lupus database...53 viii

9 3.1.4 Deceased patients Documentation of cardiac risk factors Definition and recording of cardiac events Missing data Gaps between visits Loss to follow-up...58 Table Characteristics of patients in the University of Toronto Lupus Database as of August Table Data collection characteristics for blood pressure, total cholesterol, lowdensity lipoprotein cholesterol and high sensitivity C-reactive protein...62 Chapter Variability over Time and Correlates of Total Cholesterol, Systolic and Diastolic Blood Pressure in SLE Abstract Rationale and objective Patients Methods TC, SBP and DBP and other variables Statistical analysis Results Discussion Conclusion...73 Table Characteristics of patients (n=1260)...74 Table Number, frequency and values of total cholesterol, systolic blood pressure and diastolic blood pressure measurements...76 Table Proportion of patients with normal and elevated total cholesterol, systolic blood pressure and diastolic blood pressure at baseline and during follow-up...77 ix ix

10 Table Total, between and within patient variance in total cholesterol, systolic blood pressure and diastolic blood pressure during follow-up...78 Table Independent correlates of total cholesterol...79 Table Independent correlates of total cholesterol in women only...80 Table Independent correlates of systolic blood pressure...82 Table Independent correlates of systolic blood pressure in women only...83 Table Independent correlates of diastolic blood pressure...85 Table Independent correlates of diastolic blood pressure in women only...86 Chapter Approaches to Summarizing Dynamic Cardiac Risk Factors in SLE: The Examples of Cholesterol and Blood Pressure Abstract Rationale and objective Patients Methods Measurement of TC, SBP and DBP Coronary events and event visits Mean Time-adjusted mean (AM) Area-under-the-curve (AUC) Correlations Gap analysis Results Summary measures for TC Summary measures for BP Gap analysis...95 x x

11 5.6 Discussion Conclusion Figure Schematic representation of hypothetical examples of individual patient data used for calculation of summary measures Figure Diagrammatic representation of the calculation of a time-adjusted mean Figure Diagrammatic representation of visits used to calculate summary measures for each of TC, SBP and DBP 106 Figure Depiction of the method used to perform gap analysis Table Patient characteristics Table Frequency and values of total cholesterol measurements Table Frequency and values of systolic and diastolic blood pressure measurements Table Correlation matrix for total cholesterol Table Correlation matrix for systolic blood pressure Table Correlation matrix for diastolic blood pressure Figure Absolute difference between true and evaluated mean cholesterol Figure Absolute difference between true and evaluated AM cholesterol Figure Percent difference between true and evaluated AUC cholesterol Figure Absolute difference between true and evaluated mean systolic blood pressure Figure Absolute difference between true and evaluated AM systolic blood pressure Figure Percent difference between true and evaluated AUC systolic blood pressure Figure Absolute difference between true and evaluated mean diastolic blood pressure Figure Absolute difference between true and evaluated AM diastolic blood pressure xi xi

12 Figure Percent difference between true and evaluated AUC diastolic blood pressure Chapter Importance of Remote, Recent and Cumulative Exposure to Risk Factors for Atherosclerotic Coronary Artery Disease in SLE Abstract Rationale and objective Patients Methods Calculation of summary measures Predictor variables and other covariates Interaction terms Outcome variables and outcome visits Univariate comparisons Proportionality of hazard Cox models Time-dependent covariate models Additional models Results Univariate comparisons Proportional hazards multiple regression models TC Models SBP Models DBP models Discussion Conclusions xii xii

13 xiii Table Univariate comparison of patients with and without coronary events used in TC models Table Proportional hazards models for coronary events using various measures of cholesterol Table Proportional hazards models for coronary events using various measures of cholesterol, including only significant covariates and interaction terms Table Proportional hazards models for coronary events using various measures of cholesterol defined dichotomously Table Proportional hazards models for coronary events using various measures of systolic blood pressure Table Proportional hazards models for coronary events using various measures of systolic blood pressure, including only significant covariates Table Proportional hazards models for coronary events using various measures of diastolic blood pressure Table Proportional hazards models for coronary events using various measures of diastolic blood pressure, including only significant covariates Table Proportional hazards models for coronary events using various measures of blood pressure defined dichotomously Chapter Application of Cumulative-Exposure Measures to Lipid and Lipoprotein Markers of Coronary Risk in SLE Abstract Rationale and objective Patients Methods Calculation of summary measures Other independent variables Outcome variables and outcome visits Univariate comparisons Proportionality of hazard xiii

14 xiv Time-dependent covariate models Incremental cut-point analysis Results Discussion Conclusions Table Patient characteristics Table Frequency and values of lipid measurements (1,492 measurements) Table Correlation matrix for LDL-C Table Correlation matrix for HDL-C Table Correlation matrix for TC:HDL-C Table Correlation matrix for TG Table Univariate comparisons of patients with and without coronary events used in lipid models Table Proportional hazards regression models for coronary events using various measures of low-density lipoprotein cholesterol Table Proportional hazards regression models for coronary events using various measures of high-density lipoprotein cholesterol Table Proportional hazards regression models for coronary events using various measures of total cholesterol to high-density lipoprotein cholesterol ratio Table Proportional hazards regression models for coronary events using various measures of triglycerides Table Time-dependent proportional hazards regression models for coronary events including only lipid measures and statistically significant covariates Table Cut-points for various measures of LDL-C determined using timedependent multivariate proportional hazards regression modeling for CAD events Table Test properties for the prediction of coronary event at a particular visit, based on LDL-C cut-point at (or up to) the prior visit Table Cut-point analysis for TC:HDL-C ratio using time-dependent covariate proportional hazards regression models xiv

15 Table Test properties for the prediction of coronary event at a particular visit, based on TC:HDL-C cut-point at (or up to) the prior visit Table Cut-points for various measures of TG determined using time-dependent multivariate proportional hazards regression modeling for CAD events Table Test properties for the prediction of coronary event at a particular visit, based on TG cut-point at (or up to) the prior visit Chapter Variability Over Time and Correlates of High Sensitivity C-reactive Protein in SLE Abstract Rationale and objective Methods Study population xv Inception cohort Prevalent cohort Disease activity measures and other variables Laboratory assessment Statistical analysis Results Variability of hscrp in the inception cohort Variability of hscrp in the prevalent cohort Effect of infection and disease activity on variability of hscrp Correlates of hscrp Discussion Conclusions Table Characteristics of patients in the inception and prevalent cohorts xv

16 xvi Table Risk quartile (Q 1-4 ) distribution of first hs-crp measurement in each of the inception and prevalent cohorts and changes over time Table Univariate correlation between of hscrp and age, disease duration and SLEDAI-2K score Table Univariate analyses Table Multivariate linear regression of variables independently associated with log 10 hscrp Chapter HsCRP as an Independent Risk Factor for CAD in SLE Abstract Rationale and objective Methods Patients HsCRP measurements Calculation of summary measures of hscrp Framingham risk score and measurement of disease activity Other independent variables Outcome variables and outcome visits Statistical analysis Time-dependent covariate regression models Results Discussion Conclusions Acknowledgements Table Baseline characteristics of study participants Table Univariate comparison of characteristics of patients with and without clinical CAD xvi

17 xvii Table Time-dependent proportional hazards regression analysis for CAD outcomes with log 10 transformed measures of hscrp Table Time-dependent proportional hazards regression analysis for CAD outcomes with hscrp quartiles and tertiles Table Time-dependent proportional hazards regression analysis for CAD outcomes with hscrp defined as normal or abnormal based on incremental quartile cut-points Table Time-dependent proportional hazards regression analysis for CAD events with hscrp defined dichotomously as normal or abnormal based on incremental tertile cut-points Chapter Overall Summary and Discussion Summary Discussion Originality and importance Future work Bibliography xvii

18 xviii List of Tables Table Summary of risk factors associated with the development of clinical coronary artery disease in selected studies...42 Table NCEP-ATP III classification of total cholesterol in relation to CAD risk...44 Table NCEP-ATP III recommended goals for LDL-C...45 Table Characteristics of patients in the University of Toronto Lupus Database as of August Table Data collection characteristics for blood pressure, total cholesterol, lowdensity lipoprotein cholesterol and high sensitivity C-reactive protein...62 Table Characteristics of patients (n=1260)...74 Table Number, frequency and values of total cholesterol, systolic blood pressure and diastolic blood pressure measurements...76 Table Proportion of patients with normal and elevated total cholesterol, systolic blood pressure and diastolic blood pressure at baseline and during follow-up...77 Table Total, between and within patient variance in total cholesterol, systolic blood pressure and diastolic blood pressure during follow-up...78 Table Independent correlates of total cholesterol...79 Table Independent correlates of total cholesterol in women only...80 Table Independent correlates of systolic blood pressure...82 Table Independent correlates of systolic blood pressure in women only...83 Table Independent correlates of diastolic blood pressure...85 Table Independent correlates of diastolic blood pressure in women only...86 Table Patient characteristics Table Frequency and values of total cholesterol measurements Table Frequency and values of systolic and diastolic blood pressure measurements Table Correlation matrix for total cholesterol Table Correlation matrix for systolic blood pressure xviii

19 xix Table Correlation matrix for diastolic blood pressure Table Univariate comparison of patients with and without coronary events used in TC models Table Proportional hazards models for coronary events using various measures of cholesterol Table Proportional hazards models for coronary events using various measures of cholesterol, including only significant covariates and interaction terms Table Proportional hazards models for coronary events using various measures of cholesterol defined dichotomously Table Proportional hazards models for coronary events using various measures of systolic blood pressure Table Proportional hazards models for coronary events using various measures of systolic blood pressure, including only significant covariates Table Proportional hazards models for coronary events using various measures of diastolic blood pressure Table Proportional hazards models for coronary events using various measures of diastolic blood pressure, including only significant covariates Table Proportional hazards models for coronary events using various measures of blood pressure defined dichotomously Table Patient characteristics Table Frequency and values of lipid measurements (1,492 measurements) Table Correlation matrix for LDL-C Table Correlation matrix for HDL-C Table Correlation matrix for TC:HDL-C Table Correlation matrix for TG Table Univariate comparisons of patients with and without coronary events used in lipid models Table Proportional hazards regression models for coronary events using various measures of low-density lipoprotein cholesterol Table Proportional hazards regression models for coronary events using various measures of high-density lipoprotein cholesterol xix

20 Table Proportional hazards regression models for coronary events using various measures of total cholesterol to high-density lipoprotein cholesterol ratio Table Proportional hazards regression models for coronary events using various measures of triglycerides Table Time-dependent proportional hazards regression models for coronary events including only lipid measures and statistically significant covariates Table Cut-points for various measures of LDL-C determined using timedependent multivariate proportional hazards regression modeling for CAD events Table Test properties for the prediction of coronary event at a particular visit, based on LDL-C cut-point at (or up to) the prior visit Table Cut-point analysis for TC:HDL-C ratio using time-dependent covariate proportional hazards regression models Table Test properties for the prediction of coronary event at a particular visit, based on TC:HDL-C cut-point at (or up to) the prior visit Table Cut-points for various measures of TG determined using time-dependent multivariate proportional hazards regression modeling for CAD events Table Test properties for the prediction of coronary event at a particular visit, based on TG cut-point at (or up to) the prior visit Table Characteristics of patients in the inception and prevalent cohorts Table Risk quartile (Q 1-4 ) distribution of first hs-crp measurement in each of the inception and prevalent cohorts and changes over time Table Univariate correlation between of hscrp and age, disease duration and SLEDAI-2K score Table Univariate analyses Table Multivariate linear regression of variables independently associated with log 10 hscrp Table Baseline characteristics of study participants Table Univariate comparison of characteristics of patients with and without clinical CAD Table Time-dependent proportional hazards regression analysis for CAD outcomes with log 10 transformed measures of hscrp xx xx

21 xxi Table Time-dependent proportional hazards regression analysis for CAD outcomes with hscrp quartiles and tertiles Table Time-dependent proportional hazards regression analysis for CAD outcomes with hscrp defined as normal or abnormal based on incremental quartile cut-points Table Time-dependent proportional hazards regression analysis for CAD events with hscrp defined dichotomously as normal or abnormal based on incremental tertile cut-points xxi

22 xxii List of Figures Figure Model for estimating the ten-year risk of coronary artery disease using data from the Framingham Heart Study...43 Figure SLE Disease Activity Index 2000 (SLEDAI-2K)...46 Figure A depiction of the overall framework of the thesis...50 Figure Schematic representation of hypothetical examples of individual patient data used for calculation of summary measures Figure Diagrammatic representation of the calculation of a time-adjusted mean Figure Diagrammatic representation of visits used to calculate summary measures for each of TC, SBP and DBP 106 Figure Depiction of the method used to perform gap analysis Figure Absolute difference between true and evaluated mean cholesterol Figure Absolute difference between true and evaluated AM cholesterol Figure Percent difference between true and evaluated AUC cholesterol Figure Absolute difference between true and evaluated mean systolic blood pressure Figure Absolute difference between true and evaluated AM systolic blood pressure Figure Percent difference between true and evaluated AUC systolic blood pressure Figure Absolute difference between true and evaluated mean diastolic blood pressure Figure Absolute difference between true and evaluated AM diastolic blood pressure Figure Percent difference between true and evaluated AUC diastolic blood pressure xxii

23 xxiii List of Abbreviations AM, time-adjusted mean ApoB, apolipoprotein B BAU, brachial artery ulstrasound BMI, body mass index BP, blood pressure CAC, coronary artery calcium CAD, coronary artery disease CDC/AHA, Centres for Disease Control / American Heart Association CI, confidence interval; 95% CI, 95% confidence interval cimt, carotid intima media thickness CRP, c-reactive protein DBP, diastolic blood pressure EBCT, electron beam computed tomography FMD, flow-mediated vasodilation HDL-C, high-density lipoprotein cholesterol hscrp, high-sensitivity C-reactive protein ICAM-1, intercellular adhesion molecule-1 JNC, Joint National Committee LDL-C, low-density lipoprotein cholesterol Lp(a), lipoprotein (a) mg/dl, milligram per decilitre mmhg, millimeters of mercury mg/l, milligram per litre MI, myocardial infarction xxiii

24 xxiv mmol/l, milli-moles per litre MPS, myocardial perfusion scintigraphy NCEP-ATPIII, National Cholesterol Education Program Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults III PWV, pulse wave velocity Q x, quartiles RR, relative risk SBP, systolic blood pressure SCD, sudden cardiac death SLE, systemic lupus erythematosus SLEDAI-2K, Systemic Lupus Erythematosus Disease Activity Index 2000 SLICC/ACR DI, Systemic Lupus International Collaborating Clinics / American College of Rheumatology Damage Index SPECT, Single Photon Emission Computed Tomography TC, total cholesterol TC:HDL-C ratio, total cholesterol to high-density lipoprotein cholesterol ratio TG, triglyceride umol/l, micro-moles per litre VCAM-1, vascular adhesion molecule-1 VLDLC, very low-density lipoprotein cholesterol vs, versus xxiv

25 1 Chapter 1 Introduction Systemic Lupus Erythematosus (SLE) is a prototypic multisystem autoimmune disease with an estimated prevalence of 1 in 1000, nine to one female predominance, usual onset in childbearing years and a course characterized by exacerbations and remissions [1, 2]. The advent of corticosteroids in the 1950s dramatically improved the short-term prognosis of patients with SLE from frequently fatal, with less than 50% survival at 5 years, to 93% five-year and 85% ten-year survival [3]. Over the last three decades there has been a further reduction in mortality. Among a cohort of over twelve hundred patients followed prospectively in the University of Toronto Lupus Clinic, the standardized mortality ratio has declined from (95% CI: 9.13, 17.39) in the period to 3.46 (95% CI: 2.71, 4.40) in the period [4]. This reduction in mortality has been attributed to improved treatment with immunosuppressive drugs and antimalarials, more judicious use of corticosteroids, better recognition and management of complications such as infection and renal failure, as well as secular trends related to the calendar period. In the 1970s, Urowitz and colleagues were the first to observe that many patients who survive early complication such as organ failure and sepsis go on to develop premature coronary artery disease (CAD), now recognized to be a major cause of morbidity and mortality, late in the course of SLE [5-7]. This bimodal pattern of mortality was first suggested by case series and later confirmed in prospective cohort studies. Despite improved overall prognosis, the prevalence of CAD among patients with SLE has remained relatively unchanged in recent decades [4]. Since the first description of the association between CAD and SLE, research has been directed towards determining risk factors for coronary events in order to identify the subset of patients at greatest risk, and in order to devise strategies for the prevention of this serious adverse outcome. In this introductory chapter the evidence of association between CAD and SLE will be reviewed, followed by a detailed discussion of what is currently known about the role of 1

26 traditional and non-traditional risk factors for CAD in SLE. This chapter will also include a discussion of statistical methods that may be applied to assessment of cardiac risk factors in the setting of SLE. The introduction will conclude with a summary of background points that are of key relevance to the hypothesis and objectives of this thesis. 1.1 The association of SLE with premature coronary artery disease (CAD) Case series 2 The first cases suggesting an association between SLE and atherosclerotic CAD were reported in the 1960s [8]. In 1976, the bimodal pattern of mortality in SLE was described based on a case series of 11 deaths among 81 patients followed for 5 years. Six of the eleven deaths occurred within the first year after diagnosis in patients who had active disease; in four of these a septic episode contributed to the death. All five patients who died late in the course of disease had a myocardial infarct (MI) close to the time of death [7]. Subsequently a case series described premature CAD with onset of symptoms before the age of 35 years in six women with SLE [9]. A larger case series published in 1985 and a case-control autopsy study in 1987 further supported these observations [6, 10]. Of note, case series to date have shown that the premature CAD seen in SLE is anatomically and histologically indistinguishable from classic atherosclerotic lesions seen in the general population [10] Cohort studies The first large prospective lupus cohort was established in Toronto, Canada in In the earliest reports the prevalence of symptomatic CAD, namely angina, MI or sudden cardiac death (SCD) in this SLE cohort was 8.9% [5]. In a later report, among 665 patients followed in this clinic to 1993, there were 124 (18.6%) deaths. In 31 (25%) the cause of death was an acute vascular event including MI or SCD. These causes accounted for 17% of early deaths within five years compared with 30% of late deaths beyond five 2

27 years post diagnosis [3]. In addition, in the 40 autopsies performed, moderate to severe atherosclerosis of the major vessels was found in 21 (54%) regardless of the cause of death. In a recent report from the Toronto cohort, among 1087 patients followed from 1970 to 2004, 118 (10.9%) patients had at least one atherosclerotic vascular event either cardiac, cerebrovascular or peripheral vascular. The breakdown of coronary events was as follows: MI 2.2%, angina 7.2%, SCD 0.3%. Among 561 inception patients followed from diagnosis, the prevalence of atherosclerotic vascular events was 9.6% (coronary events: MI 2.5%, angina 8.4%, SCD 0.0%). The mean (standard deviation; SD) age at first atherosclerotic event for the cohort overall and for the inception patients was 51.1 (12.3) and 53.7 (12.4) years, respectively [11]. The analysis of changing patterns in mortality and disease outcomes for patients in the Toronto cohort referred to earlier, wherein patients were divided into 4 entry cohorts and followed through four 9-year calendar periods defined over the same intervals, showed that although the prevalence of CAD decreased with later cohorts for the corresponding calendar period, it increased with follow-up period, reaching 27.6% in the first cohort after 3 decades [4]. The prevalence of symptomatic CAD in more recently established lupus cohorts in Baltimore and Pittsburgh is 8.3% and 6.6% respectively, rates that are similar to the Toronto cohort (10%) [5, 12, 13]. Although there are some differences in the demographic composition of these large cohorts, the mean age at the time of the first coronary event is similar, averaging at 48 to 49 years. Likewise the mean lupus disease duration at the time of the first coronary event is similar in the three cohorts averaging around 7 to 10 years. Manzi et al. compared the age-specific incidence rates of MI and angina in women with SLE seen at the University of Pittsburgh Medical Centre from 1980 to 1993 with women in the Framingham offspring study [12]. Women with SLE in the 35-to 44- year age group were over 50 times more likely to have an MI than women of similar age in the Framingham offspring cohort (rate ratio 52.43, 95% CI: 21.6, 98.5). Overall women with SLE were 5 to 6 times more likely to have MI than their population peers. Two thirds of all first cardiac events after the diagnosis of lupus were in women under the age of 55 years. When the annual incidence of MI among patients attending the Toronto lupus clinic was compared with background population rates for the province of Ontario, the 3 3

28 rate of MI in the lupus cohort was 5 per 1000 persons compared with 1 per 1000 persons in the general population for [14]. Furthermore the mean age of MI in the lupus clinic was 49 years compared with the peak incidence in the general population of 65 to 74 years. Overall, the prevalence of symptomatic CAD in SLE is approximately 10%, while the annual incidence of myocardial infarction is estimated to be 0.5% to 1.5% [14] Community based studies 4 A community based study by Ward using the California Discharge Database found that women with SLE aged between 18 and 44 years were more likely than age-matched controls to have been admitted to hospital with MI (odds ratio 2.27, 95% CI: 1.08, 3.46) [15]. In this study it was estimated that MI is overall 8.5 times more likely in women with SLE than in the general population. In a case-control study from the UK General Practice Research Database, Fischer et al. found that in patients presenting with their first MI, the adjusted odds ratio (adjusted for classic risk factors, previous angina and kidney disease) for a diagnosis of SLE was 2.67 (95% CI: 1.34, 5.34) [16]. In cases aged less than 70 years, the adjusted odds ratio was 3.47 (95% CI: 1.47, 8.06). Collectively, studies to date confirm that women with SLE have a significantly increased relative risk of MI compared with women in the general population and that the onset of CAD is premature with loss of pre-menopausal protection The link between coronary and non-coronary atherosclerosis in SLE The presence of vascular disease in one territory significantly increases the likelihood of disease in other vascular distributions [17-19]. In the 2002 National Cholesterol Education Program report, the presence of non-coronary atherosclerotic vascular disease was considered to carry the same risk for future cardiac events as the presence of CAD, thereby classifying it as a CAD risk equivalent [20]. 4

29 Manzi et al. performed carotid ultrasound in 175 women with SLE and found carotid plaque in 70 (40%) of these patients [21]. Twenty-six (15%) of the women in the study had a previous atherosclerotic vascular event (10 coronary, 11 cerbrovascular, 5 both). Independent determinants of carotid plaque included a number of traditional risk factors, some demographic and disease-related variables as well as a previous coronary event. The latter was also independently associated with severity of plaque. Likewise, Roman et al. compared the prevalence and correlates of carotid plaque in 197 SLE patients with 197 sex and age-matched controls [22]. Carotid plaque was more prevalent in SLE patients than controls (37.1% versus 15.2%, p<0.001). SLE patients who had carotid plaque were more likely to have pre-existing clinical CAD. Due to the small number of cardiac events, in some risk factor studies, all atherosclerotic vascular events coronary, cerebral and peripheral have been combined to maximize statistical power. However, caution is needed with this approach as there may be subtle but important differences in risk factors and the strength of association between these risk factors and specific vascular outcomes. 1.2 Cardiac risk factors in SLE Longitudinal cohort studies of cardiac risk factors in SLE 5 Due to the relatively low prevalence of SLE, most coronary risk factor studies in this disease are limited by small numbers of affected patients. In addition, differences in the racial and ethnic composition of SLE cohorts (57% blacks in the Baltimore cohort versus 7% blacks in the Toronto cohort) may also confound the findings of these studies. However, despite these limitations, several longitudinal cohort studies have shown that traditional risk factors, in particular hypercholesterolemia and hypertension are independently associated with CAD in SLE [5, 12, 13]. Three large studies of cohorts in Baltimore, Pittsburgh and Toronto have evaluated the risk factors for CAD in patients with SLE. In the prospective study of 229 patients from the Baltimore cohort, one or more coronary event(s) occurred in 19 (8.3%) patients over four years of follow-up [13]. In multiple regression analysis, compared with patients who 5

30 did not have CAD, patients who did were more likely to be older at diagnosis of SLE (31.7 versus 28.9 years, p=0.004), have a longer mean duration of SLE (12.3 versus 7.2 years, p<0.0001), have a longer mean duration of prednisone use (14.3 versus 7.2 years, p<0.0001), have a higher mean cholesterol (271.2 versus mg/dl, i.e versus 5.58 mmol/l, p<0.0001) and have a history of hypertension (odds ratio 3.5, 95% CI: 1.3, 9.6) or antihypertensive use (odds ratio 5.5, 95% CI: 1.8, 17.2). Older age at clinic entry and obesity were also significantly associated with CAD. In this study there were no significant associations with other known cardiac risk factors such as smoking, diabetes, family history of coronary disease, race or sex. In the Pittsburgh cohort of 498 women, 33 patients experienced cardiac events (11 MI, 10 angina, 12 both) [12]. In multiple regression analysis older age at lupus diagnosis (39.0 versus 34.0 years, p=0.02), longer duration of lupus (13.0 versus 10.0 years, p=0.01), longer duration of corticosteroid use (11.0 versus 7.0 years, p = 0.002), hypercholesterolemia (18.0% versus 4.0%, p=0.003) and postmenopausal status (48.0% versus 29.0%, p=0.03) were more common in SLE patients who had a coronary event than in those who did not. In a Cox proportional hazards model controlling for age, lupus disease duration (rate ratio 0.83, 95% CI: 0.74 to 0.94), hypercholesterolemia (rate ratio 3.35, 95% CI: 1.34, 8.36) and older age at lupus diagnosis (rate ratio 1.21, 95% CI: 1.09, 1.35) were significantly associated with cardiovascular events. In the first report of coronary events and cardiac risk factors from the Toronto Lupus Cohort, significant risk factors for coronary events were older age at SLE diagnosis, hypertension, hypercholesterolemia, hypertriglyceridemia and diabetes mellitus [5]. Other lupus related factors significantly associated with coronary events were previous pericarditis, myocarditis and congestive heart failure. When atherosclerotic vascular events in the Toronto cohort were reviewed in 2004, among 1087 patients, 118 experienced one or more coronary events (first events: MI 24, angina 76, SCD 3) over a mean follow-up of 8.4 years [11]. When the patients who had events were matched for sex, date of first clinic visit, age at first clinic visit and duration of follow-up, with 118 who had SLE and no events, hypertension (68.6% versus 44.1%, p=0.0002), smoking (55.1% versus 41.5%, p=0.02) and elevated cholesterol (89.6% verus 74.5%, p=0.008) were significant risk factors for atherosclerotic vascular events. In multiple regression 6 6

31 analysis, presence of more than one traditional cardiac risk factor (relative risk 1.76, 95% CI: 1.26, 2.47, p=0.0001) as well as the disease-related variables vasculitis (relative risk 2.26, 95% CI: 1.22, 4.17, p=0.009) and neuropsychiatric involvement (relative risk 2.19, 95% CI: 1.05, 4.59, p=0.004) were significantly associated with atherosclerotic vascular events. In summary, some risk factors are consistent across the cohorts. Specifically, older age at diagnosis of SLE and hypercholesterolemia are shown to be significant risk factors for CAD in all three of Toronto, Baltimore and Pittsburgh cohorts, while longer disease duration and hypertension are significant risk factors in two of these cohorts. Risk factors associated with the development of clinical CAD in these three cohorts are summarized in Table Traditional cardiac risk factors in the general population 7 The population-based Framingham Heart Study has identified several traditional risk factors that together enable estimation of an apparently healthy asymptomatic individual s 10-year absolute risk of coronary events (angina, MI and coronary death) [23-26]. The Framingham risk prediction algorithm for a patient without diabetes mellitus, chronic renal disease or clinically evident CAD takes into account age, gender, systolic and diastolic blood pressure (SBP and DBP; average of two office measurements [mmhg]), total cholesterol (TC) and high density lipoprotein cholesterol - HDL-C (one measurement [mmol/l]) and smoking (current or within past year) [23]. An important characteristic of this model is the recognition that many of the risk factors (e.g. age, hypertension and lipids) produce a graded increase in risk. Figure 1.1 shows the Framingham scoring system for calculating a person s risk of CAD within a ten-year timeframe. For example, using this algorithm, a 50 year old woman who has a total cholesterol level of 6.47 mmol/l, an HDL-C level of 1.55 mmol/l and a systolic blood pressure of 160 mmhg and who is a non-smoker would have a total risk score of 11. Her ten-year risk for CAD would therefore be 11%; the average risk for an average person of her age in the Framingham study population is 8%. As a cautionary note, the 7

32 Framingham tables underestimate coronary risk if the low-density lipoprotein cholesterol (LDL-C) level is > 6.0 mmol/l [23]. A validation study has found that the Framingham risk model performs well for prediction of coronary events in white and black men and women [27]. However, this model is not without limitations. The model does not consider lifetime risk, which might be substantially higher than ten-year risk and amenable to aggressive risk factor reduction [28-30]. While the woman in the above example has a ten-year risk of 11%, her lifetime risk is 50% [28]. In addition, the potential modifying role of other risk factors must also influence clinical judgment because no accepted algorithm exists for adjusting the Framingham risk estimate for other variables. Also, importantly, as the Framingham model was derived from data in the general population, it was not developed to be applicable to patients with extreme or unusual risk factors. Cardiac risk factors identified in other studies include elevated LDL-C, obesity and increased body mass index [BMI; weight (kg) / height 2 (m)], family history of CAD, increased total-to-hdl cholesterol ratio (TC:HDL-C) and the metabolic syndrome, which is a constellation of abdominal obesity, hypertension, diabetes, and dyslipidemia [31-39]. Among these, LDL-C and family history of CAD are also usually referred to as traditional risk factors. There are, in addition, a number of emerging novel cardiac risk factors in the general population which include several lipid abnormalities such as hypertriglyceridemia, increased Lp(a), increased non-hdl-cholesterol, increased apolipoprotein B (ApoB) and small dense LDL particles [40-45]. Markers of inflammation such as high-sensitivity C-reactive protein are also among these novel risk factors [46] Sex and age Cardiovascular risk factors promote CAD in either sex at all ages but with different strengths. In other words, in the general population, sex and age may act as effect modifiers for other risk factors. For example, the incidence of MI is increased six-fold in women and three-fold in men who smoke at least 20 cigarettes per day compared to subjects who have never smoked [47, 48]. The Framingham study found that the relative 8 8

33 importance of systolic blood pressure (SBP), diastolic blood pressure (DBP), and pulse pressure (the difference between the systolic and diastolic blood pressures) in CAD changes with age [49]. In patients <50 years of age, DBP was the strongest predictor; in those 50 to 59 years of age, all three blood pressure indices were comparable predictors while in those 60 years of age, pulse pressure was the strongest predictor. Dyslipidemia has a diminished impact with advancing age, but the lower relative risk is offset by the high absolute risk in older patients [50, 51] Family history Family history is a significant independent risk factor for CAD, particularly among younger individuals with a family history of premature disease (below age 55 years). The Physician's Health Study of 22,071 men followed for 13 years and the Women's Health Study of 39,876 women followed for 6.2 years have shown that compared to no parental history of an MI, a maternal history, a paternal history, and both maternal and paternal history was associated with a relative risk of cardiovascular disease of 1.71, 1.40, and 1.85 in men and 1.46, 1.15, and 2.05 in women [39]. However, an analysis of reliability of self-reported family history of CAD or cardiac risk factors among participants in the Framingham Offspring Study revealed that the positive predictive value of an affirmative statement regarding family history of cardiac death was only 66% for fathers and 47% for mothers, while the predictive value of a negative statement was much better at 90% [52]. These findings suggest that there is some value in obtaining family history information, but that self-reported information might not be accurate. They also suggest that the additional contribution of family history to CAD risk estimation after inclusion of other traditional risk factors is relatively modest Diabetes mellitus There is ample evidence that diabetes mellitus is associated with a substantially increased risk of coronary events. Diabetes mellitus is considered a CAD equivalent and automatically elevated to the highest risk category. Therefore the Framingham risk model is not applicable to patients with diabetes. In an analysis of over 13,000 participants in the Copenhagen Heart Study, the relative risk of MI or stroke was increased two to three 9 9

34 fold in those with type 2 diabetes, and the risk of death was increased two fold, independently of other coronary risk factors [53]. Guidelines published by the National Cholesterol Education Program and the sixth Joint National Committee on recognition and management of hypertension have provided a framework to treat coronary risk factors aggressively in diabetics [20, 54]. There is compelling evidence of the value of aggressive therapy of serum cholesterol (goal LDL-C <100 mg/dl [2.6 mmol/l]) and hypertension in patients with diabetes (goal SBP less than 130 mmhg) [20, 55, 56]. There is even some evidence that more aggressive target LDL-C goals of 70 to 80 mg/dl (1.8 to 2.1 mmol/l) may be appropriate [57, 58]. It has been asserted that approximately one-half of all patients suffering a coronary event have no established risk factors other than age and gender, a claim that has contributed to efforts to identify other markers of cardiovascular risk [59]. However, two retrospective analyses have cast doubt over the accuracy of this assertion, at least in the general population. A report based on data from three observational studies - the Framingham Heart Study, the Multiple Risk Factor Intervention Trial (MRFIT), and the Chicago Heart Association Detection Project in Industry - included more than 380,000 subjects, 21,000 of whom died of coronary events. Major cardiac risk factors were defined as TC 240 mg/dl ( 6.22 mmol/l), SBP 140 mmhg, DBP 90 mmhg, smoking, and diabetes. Study subjects were stratified by age and gender. Among subjects dying of coronary events, exposure to at least one risk factor ranged from 87 percent (for men aged 40 to 59 in the MRFIT trial) to 100 percent (for women aged 18 to 39 in the Framingham Heart Study)[60]. Another report based on 14 randomized clinical trials of CAD, included more than 120,000 subjects with ST elevation MI, non-st elevation acute coronary syndrome, or percutaneous coronary intervention. Risk factors were defined by information collected at the time of study enrollment, including smoking, diabetes, hypertension, and hyperlipidemia. At least one of these four risk factors was present in 85 percent of women and 81 percent of men. When stratified by age, the lowest prevalence of at least one risk factor was seen among subjects >75 years old (77 percent of women and 65 percent of men) [61]

35 Hyperlipidemia Total cholesterol (TC) Epidemiologic data have documented a continuous, graded relationship between the serum TC concentration and coronary risk [62, 63]. The causal role of TC in this relationship is suggested by clinical trials, which have demonstrated that targeted lowering of TC in patients with hypercholesterolemia reduces CAD morbidity. A metaanalysis of 38 primary and secondary prevention trials, for example, found that for every 10 percent reduction in serum TC, CAD mortality would be reduced by 15 percent and total mortality risk by 11 percent [64]. The National Cholesterol Education Program Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (NCEP-ATP) has designated a serum TC level <5.17 mmol/l optimal in the general population (Table 1.3)[65]. The serum cholesterol concentration in the blood is comprised of several major fractions, which are categorized according to their relative density. Serum TC has been used to screen adults at risk for CAD, but more recent data emphasize the advantages in knowing the concentrations of lipid fractions, such as low density lipoprotein cholesterol (LDL-C) and high density lipoprotein cholesterol (HDL-C) Lipoprotein measurement A standard serum lipid profile consists of TC, triglycerides (TG), and HDL-C. Once the TC, TG, and HDL-C values are known, the LDL-C concentration can be estimated from the Friedewald formula, where LDL-C is equal to TC minus very low density lipoprotein cholesterol (VLDL-C) and HDL-C [66]. Lipoprotein analysis should be performed after 9 to 12 hours of fasting to minimize the influence of postprandial hyperlipidemia. Serum TC and HDL-C can be measured in fasting or nonfasting individuals; there are only small clinically insignificant differences in these values when measured in the fasting or nonfasting state [67]. TC can vary by 4 to 11 percent within an individual due to multiple factors including stress, minor illness, and posture [68]. Values may also vary 11

36 between different laboratories, with data suggesting that a single measurement of serum TC can vary as much as 14 percent [68]. Thus ideally more than one measurement of TC should be obtained when treatment considerations demand a precise determination. Measurement of serum HDL-C and TG may demonstrate even greater variability over time [69]. Despite some variability of serum lipids over time, The Bogalusa Heart Study has shown that in the general population, risk factors for CAD often begin in childhood, persist over time and correlate with surrogate markers of CAD such as carotid intimamedia thickness (IMT) in young adults [70, 71]. This phenomenon of persistence of a cardiovascular risk factor over time is known as tracking [71]. Serum TC and LDL-C track at a high order, while BP tracks at lower order [70, 71] Low-density lipoprotein cholesterol (LDL-C) 12 High concentrations of LDL-C are a particularly important risk factor for atherosclerosis [31, 32]. The oxidative modification of LDL appears to be one mechanism by which LDL promotes atherosclerosis; oxidized LDL in turn may lead to atherogenesis via a number of mechanisms [72, 73]. Higher LDL-C concentrations have been associated with an increased incidence of CAD in a large number of studies [38]. Elevated plasma concentrations of oxidized LDL-C are also associated with CAD [74]. The robust relationship between total cholesterol (TC) and cardiovascular disease found in earlier epidemiologic studies is likely a reflection of the fact that most of TC is contained in LDL-C [20]. LDL-C levels serve as the primary target of lipid lowering treatment strategies. The Third Report of the Expert Panel on Detection, Evaluation and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III, or ATP III) classified the risk associated with various LDL cholesterol concentrations as shown in Table 1.3 [20]. NCEP ATPIII screening guidelines for lipid abnormalities published in 2001 recommended that screening be performed at least once every five years for all persons age 20 and over [65]. The desirable LDL-C level for individuals without known CAD who have 0 to one risk factor is <4.1 mmol/l while a desirable serum LDL cholesterol concentration for individuals without known CHD who have two or more risk factors (or 12

37 a ten-year Framingham risk score 20%) is <3.4 mmol/l. Patients with borderline-high cholesterol and less than two risk factors should be re-screened within one to two years. In an update to the NCEP III guidelines released in 2004, based on new clinical trial data, the proposed desirable LDL-C level for those in the highest risk category, including those with CAD or CAD equivalent or Framingham risk score >20%, once <2.6 mmol/l has now been reduced to 2.0 mmol/l [57]. CAD equivalents include diabetes mellitus, symptomatic carotid artery disease, peripheral arterial disease, abdominal aortic aneurysm and multiple risk factors that confer a 10-year Framingham risk of CAD >20% (Risk does not need to be assessed in people without CAD who have 0 to 1 risk factors since individuals in this category have a 10-year risk of CAD that is <10%) [20]. In addition to the conditions identified by ATP III as CAD equivalents, chronic renal insufficiency (defined by a plasma creatinine concentration that exceeds 1.5 mg/dl [133 µmol/l] or an estimated glomerular filtration rate that is less than 60 ml/min per 1.73 m 2 ) is considered to be a CAD equivalent [20]. Major CAD risk factors other than LDL-C are cigarette smoking, hypertension (BP 140/90), low HDL-C (<40 mg/dl [1.03 mmol/l]), family history of premature CAD (in male first degree relatives <55 years, in female first degree relative <65 years) and age (men 45 years, women 55 years). HDL-C 60 mg/dl (1.55 mmol/l) counts as a "negative" (protective) risk factor; its presence removes one risk factor from the total count [20] High density lipoprotein cholesterol (HDL-C) A low level of HDL-C is another important risk factor for atherosclerosis [75]. Comparisons of lipid profile to coronary angiography have shown an association between low HDL-C values and anatomic coronary atherosclerosis [76]. Low serum HDL-C is also associated with increased risk of symptomatic CAD [77]. Based upon data from the Framingham Heart Study, the risk for MI increases by about 25% for every 5 mg/dl (0.13 mmol/l) decrement in HDL-C below median values for men and women [75]. In 13 13

38 general, HDL-C below 1.0 mmol/l is considered a cardiac risk factor, while in contrast, a high serum HDL-C (above 60 mg/dl or 1.6 mmol/l) is cardioprotective [75]. HDL-C are highly heterogeneous in terms of both structure and function. HDL-C subclasses may differ in anti-atherogenic properties [78]. While some HDL-C particles are cardio-protective, pro-inflammatory HDL-C may in fact be associated with increased risk of CAD [79]. Measurement of total HDL-C does not give a sense of the proportion comprised of pro-atherogenic and anti-atherogenic subfractions Total cholesterol to HDL-C ratio (TC:HDL-C) Data from the Lipid Research Clinics and the Framingham Heart Study suggest that the TC (or LDL-C) to HDL-C ratio may have greater predictive value for CAD than serum TC or LDL-C alone [37]. Among men, a ratio of 6.4 or more identified a group at 2 to 14 percent greater risk than predicted from serum TC or LDL-C alone. Among women, a ratio of 5.6 or more identified a group at 25 to 45 percent greater risk than predicted from serum TC or LDL-C alone. Similarly, a prospective analysis from the Women's Health Study found that the TC:HDL-C ratio was highly predictive of cardiovascular events [80]. Comparing women in the highest and lowest quintiles for the ratio, the hazard ratio for an event was 3.8 (95% CI ). This was superior to the prognostic value of other lipid parameters and ratios including those using apolipoprotein measurements. Risk tables have been constructed that use both the ratio and presence or absence of other risk factors to predict coronary risk and possible indications for therapy [81]. TC:HDL-C ratio is subsumed in the 10-year Framingham risk assessment, which includes both TC and HDL-C level. Broadly, it is advocated that in the general population, target TC:HDLC ratio should be 5 or less and that in high-risk patients, such as those with multiple risk factors or CAD-equivalents including diabetes, the target should be 4.5 or less [82]. However, ATP III does not define the TC:HDL-C ratio as a specified lipid target of therapy due to concerns that treatment of ratios will divert priority from specific lipoprotein fractions as targets of therapy [20]

39 Triglycerides (TG) Hypertriglyceridemia is associated with an increased risk for CAD. As an example, in a meta-analysis of prospective population-based studies that evaluated the association between the serum TG concentration and incident cardiovascular disease, pooled analysis of 46,413 men enrolled in 16 studies was associated with a univariate risk ratio (RR) for TG of 1.32 (95% CI ) [41]. The five studies of nearly 10,800 women were associated with a univariate RR of 1.76 (95% CI ). Hypertriglyceridemia is often associated with reduced levels of HDL-C [83, 84]. Although the risk ratios for hypertriglyceridemia in both men and women decreased when the HDL-C concentration was included in the analysis, the risk ratios remained significant. Although elevated serum TG is considered a risk factor for CAD by some, the close association of TG with other lipid factors such as low HDL-C has brought into question its independence as a risk factor. Among the general population, knowing a patient's triglyceride level does not appear to improve the ability to estimate CAD risk beyond the estimate provided by measurement of serum cholesterol subfractions alone [85]. NCEP-ATPIII guidelines classify serum triglycerides such that desirable level is less than 1.7 mmol/l, with borderline high level between 1.7 and 2.2 mmol/l, and high level 2.3 mmol/l and over [20] Hypertension In the general population, the risk of MI and stroke first becomes apparent as blood pressure (BP) rises above 110/75 mmhg, and at any BP is affected by the presence of other risk factors [86, 87]. Although these data do not prove a cause-and-effect relationship, projections have been made for the predicted decrease in mortality to be expected from the 5 to 6 mmhg average fall in DBP achieved in the multiple clinical trials performed over the past 30 years (Cook, 1995 #276; Psaty, 2001 #278; Whelton, 2002 #275). The estimated benefit is a 40 percent reduction in stroke mortality and a 25 percent reduction in coronary mortality [88]. 15

40 Based on an average of two or more properly-measured readings at each of two or more visits after an initial screen, The Joint National Committee (JNC) defines hypertension as systolic BP (SBP) 140 mmhg or diastolic BP (DBP) 90 mmhg and prehypertension as SBP mmhg or DBP mmhg [86]. Thus the JNC recognizes that the correlation between the risk of adverse outcomes and BP is a continuous variable in which there is an increased incidence of poor outcomes as BP rises, even within the normal range. The increase in cardiovascular risk has primarily been described in terms of systolic and diastolic hypertension [49, 89]. Sequential studies have shown that BP can drop by an average of 10 to 15 mmhg between clinic visits [90, 91]. Thus many patients considered to be hypertensive at initial visits to a clinic are in fact normotensive. Although BP at the time of risk assessment (current BP) is typically used in most prediction algorithms, this does not accurately reflect an individual's past BP experience. Two analyses demonstrate the importance of inclusion of past BP into risk prediction models, since the duration as well as the degree of hypertension are both risk factors. A report from the Framingham Heart Study found that recent and remote antecedent BP predicted cardiovascular risk incrementally over current BP [92]. This effect was seen in men and women, younger and older subjects, and lower and higher BP groups. In another study, 1604 men whose risk status was first assessed when they were aged 45 to 64 and free of clinically obvious cardiovascular disease were then reassessed 25 years later when they were aged 70 to 90 [93]. Most patients changed their risk status over this time period, moving forward or backward. When the current risk status of the 342 subjects who developed cardiovascular disease (MI, angina, coronary bypass surgery, angioplasty, or stroke) during the follow-up period was compared with that of the 279 who remained healthy, there were few differences. However, when the original risk status was used, those patients who remained healthy had had significantly lower BP (121/79 versus 134/83 mmhg) and plasma TC levels (211 versus 226 mg/dl [5.45 versus 5.84 mmol/l]) 25 years earlier. Overall, these studies show that the use of long-term average BP may be more accurate in predicting CAD risk than single-point or current reading

41 Goal blood pressure Based on clinical trials evidence, the goal of antihypertensive therapy in patients with uncomplicated combined systolic and diastolic hypertension is a BP below 140/90 mmhg; treatment goals are determined by the higher BP category [86]. SBP is the greater predictor of risk in patients over 50 to 60 years of age [49]. However, caution is needed not to inadvertently lower the DBP to below 65 mmhg to attain a goal SBP <140 mmhg, since this level of DBP has been associated with an increased risk of stroke [94, 95]. A number of clinical trials and guidelines suggest a goal BP of <130/80 mmhg in two clinical settings, diabetes mellitus and proteinuric chronic kidney disease [86, 96]. Based on findings in the HOPE (Heart Outcomes Prevention Evaluation), EUROPA (European Trial on Reduction of Cardiac Events with Perindopril in Stable Coronary Artery Disease), and CAMELOT (Comparison of Amlodipine vs Enalapril to Limit Occurrences of Thrombosis) trials, a goal BP <130/80 mmhg is also recommended in patients with known atherosclerotic cardiovascular disease [97-99] Prehypertension and borderline hypertension Long-term follow-up of patients destined to develop essential (primary) hypertension demonstrates that BP readings gradually increase over time. They may initially be normal, then prehypertensive (or high-normal), and then intermittently elevated; however, the readings may show considerable variability [86]. Multiple epidemiologic studies have demonstrated an increase in cardiovascular risk in patients with prehypertension and borderline hypertension. The Framingham Heart Study examined the risk of cardiovascular disease at 10-year follow-up among participants with high-normal BP at baseline examination [100]. After adjustment for cardiovascular risk factors and when compared to those with optimal BP, the hazard ratios for a cardiovascular event at 10 years for those with high-normal values (130 to 139/85 to 89 mmhg) were 2.5 and 1.6 for women and men, respectively. An increased hazard ratio was also observed in those with values 120 to 129/80 to 84 mmhg compared with participants with optimal BP (<120/<80 mmhg). 17

42 Patients with prehypertension appear to have a greater prevalence of traditional cardiovascular risk factors than those with normal BP. Based on data from the 1999 to 2000 National Health and Nutrition Examination Survey (NHANES), the presence of at least one adverse risk factor (above-normal TC level, overweight/obesity, and diabetes mellitus) was significantly more likely among prehypertensive than normotensive individuals (RR of 1.65) [101]. However, the increase in cardiovascular risk associated with prehypertension cannot be explained entirely by a higher prevalence of other cardiovascular factors. This was illustrated in a substudy of the Strong Heart Study, in which the risk of incident cardiovascular events at 12 years was higher in patients with prehypertension and diabetes than in those with either risk factor alone [102] White coat hypertension Many patients are anxious when visiting the physician, leading to an office BP that may be substantially higher than BP during normal daily activities. If office readings average >140/90 and out of office readings average <135/85, the diagnosis is white coat hypertension. Sequential studies have shown that, in patients diagnosed as being hypertensive on a first visit to a new physician, there is a mean 15 and 7 mmhg fall in the systolic and diastolic BP, respectively, by the third visit [86], with some patients not reaching a stable value until the sixth visit [91]. Thus, it has been recommended that a patient with mild to moderate elevation in BP should not be diagnosed with hypertension unless the BP remains elevated after three to six visits or there is evidence of ongoing end-organ damage Interaction between SBP and TC 18 There is an interaction between SBP and TC that was illustrated in a large study of 380,000 individuals from Asia, Australia, and New Zealand [103]. Each 10 mmhg increase in SBP was associated with a 21 to 34 percent increase in risk at all levels of serum TC and higher levels of TC increased risk at all levels of SBP. These increases in risk were more than additive but less than multiplicative. Adjustment for other risk 18

43 factors had no appreciable effect on these findings. Patients in the higher categories of both TC ( 240 mg/dl [6.25 mmol/l]) and SBP ( 160 mmhg) had a seven-fold increase in CAD and an eight-fold increase in stroke compared to patients in the lowest category of both TC (less than 183 mg/dl [4.75 ml/mmol/l]) and SBP (less than 130 mmhg) Role of traditional cardiac risk factors in SLE 19 In a retrospective cohort study, Rahman et al. compared the prevalence of traditional cardiac risk factors (hypertension, hypercholesterolemia, diabetes, smoking and family history) in 35 patients with SLE (27 women, 8 men) who had coronary artery events during the course of their illness, with 397 age- and sex-matched non-sle controls (83 women, 314 men) who had premature coronary artery disease (aged less than 60 years at time of first event and angiographically documented coronary artery disease) [104]. Women who had SLE had significantly fewer coronary risk factors per cardiac event than female controls (2.0 ± 0.77 versus 2.9 ± 1.19, p=0.0008). Bruce et al. compared the prevalence of traditional coronary risk factors and the ten-year Framingham risk of a coronary event in 250 female SLE patients and 250 age-matched controls [105]. They found that although hypertension and diabetes were significantly more common among SLE patients (Hypertension: 33% vs 13%, p=0.001, diabetes: 5% vs 1%, p=0.0066) and SLE patients had a higher mean number of coronary risk factors per person (1.01 ± 1.0 versus 0.72 ± 1.0, p=0.0014), the ten-year risk of a coronary event, using the Framingham multiple risk factor assessment formula was the same in SLE patients and controls (3.2%). However, in a case-control study, Haque et al. showed that patients with SLE who have clinical CAD have had more exposure to all traditional CAD risk factors than SLE patients without clinical CAD [106]. In a seminal paper, Esdaile and colleagues sought to determine to what extent the risk of cardiovascular disease in SLE could not be explained by common risk factors [107]. The participants at two SLE registries were assessed retrospectively for the presence of cerebral and cardiac vascular outcomes. For each patient the probability of the given 19

44 outcome was estimated based on the individual s risk profile and the Framingham multiple logistic regression model. After controlling for traditional risk factors at baseline, the increase in relative risk (observed compared with baseline) for all coronary events was 7.5 (95% CI 5.1 to 10.4) and the relative risk for stroke was 7.9 (95% CI 4.0 to 13.6). The investigators concluded that there is a substantial increase in risk of CAD and stroke in patients with SLE that cannot be fully explained by traditional Framingham risk factors alone. A number of other studies using surrogate markers of CAD such as carotid intima-media thickness (IMT) or brachial artery flow-mediated vasodilation (FMD) as the outcome of interest have confirmed the importance of both traditional and as yet undefined lupusassociated factors in SLE-related CAD [108]. Overall, although traditional risk factors may play a relatively small role in SLE-related CAD, they are potentially reversible. Some experts have advocated that SLE should be considered a CAD equivalent and thus elevated to the highest category of baseline risk for treatment of risk factors such as lipids (aiming for LDL-C < 2.6 mmol/l) and BP (aiming for < 130/80), much like diabetes mellitus [109, 110]. However, unlike diabetes, there is currently no evidence that such aggressive therapy reduces the incidence of coronary events in SLE Assessment of lipid risk factors in SLE 20 As discussed earlier, hypercholesterolemia has been shown in three large cohort studies to be a significant risk factor for CAD in SLE [5, 12, 13]. However, although these studies used a single-point-in-time measurement of TC in their analyses, the cut-point level of TC for definition of normal and abnormal has not been consistent across all studies. By and large, hypercholesterolemia has been defined as a level greater than the recommended target level for the general population i.e. > 5.2 mmol/l. In a study of 134 patients from the Toronto cohort who were followed from the time of diagnosis of SLE, patients with sustained hypercholesterolemia in the first 3 years of their disease were shown to be at greatest risk of cardiovascular events over 12 to14 years 20

45 follow-up compared with those who had persistently normal cholesterol or variable hypercholesterolemia in the first 3 years of disease [111]. In multiple logistic regression analysis the best predictors of sustained hypercholesterolemia were cumulative dose of steroid (OR 1.11 per gram, CI , p<0.001), lack of anti-malarial therapy (antimalarial OR 0.08, 95% CI , p<0.001) and age of onset of SLE > 35 years (OR 6.46, 95% CI , p=0.006). The ideal lipid marker of cardiovascular risk in SLE patients is not known. Studies to date have primarily addressed the role of elevated TC and found this to be significantly associated with cardiovascular events. The role of LDL-C as an independent risk factor for CAD in SLE has not been thoroughly evaluated. Studies to date have described lipid abnormalities and their correlates in lupus patients but have not evaluated the predictive role of these lipid markers for coronary events in SLE. In the Toronto risk factor study, there was no statistically significant difference in LDL-C level between lupus patients and matched controls, while in a paper by Manzi et al., LDL-C was an independent determinant of focal plaque in patients with SLE [21, 105]. Telles et al. also found that LDL-C >100mg/dL (i.e. > 2.58 mmol/l, p=0.044) was an independent predictor of carotid plaque in a group of Brazilian patients with SLE [112]. Ideal targets for lipids in SLE patients are not known. Some experts advocate including patients with SLE in the high risk category, along with patients who have diabetes mellitus or known CAD, which in light of the most recent recommendations of the American Heart Association would mean aiming for LDL-C less than 2.0 mmol/l and TC:HDL-C ratio less than 4.0. However, there is currently no evidence that aggressive lowering of lipids leads to a reduction in CAD risk among patients with SLE. In a 2-year double blind clinical trial of 200 patients with SLE, those randomised to receive atorvastatin 40 mg had slowing of the progression of carotid intima-media thickness compared with patients receiving placebo [113]. However, there was no effect on coronary calcium progression, a marker of atherosclerotic CAD. Furthermore, patients in the treatment arm had significantly more muscle and liver toxicity than those on placebo. Preliminary analysis of data from an ongoing randomised controlled trial of health improvement and cardiovascular disease prevention has shown a reduction in the 21 21

46 estimated 8-year risk of CAD in patients with SLE through patient education and management of traditional risk factors [114] Hypertension and CAD in SLE 22 As previously discussed, hypertension has been shown to be a significant risk factor for CAD in both the Toronto and Baltimore lupus cohorts [5, 13]. In a study of an inception cohort of 150 patients with SLE (75 hypertensive and 75 normotensive), 17 (22.7%) hypertensive patients had at least one atherosclerotic vascular event compared to 6 normotensive patients (p=0.022) [115]. The groups were comparable with respect to other risk factors, except for a higher frequency of hypercholesterolemia (58.7% vs. 33.3%, p=0.003), azotemia (24.0% vs. 5.3%, p=0.001) and steroid use (82.7% vs. 69.3%, p=0.038) among the hypertensive group. In multiple regression analysis, the best predictor of a vascular event was hypercholesterolemia, defined as two successive nonfasting TC levels above 5.2 mmol/l or treatment with a lipid lowering agent (odds ratio 6.9, 95% CI: 2.4, 24.8, p<0.001). The investigators concluded that systemic hypertension is associated with an increased frequency of vascular events in SLE and that this is best explained by its association with hypercholesterolemia. As discussed earlier, the association between hypertension and hypercholesterolemia is well described in the general population. Furthermore, in the general population there is an interaction between these two risk factors such that the combined risk associated with the presence of both risk factors is more than additive [103]. Unlike hypercholesterolemia, hypertension has been more consistently defined in lupus studies to date as systolic blood pressure (SBP) 140 or diastolic BP (DBP) 90 mmhg. In some studies use of an antihypertensive agent for the treatment of hypertension is also included in the definition. As discussed, sequential studies in the general population have shown that BP can drop by an average of 10 to 15 mmhg between clinic visits [90, 91]. Thus many patients considered to be hypertensive at initial visits to a clinic are in fact normotensive. Therefore, in studies to date of risk factors for CAD in SLE, defining hypertension based on one abnormal reading may have resulted in misclassification of patients. Furthermore, 22

47 the cut-off for normal and abnormal BP in patients with SLE is unclear. For patients at average risk the JNC recommends a BP goal of below 140/90 mmhg [86]. However, based on a number of clinical trials, JNC guidelines suggest a goal BP of <130/80 mmhg in high-risk patients with diabetes mellitus or chronic renal failure [86, 96]. In absence of evidence to the contrary, currently, the accepted definition of hypertension in patients with SLE is BP 140/90 mmhg, although some experts argue that patients with SLE, like patients with diabetes, should be considered high risk, with BP targets of <130/80 mmhg [109, 110]. 1.3 Methodological considerations in studies of coronary risk factors in SLE Time-to-event analysis using time-dependent covariates 23 Most studies to date of risk factors for CAD in SLE have used the Cox proportional hazards regression model to determine the independent contribution of each risk factor to coronary event outcomes. The Cox model is a semi-parametric method of survival (or time to event ) analysis wherein there are no underlying assumptions regarding the distribution of the hazard function or baseline hazard. In this model the hazard function h(t) is expressed as h(t) = h 0 (t) * exp(xβ), where Xβ = β 1 X 1 + β 2 X 2 + +β j X j ; X 1-j are covariates and β 1-j their respective β coefficients. h(t) is the hazard at time t and h 0 (t) is the baseline hazard, which remains unspecified. Alternatively, the Cox hazard function may be expressed as ln h(t) = h 0 (t) + β 1 X 1 + β 2 X 2 + +β j X j. The latter expression of the hazard function highlights one of the major assumptions of the Cox model, which is that the relationship between the hazard and continuous covariates is log-linear. The second major assumption of the Cox model is that the hazard ratio is constant over time. This is also referred to as the proportionality of hazard assumption and can be tested using log-log graphs of ln(-lns(t)) [natural log of the negative natural log of survival function] versus ln(t) [natural log of time] for a simple two-sample comparison case. For models with many covariates, an overall proportionality test statistic may be determined by including multiple interaction terms (covariate * ln(t)) in the model. A proportionality 23

48 test statistic p value >0.05 indicates absence of evidence to contradict the proportionality assumption and, provided other assumptions are met, indicates that a proportional hazards model may be safely applied to the dataset. Finally, an assumption that applies to all of survival analysis including Cox models is the issue of non-informative censoring. To satisfy this assumption, the design of the underlying study must ensure that the mechanisms giving rise to censoring of individual subjects are not related to the probability of an event occurring. A conventional survival analysis model such as the Cox model allows us to look at how long people are in one state (e.g. free of coronary event) followed by a discrete outcome (e.g. MI). It can handle situations in which people enter the study at different times and are followed for varying periods (e.g. follow-up of patients in the lupus clinic); it also allows us to compare two or more groups (e.g. high vs. normal TC). However, this type of analysis cannot handle situations where predictor variables change over time and are measured at several time points prior to outcome. In this setting a survival analysis method using time-dependent covariates is more appropriate. This model allows for all evaluations of independent variables through time to be used and relates these to outcome at each time point. However, as discussed earlier, the assumptions of Cox regression must first be met before this model can be applied to a dataset. Of note, although use of covariates in a time-dependent manner enables use of all data available, at each time point (t i ), the hazard of an event is related to the value of the covariate measured at the prior time point (t i-1 ), with no retention of memory for values recorded from the first visit to t i-1. To date, studies of cardiac risk factors in SLE have used Cox proportional hazards models with single point in time measurements of potential risk factors thus failing to capture cumulative exposure to risk factors over time Summarising dynamic cardiac risk factors over time 24 One approach to summarising dynamic risk factors such as lipids and BP measured over multiple visits would be to evaluate the arithmetic mean or average of all available observations. This approach is unaffected by the length of time between visits and therefore does not incorporate a sense of how long the variable remains at a given level. 24

49 Because a time-adjusted mean includes in its evaluation the time interval between visits, the contribution of X i (where X is the variable observed at the ith visit) to the overall mean is larger or smaller if the time interval between visits (i.e. measurements) is larger or smaller. The downside to the use of a time-adjusted mean is that where there are large gaps between visits, the contribution of the variable measured at the visit just prior to the gap interval is magnified. This may lead to bias in either direction, as the true status of the variable in the gap period is unknown. The advantage of both arithmetic and time-adjusted means is that they are measured in the same units as the original variable and are therefore potentially more clinically meaningful. Furthermore, arithmetic and time-adjusted means may rise and fall over time. On the other hand, area-under-the-curve continues to rise over time and because it is the product of a covariate and unit of time, it is measured in units that are less meaningful to clinicians. Although the arithmetic and adjusted means are closely correlated, time-adjusted means take into account the skewing effect of transient outlier observations such as brief elevations in lipids or BP that may occur during treatment with high dose corticosteroids Use of summary measures of cardiac risk factors in a time-dependent survival model 25 As with single-point measurements, summary measures such as mean and time-adjusted mean (AM) of covariates measured over multiple visits may be related to outcome in a time-dependent model. However, unlike the case of single-point-in-time measurements, use of summary measures means that at each time point the hazard of outcome is related to a variable that captures cumulative exposure to the potential risk factor from t 1 to t i-1. The use of measures that summarise cardiac risk exposure over time is novel and to date has not been evaluated in SLE. 1.4 Non-traditional risk factors for CAD in SLE As traditional risk factors only partly account for the increased risk of CAD in SLE, there has been much interest in identification of novel risk factors. These include 25

50 homocysteine, which in both the Physicians Health Study and the Framingham Heart Study has been shown to be associated with increased relative risk of coronary events and stroke [116, 117]. In one study, after adjustment for established cardiovascular risk factors such as hypercholesterolemia and hypertension, total plasma homocysteine concentration greater than 14.1 mmol/l was an independent risk factor for thromboembolic stroke (odds ratio 2.44 [ ], p=0.04) and arterial thromboses overall, including MI and limb ischemia (odds ratio 3.49 [ ], p=0.05) in SLE [118]. Similarly, Roman et al. have shown that higher homocysteine concentration is an independent predictor of progression of carotid plaque [119]. In the Toronto Risk Factor Study, elevated homocysteine level (>15 µmol/l) was seen more commonly in patients with SLE than in matched controls (11.6% vs 0.8%, p<0.01) [105]. In the general population and in patients with SLE, numerous other novel risk factors for CAD are currently under investigation. These include soluble intercellular adhesion molecule-1 (ICAM-1) and soluble vascular adhesion molecule-1 (VCAM-1), anti-endothelial cell antibodies and atherogenic lipoproteins such as apolipoprotein B [ ]. Although many such risk factors have been proposed, most have yet to be conclusively linked to coronary events in the general population or in patients with SLE. Among the nontraditional risk factors for CAD in SLE, C-reactive protein has received the greatest attention High sensitivity C-reactive protein (hscrp) 26 Inflammation is increasingly recognised as an important contributor to atherogenesis [126]. It is therefore perhaps not surprising that a chronic inflammatory disease such as SLE is associated with increased risk of premature CAD. Protein markers of inflammation have been studied as indicators of cardiovascular risk. The most extensively studied biomarker of inflammation in cardiovascular diseases is C-reactive protein (CRP), for which standardized high-sensitivity assays (hs-crp) are widely available [127, 128]. Data from dozens of epidemiologic studies have shown a significant association between elevated serum or plasma concentrations of CRP and the prevalence of underlying atherosclerosis, the risk of recurrent cardiovascular events among patients with established disease, and the incidence of first cardiovascular events among 26

51 individuals at risk for atherosclerosis [129]. In both the Physicians Heath Study and the Women s health study, baseline hscrp concentration was shown to be an independent risk factor for cardiovascular disease [130, 131]. Ridker et al. have shown that hscrp levels may be divided into quartiles based on relative risk of coronary events in the general population at 5 years [130, 132]. The quartiles (Q x ) in mg/dl and associated relative risks (RR) are Q 1 <0.08, RR 1.0; Q to <0.16, RR 1.2; Q to 0.35, RR 1.3; Q 4 >0.35, RR 1.7. More recently, in the JUPITER (Justification for the Use of Statins in Primary Prevention: An Intervention Trial Evaluating Rosuvastatin) study, it has been shown that treatment of elevated hscrp level with a statin prevents vascular events in men and women, independently of the lipid lowering effect (Ridker, 2008 #261). This landmark study by Ridker and colleagues suggests that hscrp, hereto a screening tool for assessment of coronary risk in the general population, may in fact be a treatable target for prevention of atherosclerotic events. However, it is possible that the cardioprotective effect of statins in this case was mediated through mechanisms other than reduction in hscrp such as endothelial effects, and the findings require confirmation in further studies. High sensitivity methods for measurement of CRP (hs-crp) detect concentrations down to 0.3 mg/l. These assays are necessary for cardiovascular risk stratification, which is based upon discrimination of CRP levels extending below 3 mg/l. Although the precision of the hs-crp assay is 0.3 mg/l, the CDC/AHA recommendation of two measurements to confirm a stable value is a reflection of the fact that systemic inflammatory status can fluctuate over time. In a study of 113 healthy adults, the variability in five serum CRP measurements over one year was similar to that for serum total cholesterol [133]. In contrast, a greater degree of variability was noted in a study in which serial measurements of serum CRP were obtained in 159 patients with stable ischemic heart disease [134]. Using categories of low, average, and high risk, defined on the basis of serum CRP <1, 1 to 3, and >3 mg/l, 40 percent of patients changed risk category between the first and second measurements. Similar fluctuations were noted in interleukin (IL)-6, another inflammatory marker. These observations raise the possibility that patients with cardiovascular disease have a dynamic systemic inflammatory status that may affect the acute coronary risk

52 In SLE, CRP has been shown to be associated with intercurrent infection and some clinical features particularly in the pulmonary and musculoskeletal systems [135, 136]. However, in contrast to the general population, most studies to date have not shown CRP to be correlated with prevalent vascular disease in SLE, detected using modalities such as carotid, aortic or femoral ultrasound or helical computed tomography used to detect coronary calcium [21, 137, 138]. Similarly, in a prospective cohort of 78 patients followed five years, CRP was not an independent predictor for development of preclinical atherosclerosis detected by ultrasound measurement of carotid IMT [139]. In a cross-sectional study of 214 women form the Pittsburgh SLE cohort who did not have clinical cardiovascular disease, women with carotid plaque had significantly higher median CRP levels than did women without plaque [21]. However, in regression analysis, CRP was not an independent determinant of plaque. Among 546 patients in the LUMINA (Lupus in Minority Populations: Nature vs Nurture) lupus cohort, over a median follow-up of 73.8 months, 34 (6.2%) developed one or more vascular events. Independent predictors of vascular events were older age, current smoking, longer follow-up, elevated CRP (defined as a level greater than the highest quintile for the distribution of the study patients values; >16.5 mg/l) and presence of any antiphospholipid antibody [140] SLE disease activity measures 28 The SLE Disease Activity Index (SLEDAI) and its revised form SLEDAI-2K were designed to evaluate disease activity at the time of each clinic visit [141, 142]. The SLEDAI-2K is a multifactorial index of 24 descriptors of disease activity in 9 organ systems (Figure 1.1). Some of the items are clinical features that are currently present or have been present within the last 10 days and others are laboratory measures taken at the time of visit. Each of the items in the SLEDAI-2K is clearly defined and inherent to all definitions is attribution to SLE itself. The maximum theoretical score in the SLEDAI-2K is 105 but in practice, very few patients have scores greater than 45. The SLEDAI has been validated and shown to be responsive to change over time [143, 144]. Based on a study correlating SLEDAI values with clinician assessments, flare has been defined as an increase in SLEDAI of 4 or more points and improvement as a reduction in SLEDAI of 4 28

53 or more points between visits at least 3 months apart [145]. Compared with the original SLEDAI, the modified SLEDAI-2K reflects persistent disease activity in some descriptors, namely proteinuria, rash, alopecia and mucous membrane lesions that had previously only applied to new or recurrent occurrences [142]. There are a number of other validated indices for measuring disease activity in SLE including the British Isles Lupus Assessment Group (BILAG) index [146]. However, to date most studies of cardiac risk factors in SLE have used the SLEDAI as the instrument for assessing disease activity. As discussed earlier, classic risk factors do not fully account for the increased risk of CAD in SLE. Given the chronic inflammatory nature of SLE and the important role of inflammation in atherosclerosis, the role of lupus disease activity as a predictor of coronary risk has been evaluated. As the course of SLE is characterised by periods of activity and quiescence, single-point-in-time measurement of disease activity does not adequately reflect cumulative activity over the course of illness. While SLEDAI-2K at presentation is not predictive of coronary events, when the time-adjusted mean SLEDAI (AMS) is used to summarise disease activity over time, for each unit increase in AMS, the hazard ratio for coronary events is seen to increase by 8% (hazard ratio 1.08, 95% CI 1.00, 1.16, p=0.046) [147, 148]. In the same model, male gender, older age at diagnosis, longer disease duration and prior exposure to immunosppressives a possible surrogate for disease activity- is also predictive of coronary events. Karp et al. have also shown that recent lupus disease activity measured using the SLEDAI-2K correlates with higher values of several well-recognised coronary risk factors and overall two-year coronary heart disease risk [149]. Specifically, a 6-point increase in average SLEDAI score in the past year was associated with a statistically significant increase of 0.13 mmol/l in triglycerides, 3.4 mmhg in systolic BP, 0.04 mmol/l in blood glucose, a decrease of 0.06 mmol/l in HDL cholesterol and a 5% increase in probability of CAD in the next two years even when adjusted for corticosteroid use and other covariates. The same investigators found very significant interactions between the average daily corticosteroid dose and the average SLEDAI-2K score for three outcomes: total cholesterol (p<0.0001), systolic BP (p<0.0001) and two

54 year coronary heart disease risk (0.0015). In each case the impact of increasing the average corticosteroid dose on increasing the level of a risk factor was stronger for patients with higher SLE disease activity in the past year. A lupus pattern of dyslipidemia occurring in times of disease activity has been described. High levels of VLDL-C and TG and low levels of HDL-C - the 'lupus pattern' - were observed in inactive SLE patients compared to controls (P < 0.05) [150]. Active disease enhanced this difference inducing a more striking increase in VLDL-C and TG levels and also a decrease in HDL-C and LDL-C levels compared to inactive SLE patients (P < 0.05), characterizing the 'active lupus pattern'. Moreover, a significant correlation was found between SLEDAI scores and all lipid fractions SLE disease and treatment related factors 30 Several SLE disease and treatment related factors have been shown to contribute to the risk of CAD, although the findings have not been consistent across all studies. In the Toronto cohort, along with classic cardiac risk factors such as smoking, vasculitis (relative risk 2.26, p=0.009) and neuropsychiatric manifestations (relative risk 2.19, p=0.004) were independently associated with atherosclerotic vascular events [11]. The investigators speculated that these two features may be surrogates for overall disease activity. In a study by Roman et al., independent predictors of carotid plaque, a surrogate marker for coronary atherosclerosis, in SLE patients were a longer duration of disease, a higher Systemic Lupus International Collaborating Clinics / American College of Rheumatology (SLICC/ACR) damage index score, the absence of anti-smith antibodies and a lower incidence of cyclophosphamide use [22]. The latter association suggested that less exposure to therapy might itself be a risk factor and that uncontrolled disease activity may be the hidden culprit. Two prospective cohort studies have shown that longer mean duration of prednisone use, a possible surrogate marker for disease activity over time, is a risk factor for coronary events in SLE [12, 13]. Corticosteroids are known to elevate levels of atherogenic lipids such as total, low density and very low density lipoprotein cholesterol as well as blood pressure and blood glucose [151]. However, as corticosteroids are used to treat active 30

55 disease, it is difficult to tease apart the effect of disease itself from treatment. Indeed by controlling disease activity and inflammation, steroids may also play a cardioprotective role. In the study by Karp et al., even after adjustment for SLE activity and other potential confounders, a 10 mg increase in the average daily prednisone-equivalent dose in the preceding year was associated with a statistically significant increase of 0.41 mmol/l in total serum cholesterol, 0.08 mmol/l in HDL cholesterol, 0.22 mmol/l in LDL cholesterol, 0.06 mmol/l in ApoB, 0.15 mmol/l in triglycerides, 1.6 mmhg in systolic BP, 0.4 kg/m 2 in BMI and 0.02 mmol/l in blood glucose level, as well as a 16% increase in the estimated 2-year CHD risk [149]. Although antimalarials have not been shown to be independently protective against CAD in SLE, they may have many advantages in regard to atherosclerotic risk. Several investigators have found that antimalarials significantly lower total and LDL cholesterol concentrations and elevate HDL cholesterol [ ]. In addition, when taken concomitantly with steroids, antimalarials can reduce serum cholesterol by up to 10% over 3 to 6 months [154]. Antimalarials also have anti-inflammatory and anticoagulant effects and lower fasting blood glucose concentration [155, 156]. A recent systematic review found high levels of evidence that antimalarials prevent lupus flares and increase the long-term survival of patients with SLE [157]. Combined with a good safety profile, the multitude of beneficial effects of antimalarials led the authors to conclude that antimalarials, in particular hydroxychloroquine, should be administered lifelong to most patients with SLE. MI and stroke are features of the anti-phospholipid syndrome, wherein anticardiolipin antibodies (ACA) and lupus anticoagulant (LAC) cause thrombosis. In the LUMINA cohort, the presence of antiphospholipid antibodies was an independent predictor of vascular events [140]. However, antiphospholipid antibodies (APLA) are heterogeneous and certain APLA are thought to be proatherogenic by promoting the uptake of oxidized LDL into macrophages [158]. Clinical studies have been divided as to whether APLA are associated with the development of atherosclerosis. Several studies have found no clear association of APLA/LAC with subclinical atherosclerosis, while others have found an 31 31

56 association between these antibodies and increased carotid IMT and presence of carotid plaque, both markers of cardiovascular disease [22, 138, 139, 159]. 1.5 Markers of subclinical CAD in SLE 32 As the evolution of atherosclerotic vascular disease is a continuum, starting with the earliest pathophysiologic changes in blood vessels and culminating in occlusive lesions that lead to coronary events, study of the prevalence and correlates of subclinical CAD could provide useful information for assessing risk and intervening to prevent the occurrence of actual coronary events [160]. The endothelium plays a central role in the pathogenesis of atherosclerosis [126]. Endothelial dysfunction is one of the earliest detectable functional abnormalities of the coronary circulation, occurring before the appearance of visible atherosclerotic lesions, and has been demonstrated in patients with cardiovascular risk factors [161]. It has been postulated that loss of normal endothelial function is a common mechanism by which cardiac risk factors lead to atherosclerosis [161]. We have reported that a significant proportion of SLE patients who have perfusion defects on scintigraphy do not have occlusive atherosclerotic plaques on coronary angiography, suggesting alternate mechanisms of ischemia such as endothelial dysfunction or coronary vasospasm [162]. We found that of 21 patients with perfusion defects on scintigraphy (10 of whom were symptomatic for coronary heart disease), 14 (67.0%) had no occlusive lesions on angiography. Sella et al. performed SPECT (Single Photon Emission Computed Tomography) 99m Tc-sestamibi myocardial scanning at rest and after dipyridamoleinduced stress in 90 patients with SLE without a history of coronary artery disease [163]. Thirty (33%) patients had perfusion defects. Of these, 21 agreed to undergo coronary angiography, wherein atherosclerotic lesions were identified in 8 of the 21 patients. Hypertension ever and postmenopausal status were significantly associated with abnormal angiographic findings. Conventional coronary angiography does not allow imaging of the coronary microvasculature where the pathophysiology may lie among patients with SLE. The prognosis of such cases of non-occlusive coronary disease is not known. However, it is possible that the course of such cases may not be benign as 32

57 evidenced by 13 patients in our study who had documented coronary syndromes without atheromatous lesions on coronary angiography [162]. It may also be postulated that if endothelial dysfunction is a precursor to atherosclerotic CAD, then non-occlusive CAD forms an intermediary step in the evolution to full-blown occlusive CAD [160]. High-resolution brachial artery ultrasound (BAU) is a widely used, noninvasive technique for measuring endothelial function and has been shown to correlate closely with coronary artery endothelial function measured using intracoronary acetylcholine challenge during angiography [ ]. Specifically, this technique measures endothelium-dependent nitric oxide mediated (flow-mediated vasodilation; FMD) and endothelium-independent (nitroglycerin-mediated; GTN-mediated) vasodilation. Endothelial function measured by FMD is significantly impaired in women with SLE compared with controls [108, 167]. To date, SLE itself, older age, menopausal status and hypertension, but none of the other traditional risk factors, have been shown to be associated with impaired FMD [167, 168]. Although blunted FMD has not been proven to be a surrogate marker of structural CAD in SLE, in asymptomatic individuals, its presence points to functional cardiovascular disease. Increased carotid intima media thickness (c-imt) and plaque, assessed by external vascular ultrasound, are associated with an increased risk of MI and stroke in men and women without a history of cardiovascular disease [169, 170]. In a study by Manzi et al., carotid plaque and IMT, surrogate measures of coronary atherosclerosis, were determined by B mode ultrasound in 179 women who had SLE, of whom 26 had previous arterial events. Sixty-eight (32%) of the women had at least one focal plaque [21]. Independent determinants of plaque were older age, higher SBP and higher LDL-C at the time of study, prolonged treatment with prednisone and a previous coronary event. Older age, elevated pulse pressure, a previous coronary event and a higher SLICC damage score were independently related to increased IMT. In a similar study of 78 patients with SLE without clinical atherosclerotic disease, Doria et al found a thickened carotid artery intima in 22 (28%) patients and plaque in 13 (17%) on duplex sonography [139]. In multiple regression analysis, age and cumulative prednisone dose were associated with carotid abnormalities, while age and hypertension during the study period (as defined by SBP > 33 33

58 140 and/or DBP > 90 mmhg and/or use of antihypertensive drugs) correlated with higher mean and maximum IMT. Likewise Roman et al. showed that women with SLE had a significantly higher prevalence of carotid plaque than a control population (37% vs 15%, p<0.001) [22]. As discussed earlier, in this study, independent predictors of plaque included demographic and disease-related variables. The same investigators later showed that older age at diagnosis, longer duration of SLE, and higher homocysteine concentration are independently related to progression of atherosclerosis [119]. Selzer et al. undertook a cross-sectional study of 214 women without clinical cardiovascular disease enrolled in the Pittsburgh Lupus Registry [138]. B-mode ultrasound was used to measure carotid plaque and IMT and Doppler probes were used to collect pulse-wave velocity (PWV) waveforms from the carotid and femoral arteries as a measure of aortic stiffness. Independent determinants of plaque included older age, higher SBP, lower HDL-C and antidepressant use. Independent determinants of the highest quartile of IMT were older age, higher pulse pressure, lower albumin, elevated C- reactive protein, high cholesterol and high glucose. Aortic stiffness was associated with older age, higher SBP, higher C3 level, lower white blood cell count, higher insulin levels and renal disease. There are numerous other methods for detection of early or subclinical disease in SLE. Among these, electron beam computed tomography (EBCT) used to measure coronary artery calcification is a sensitive and specific marker for the presence of obstructive CAD [171]. In asymptomatic adults, EBCT of the coronary arteries has been shown to predict CAD-related death and nonfatal MI and the need for revascularisation procedures at three to four years [172]. In addition, in patients undergoing angiography, the extent of coronary artery calcification (CAC) on EBCT, measured by the CAC score, has been shown to be highly predictive of future cardiac events, thus adding prognostic information [173]. Asanuma et al used EBCT to screen for the presence of coronary artery calcification in 65 patients with SLE and 69 control subjects with similar baseline demographics [174]. Patients and controls with a history of cardiovascular disease were excluded. CAC was 34 34

59 35 more frequent in patients with lupus (20 [30.8%]) than in controls (6 [8.7%]), p= The mean CAC score was also significantly higher in patients compared with controls. Similarly, Manger et al. sought to determine the prevalence and correlates of CAC in 75 female patients with SLE aged less than 50 years [175]. Overall, 21 (28%) of 75 patients had CAC. In multiple regression analysis, cigarette smoking, reduced renal function, high C3, and a cumulative steroid dose above 30 g were the most important CAC-associated factors in the lupus cohort. Von Feldt et al. have shown that homocysteine concentration, glomerular filtration rate, age, and disease duration are independently associated with CAC in patients with SLE [176]. In summary, despite differences in the technique used, recent cross-sectional and casecontrol studies have consistently demonstrated the prevalence of sub-clinical CAD in SLE to be 30 to 40%. In a therapeutic context, diagnostic modalities used to detect asymptomatic CAD may have a place in the algorithm used to screen patients for risk of future ischemic events Myocardial perfusion imaging Hosenpud et al first suggested in 1984 that myocardial perfusion defects might be common in patients with SLE [177]. They selected 26 patients with SLE irrespective of cardiac history to undergo exercise thallium-201 myocardial scintigraphy. Ten (38.5%) patients had reversible and/or fixed defects. Later Bruce et al determined the prevalence of myocardial perfusion abnormalities among 133 women with SLE [178]. A dualisotope (DIMPI) SPECT technique with thallium-201 as well as technetium 99msestamibi was used to minimize breast artifact and enhance specificity [179]. Following dipyridamole-induced stress, a perfusion defect was detected in 41 (35%) patients with no history of CAD. Eleven (85%) of 13 patients with a history of angina or MI had perfusion defects. Overall 52 (40%) patients had an abnormality of myocardial perfusion. Factors associated with an abnormal myocardial perfusion scan (DIMPI) included current hypertension, elevated cholesterol ever and elevated TC:HDL-C ratio [180]. Similarly Sella et al. found myocardial perfusion abnormalities in 23 (28%) of 82 female patients with SLE who did not have a prior history of CAD. In multiple regression analysis, lower 35

60 36 HDL-C, diabetes and current vasculitis were significantly associated with perfusion abnormalities [181]. As discussed earlier, there is poor correlation between myocardial perfusion imaging and coronary angiography. Conventional angiography does not allow imaging of the coronary microvasculature where in SLE the pathophysiology may lie, and myocardial perfusion imaging may be detecting more than just structural heart disease. In a study wherein both myocardial perfusion scanning and brachial artery ultrasound were performed in 92 women with SLE who did not have clinical CAD, we reported poor agreement between the two tests (Kappa 0.21, 95% CI: 0.1 to 0.41) [182]. There are several possible explanations for this discordance. An abnormal FMD with normal MPS is in keeping with the hypothesis that endothelial dysfunction is a precursor to myocardial ischemia, whereas normal FMD with abnormal MPS may reflect external influences such as disease factors or therapy at the time of study, which may lead to pseudo-normalisation of FMD. Furthermore these two modalities may be assessing different aspects of vascular function, for example macro- vs. micro- vascular function that may not be governed by the same regulatory mechanisms. The preliminary findings of this study indicate that brachial FMD and myocardial perfusion scintigraphy may be complementary investigations in assessing the cardiovascular health of patients with SLE and that they should not be used interchangeably The relationship between subclinical and clinical CAD The presence of a perfusion defect on dipyridamole thallium myocardial scintigraphy (MPS) has been shown to be an independent predictor of subsequent coronary events in a large unselected population in whom the majority underwent MPS to investigate possible coronary symptoms [183]. Likewise in diabetes mellitus, myocardial perfusion imaging has been shown to have high sensitivity for the detection of angiographically significant coronary stenoses [184]. Recently, we have shown that scintigraphic myocardial perfusion defects are predictive of subsequent coronary events among patients with SLE (adjusted hazard ratio 13.0), independently of disease activity and traditional Framingham risk factors [185]. Elliott et al. have shown that in SLE patients who do not 36

61 have symptomatic CAD, increased carotid IMT and the presence of carotid plaque are also predictive of subsequent coronary events, independently of traditional cardiovascular and SLE-specific risk factors (adjusted hazard ratio for cardiovascular events in presence of carotid plaque 5.97, 95% CI: 1.52, 23.38) [186]. The role of screening for presymptomatic CAD in SLE using MPS or carotid ultrasound is yet to be defined. In summary, to date, many investigators have identified risk factors for cardiovascular disease in SLE using markers of subclinical CAD such as brachial FMD or carotid IMT. From a methodological point of view, this approach increases statistical power to show associations between independent and outcome variables, as many more patients with SLE have subclinical CAD compared with clinical CAD. However, not all patients with subclinical CAD go on to develop coronary events. Furthermore, with the exception of myocardial perfusion defects and increased carotid IMT, the relationship between these surrogate markers of CAD and actual CAD events in the context of SLE is not known. In other words, not all cases of subclinical CAD are pre-symptomatic. For these reasons, where possible, use of clinical CAD events such as angina and MI as outcome variables in research studies is preferred. 1.6 Summary of background points of key relevance to thesis 37 SLE is strongly associated with premature CAD, which overall affects 1 in 10 patients. The first coronary event usually occurs a decade or so after diagnosis. Several observational cohort studies, notably those in Toronto, Baltimore and Pittsburgh, have evaluated risk factors for CAD in SLE. Hypercholesterolemia, hypertension and smoking have been shown to be independent risk factors for coronary events in SLE. Other risk factors identified in these studies include older age at SLE diagnosis, longer disease duration and longer duration of corticosteroid use. In the general population, among those who do not have clinically evident CAD or its equivalent such as diabetes mellitus or renal insufficiency, the Framingham risk prediction model based on age, gender, systolic BP, total and HDL cholesterol and smoking, enables calculation of an individual s ten-year absolute risk of coronary 37

62 events. Large retrospective studies have shown that among those that experience coronary events, over 80% have at least one traditional cardiac risk factor. In contrast, despite having a greatly increased risk of coronary events, patients with SLE have ten-year Framingham risk scores similar to the low-risk general population. Women with SLE have significantly fewer coronary risk factors per cardiac event than female controls. Indeed, traditional risk factors only partly account for the increased risk of CAD in SLE, with the relative risk of a coronary event remaining elevated at 7.5 even after controlling for these risk factors. In the general population, other traditional and non-traditional cardiac risk factors include family history, LDL-C, metabolic syndrome and hscrp. The role of these risk factors in SLE-related CAD has not been thoroughly evaluated. In the general population and also possibly in SLE, age and gender often act as effect modifiers for other risk factors. In the general population and also probably in patients with SLE, self-reported information on family history might not be accurate. The additional contribution of family history to CAD risk estimation after inclusion of other traditional risk factors is relatively modest. Epidemiologic studies have shown a continuous, graded relationship between serum TC and coronary risk. At present the desirable TC level for individuals without known CAD who have 0 to one other risk factor is <5.2 mmol/l. Evidence points to the atherogenic role of LDL-C, and the NCEP-ATP III has designated LDL-C the primary target of lipid lowering strategies. The desirable serum LDL-C concentration for individuals without known CAD who have two or more risk factors (or a ten-year Framingham risk score 20%) is <3.4 mmol/l. Based on new clinical trial data, it has been proposed that the desirable LDL-C level for those in the highest risk category, including those with CAD or CAD equivalent or Framingham risk score >20%, once <2.6 mmol/l, be reduced to 2.0 mmol/l

63 Although serum lipids have been shown to track from childhood through adolescence into adulthood, in the general population, TC has been shown to vary by 4 to 11% over time. Although the variability of serum lipids such as TC and LDL-C has not been formally evaluated in patients with SLE, one third of SLE patients have been shown to have variable hypercholesterolemia in the first three years of disease. In the general population, serum TC:HDL-C ratio and TG level are also lipid markers of coronary risk. However, the close association of TG with other lipid factors such as low HDL-C has brought into question its independence as a cardiac risk factor. In the general population, the risk of MI and stroke first becomes apparent as BP rises above 110/75 mmhg. However, hypertension is formally defined as SBP 140 mmhg or DBP 90 mmhg. In the general population there is an interaction between SBP and TC, such that the coronary risk with both risk factors combined is more than additive but less than multiplicative. A similar interaction between hypertension and hypercholesterolemia has also been suggested in SLE. In the general population, sequential studies have shown that BP can drop by an average of 10 to 15 mmhg between clinic visits. Due to the phenomenon of white coat hypertension, BP in some patients may not reach a stable value until the sixth visit or later. Although BP at the time of risk assessment (current BP) is typically used in most prediction algorithms, a report from the Framingham Heart Study found that recent and remote antecedent BP (SBP and DBP) predicted cardiovascular risk incrementally over current BP. Although the variability of BP over time in SLE has not been formally evaluated, due to changes in disease activity and treatment, it is likely that both SBP and DBP take a variable course in patients with SLE

64 As traditional risk factors only partly account for the increased risk of CAD in SLE, there has been much interest in identification of novel risk factors. In the general population, protein markers of inflammation have been studied as indicators of cardiovascular risk. The most extensively studied biomarker of inflammation in cardiovascular diseases is C-reactive protein (CRP). In the general population, baseline CRP concentration is an independent risk factor for cardiovascular disease. High sensitivity assays for CRP (hscrp) are necessary for cardiovascular risk stratification, which is based on discrimination of levels extending below 3 mg/l. Although the precision of the hs-crp assay is 0.3 mg/l, the CDC/AHA recommendation of two measurements to confirm a stable value is a reflection of the fact that systemic inflammatory status can fluctuate over time. In healthy adults, the variability of serum CRP measurements over one year is similar to that for serum TC. For cardiovascular risk stratification, hscrp levels may be divided into quartiles based on relative risk of coronary events in the general population at 5 years. In SLE, CRP has been shown to be associated with intercurrent infection and some clinical features. As CRP is a marker of inflammation, it is also possible that it may fluctuate due to changes in SLE disease activity. In contrast to the general population, most studies to date have not shown baseline hscrp to be correlated with prevalent vascular disease in SLE. Other non-traditional markers of coronary risk in SLE include disease and treatment related factors. While disease activity score (SLEDAI-2K) at presentation is not predictive of coronary events, when the adjusted mean disease activity score (AMS) is used to summarise disease activity over time, for each unit increase in AMS, the hazard for coronary events increases by 8%. Recent lupus disease activity and increases in average daily corticosteroid dose have been shown to be correlated with higher levels of several coronary risk factors such as 40 40

65 TC, SBP and overall two-year CAD risk. The impact of increasing the average daily corticosteroid dose on increasing the level of a risk factor is greater for patients with higher SLE disease activity in the past year. Study of the prevalence and correlates of subclinical CAD in SLE may provide useful information for assessing risk and intervening to prevent the occurrence of actual coronary events. There are a number of emerging techniques for the detection of subclinical cardiovascular disease, including ultrasound of brachial or carotid arteries and EBCT of coronary arteries. Overall the prevalence of subclinical CAD in SLE is 30% to 40%. Scintigraphic myocardial perfusion defects may be detected in 35% of SLE patients who do not have clinical CAD. Factors associated with myocardial perfusion defects include current hypertension, elevated TC ever and elevated TC:HDL-C ratio. Scintigraphic myocardial perfusion defects and increased carotid IMT are predictive of subsequent coronary events in patients with SLE who do not have symptoms of CAD. However, the relationship between other markers of subclinical CAD and subsequent CAD events among patients with SLE in unknown. Furthermore, not all patients with subclinical CAD go on to develop coronary events. Overall, to date, the role of traditional and non-traditional cardiac risk factors in SLE has been evaluated in multiple regression models that use a single-point-in-time baseline measurement of such variables. This fails to capture cumulative exposure over time. As these risk factors are likely to take a dynamic course in SLE, strategies that seek to use time-varying summary measures may more accurately estimate risk of coronary events

66 42 Table Summary of risk factors associated with the development of clinical coronary artery disease in selected studies Toronto, Canada Baltimore, USA Pittsburgh, USA Older age at SLE diagnosis Longer disease duration Hypercholesterolemia Hypertension Hypertriglyceridemia Longer duration of steroid use Other factors Diabetes mellitus, pericarditis, myocarditis, congestive heart failure Obesity, older age at clinic entry Post-menopausal Adapted from Bruce, I.N. Not only but also : factors that contribute to accelerated atherosclerosis and premature coronary heart disease in systemic lupus erythematosus. Rheumatology 2005;44: References: [5, 12, 13] 42

67 43 Figure Model for estimating the ten-year risk of coronary artery disease using data from the Framingham Heart Study Note: this model applies to a patient without diabetes mellitus or clinically evident cardiovascular disease. Adapted from Genest, J. et al. Recommendations for the management of dyslipidemia and the prevention of cardiovascular disease: summary of the 2003 update. CMAJ, 2003; 169(9):

68 44 Table NCEP-ATP III classification of total cholesterol in relation to CAD risk Total cholesterol <5.17 mmol/l (<200 mg/dl) Optimal 5.17 to 6.18 mmol/l (200 to 239 mg/dl) Borderline High 6.20 mmol/l (>240 mg/dl) High Reference: [65] 44

69 45 Table NCEP-ATP III recommended goals for LDL-C Reference: [65] Risk category LDL-C goal High risk: coronary heart disease or equivalent (ten-year Framingham risk >20%) < 2.58 mmol/l (<100 mg/dl) Moderate risk: two or more risk factors (tenyear Framingham risk 10% to 20%) 3.36 mmol/l ( 130 mg/dl) Low risk: 0 to 1 risk factors (ten-year Framingham risk <10%) 4.13 mmol/l ( 160 mg/dl) 45

70 46 Figure SLE Disease Activity Index 2000 (SLEDAI-2K) Reference: [142] 46

71 47 Chapter 2 Hypothesis, Objectives and Overall Framework of Thesis In chapter 1, I reviewed and summarised the literature regarding CAD in SLE, with a particular focus on the role of traditional and non-traditional risk factors. In this chapter I will present the overarching hypothesis of the thesis, delineate the objectives of the study and present an overall framework for the project. 2.1 Hypothesis The overarching hypothesis of my thesis project is that: Due to the dynamic nature of traditional and non-traditional risk factors in SLE, summary measures such as time-adjusted means better reflect risk of subsequent coronary events than single-point-in-time measurement of risk factors. From this hypothesis stem several objectives that overall seek to describe and quantify variability in certain cardiac risk factors measured serially in patients with SLE, and to derive summary measures that reflect exposure over time and may therefore better predict coronary risk than single-point-in-time measurements. 2.2 Objectives My thesis project has 6 specific objectives. These objectives are listed below, along with the thesis chapter in which each objective will be addressed Objective 1 To describe and quantify variability over time and to determine the correlates of TC, SBP and DBP in patients with SLE Chapter 4 47

72 Objective 2 To derive summary measures for TC, SBP and DBP in SLE, which reflect cumulative exposure to these factors over time Chapter Objective 3 To compare summary measures of TC, SBP and DBP with single-point-in-time measurement of these variables, both remote and recent, in terms of ability to estimate risk of subsequent coronary events Chapter Objective 4 To apply the concept of summary measures to several lipid markers of coronary risk, namely LDL-C, HDL-C, TC:HDL-C ratio and TG, in order to determine their role as independent markers of coronary risk in SLE, and to determine lupus-specific CAD risk assessment values for these lipids - Chapter Objective 5 To describe the variability over time, and to determine the correlates of hscrp, a nontraditional cardiac risk factor, in SLE Chapter Objective 6 To apply the concept of summary measures to hscrp in order to determine its role as an independent marker of coronary risk in SLE, and to determine lupus-specific CAD risk assessment values for hscrp Chapter Summary of thesis objectives In summary, the objectives of this thesis project are to describe and quantify variability over time of several cardiac risk factors in patients with SLE, and to derive summary 48

73 measures for these variables that reflect cumulative exposure over time and may therefore better estimate coronary risk than single-point measurements. The concept of deriving summary measures will first be developed using TC and BP, for which the greatest number of data points - corresponding to clinic visits - are available in the Toronto Lupus Database. Here, the potential impact of gaps between visits (i.e. measurements) will also be evaluated. The concept of summary measures will then be applied to other risk factors, including lipid subfractions (LDL-C, HDL-C, TC:HDL-C ratio and TG) and hscrp. LDL-C is an example of a risk factor that is highly correlated with TC, for which there are less frequent measurements in the lupus database. HsCRP is an example of a novel risk factor, which due to its possible association with disease activity and infection is likely to assume a more variable course over time than either TC or BP. 2.4 Overall framework of thesis 49 Figure 1 depicts the overall framework of the thesis. The outcome of interest is coronary events, namely angina, MI and sudden cardiac death. Independent variables include the traditional cardiac risk factors namely TC, SBP, DBP and lipid subfractions, the nontraditional risk factor hscrp, together with other demographic and disease related covariates. The study design is retrospective cohort. Information regarding independent and outcome variables is obtained from the University of Toronto Lupus Database, where such data are collected at each visit and recorded prospectively as part of an ongoing cohort study of the risk and prognostic factors for CAD in SLE. Firstly, the variability of lipids, BP and hscrp over time in patients with SLE will be described and quantified. Then, summary measures, namely means and time-adjusted means for each of these variables will be derived and compared with single-point measurements (recent and remote), in terms of ability to estimate coronary risk. 49

74 50 Figure A depiction of the overall framework of the thesis Abbreviations: CAD = coronary artery disease, SLE = systemic lupus erythematosus, TC = total cholesterol, SBP = systolic blood pressure, DBP = diastolic blood pressure, hscrp = high sensitivity c-reactive protein. 50

75 51 Chapter 3 Setting In chapter 2, I presented the overall hypothesis of the thesis and delineated the objectives of the project. In this chapter I will describe the setting for the project and provide details of the definition and documentation of cardiac risk factors, coronary events and other relevant demographic and disease-related data. The chapter will conclude with a discussion of the handling of missing data and large gaps between visits, and the potential impact of loss to follow-up. 3.1 The University of Toronto lupus clinic and the lupus database The University of Toronto (U of T) lupus clinic was established in Over the last four decades, the rigorous, systematic and standardised collection of clinical and laboratory data in the course of patient care has led to several important discoveries regarding the course and prognosis of SLE [187]. Indeed the association between SLE and premature CAD was first described in the Toronto lupus cohort [7]. Since this seminal observation in the mid 1970 s, further studies from this clinic, summarized in the introductory chapter, have added to the knowledge of risk and prognostic factors for coronary events in SLE Patients and visits In order to be admitted to the Lupus Clinic, a patients must fulfill four or more of the 1971, 1982 or 1997 American College of Rheumatology [188] classification criteria for SLE, or have 3 ACR criteria plus a histological lesion typical of SLE on renal or skin biopsy [ ]. Patients are normally seen at 2 to 6 monthly intervals either by the clinic directors or by fellows trained by them, including myself. 51

76 Collection of demographic and disease-related data Collection and storage of data is approved by the research ethics board of the University Health Network, and patients give informed consent on entry into the clinic. Demographic information recorded for each patient at first clinic visit includes sex, date of birth and race (Caucasian, Black, Asian or other ). At each visit patients are assessed according to a standard protocol including a complete history, examination and laboratory investigation. All documentation is based on a glossary of definitions included in the data retrieval form. At each visit, all clinical and laboratory information required for calculation of the SLE Disease Activity Index 2000 (SLEDAI-2K) score (Chapter 1, Figure 1.1) is obtained, including disease manifestations at the time of the visit or in the preceding 10 days, anti-dsdna antibody levels, complement (C3 and C4) levels, blood cell counts and urinanalysis [141, 142]. Additional information collected at each visit includes menopausal status (defined as a minimum of 12 months of amenorrhoea irrespective of cause), blood pressure (BP), weight, blood glucose, total cholesterol (TC), triglyceride level (TG) and high sensitivity C-reactive protein (hscrp) level (see section 3.2). Fasting levels of low-density lipoprotein cholesterol (LDL-C) and high-density lipoprotein cholesterol (HDL-C) are measured yearly or more frequently (see section 3.2). Coronary events (angina and MI) are recorded at each visit (see section 3.3). Anti-phospholipid antibody (APLA) and lupus anticoagulant (LAC) assays as well as autoantibody panels are performed every six months. Information regarding damage due to disease is documented every 6 months to enable calculation of the Systemic Lupus International Collaborating Clinics / American College of Rheumatology (SLICC/ACR) damage index [192]. Lupus medications, namely corticosteroids, antimalarials (chloroquine and hydroxychloroquine) and immunosuppressives (methotrexate, azathioprine, mycophenolate mofetil, cyclosporine and cyclophosphamide) are recorded at each visit. Dates of starting and stopping these medications are also recorded. However, there is only limited information on average daily dose since last visit. Treatment with each of 52

77 statins, anti-hypertensives, anti-platelet agents and hormone replacement therapy is documented as a dichotomous variable at each visit, but for these medications doses are not routinely recorded in data retrieval forms The lupus database 53 Following the clinic visit, dedicated data entry staff check to ensure that the data retrieval forms have been correctly and completely filled. The data that were recorded in paper form at the clinic are then entered into a dedicated Oracle database, specifically designed and maintained for this purpose. Field ranges and logic checks maximise accuracy of data entry. Periodic chart reviews of doctors notes and correspondence are conducted to ensure that the data retrieval forms have been accurately filled, and cross-referenced against print-outs from the database. The most recent systematic chart review of the entire cohort conducted specifically in order to ensure accuracy of recording of atherosclerotic vascular events was completed in The characteristics of patients in the University of Toronto lupus cohort as of August 2008 are summarized in Table 3.1. Of note, among 1,388 patients seen since 1970, the majority (87.5%) are female. Most patients are Caucasian (72.2%), while 9.8% are Black, 8.7% Asian and 9.3% of other racial origins. Mean age at diagnosis is 31.3 ± 14.0 years. Mean age and disease duration at clinic entry are 35.3 ± 13.7 and 4.0 ± 5.7 years respectively. There are 650 (46.8%) inception patients in the cohort, who were first seen and recruited within 12 months of diagnosis. Mean age and disease duration among the entire cohort as of the most recent clinic visit up to August 2008 is 44.4 ± 15.6 and 13.0 ± 10.1 years, respectively. There are a total of 29,267 visits recorded for the entire cohort. The mean number of visits per patient is 21.1 ± The mean interval between visits is 5.4 ± 9.5 months. One hundred and seventy two (12.4%) have had only one or two visit(s). Six hundred and ninety one patients (49.8%) have been seen at least once in the two years preceding August 2008 ( active patients ). Four hundred and twenty three (30.5%) patients have one or more gaps greater than 18 months between visits, representing 680 (2.4%) of all visit intervals. Excluding documented deaths, 486 (41.8%) patients have not been seen since prior to February 2007 ( lost to follow-up ). Overall, 53

78 from commencement of recruitment in 1970 until August 2008, 226 (16.3%) of patients have died Deceased patients 54 Primary and secondary causes of death, as appear on the death certificate, are recorded in the database for all deceased patients of the Lupus Clinic. The cause of death is recorded based on the International Classification of Diseases Version 9 (ICD-9) codes. For example, death due to myocardial infarction is coded while sudden death of presumed cardiac cause is coded Documentation of cardiac risk factors Sex and age: As outlined in section 3.1.2, sex is recorded at first clinic visit while age (in years) is calculated at each visit based on date of birth. Blood pressure: Systolic and diastolic blood pressures (SBP and DBP) are measured in millimeters of mercury (mmhg) at every visit using a manual sphygmomanometer. The patient is allowed to rest for 5 minutes in the sitting position. The reading is taken on the right arm, supported at the level of the heart. Korotkoff phase V (disappearance) is recorded as the diastolic pressure [193]. Based on recommendations of the Joint National Committee, in this thesis, hypertension is defined as systolic blood pressure 140 mmhg or diastolic blood pressure 90 mmhg [86]. Total cholesterol (TC) level: Cholesterol is measured non-fasting in plasma using a commercial assay (Boehringer Mannheim kit Indianapolis, IN) at every visit and recorded in mmol/l. As discussed in Chapter 1, there are only small, clinically insignificant differences in cholesterol when measured in the fasting or non-fasting state [67]. Based on the NCEP recommendations, in well-controlled laboratories such as ours, the analytical error in serum cholesterol is not allowed to exceed 5% and often the interassay reproducibility of reference samples is within 1-2% [31, 194]. Based on NCEP-ATP III recommendations for the general population and previously published 54

79 papers from the U of T lupus cohort, in this thesis, hypercholesterolemia is defined as cholesterol > 5.2 mmol/l [65, 111]. 55 Low-density and high-density lipoprotein cholesterol (LDL-C and HDL-C) level: Lipoproteins are separated from plasma into subfractions by ultracentrifuging. In patients with triglyceride level < 4.5 mmol/l, LDL-cholesterol concentration can be estimated from the Friedewald formula, where LDL-cholesterol is equal to total cholesterol minus VLDL-cholesterol and HDL-cholesterol [66]. In patients with triglyceride level 4.5 mmol/l, LDL-C level is determined using a more direct method, where HDL-C and LDL-C are separated from each other by manganese chloride/heparin precipitation of LDL-C from the Svedberg flotation <12 subfraction of ultracentrifuged plasma. In the U of T lupus clinic LDL-C is reported in mmol/l. Based on NCEP-ATP III recommendations for the general population and previously published papers from the U of T lupus cohort, in this thesis, elevated LDL-C is defined as a fasting plasma level > 3.4 mmol/l, while low HDL-C is defined as a fasting plasma level <0.9 mmol/l [20, 105]. Triglycerides (TGs): Triglyceride level is measured fasting in plasma using a commercial assay (Boehringer Mannheim kit Indianapolis, IN) at every visit and recorded in mmol/l. Based on NCEP-ATP III recommendations for the general population, in the U of T lupus cohort, elevated TG is defined as a fasting plasma level > 2.3 mmol/l [20, 105]. Smoking: Current smoking is defined as smoking of an average of 1 cigarette/s per day over the past month. Diabetes: Diabetes is defined as fasting plasma glucose > 7.0 mmol/l or diabetes therapy [195]. Once diabetes has been designated present, this status does not change over time, regardless of plasma glucose level. HsCRP: Since 2004, HsCRP has been measured at every visit using an immunoturbidimetric assay (Hitachi 917 analyser, Roche Diagnostics, Indianapolis, IN). The limit of quantitation of this assay is mg/dl. The assay has a sensitivity of mg/dl and very high precision with reported inter-assay coefficients of variation of 55

80 less than 10% across a range of hscrp concentrations from mg/dl to mg/dl [128]. 56 Family history of coronary artery disease: Due to the doubtful accuracy of self-report, to date, information on family history of CAD has not been regularly collected in the lupus clinic. Body mass index and waist: hip ratio: To date information on BMI and waist:hip ratio has not been regularly collected in the lupus clinic. In this thesis I aim to derive summary measures for several key cardiac risk factors in SLE, namely TC, BP (systolic and diastolic), LDL-C, HDL-C, TC:HDL-C ratio, TG and hscrp. Characteristics of data collected and stored in the lupus database for these variables are summarized in Table 3.2. It is important to note that although not all variables are measured at every visit, the measurement of these and other variables is always tied to a clinic visit. 3.3 Definition and recording of cardiac events The occurrence of a new cardiac event since the previous clinic visit is recorded in the data retrieval form and subsequently entered into the database at each visit. Irrespective of the actual event date, a coronary event that has occurred between visits V1 and V2 is recorded at V2, which then becomes the event visit. A coronary event must be presumed related to atherosclerosis and not active SLE. The diagnosis of angina and MI is also confirmed by a cardiologist. The definitions of cardiac events used in the lupus clinic data retrieval forms are as follows: Myocardial infarction (MI) is defined as one of: definite electrocardiographic (ECG) abnormalities, typical symptoms with probable ECG abnormalities and abnormal enzymes ( 2 times upper limit of normal); typical symptoms and abnormal enzymes. 56

81 Angina is defined as severe pain or discomfort over the upper or lower sternum or anterior left chest and left arm, of short duration, relieved by rest or vasodilators. Sudden cardiac death is death with undetermined cause but presumed cardiac. 3.4 Missing data 57 In the Lupus Database, there are two types of missing data. Firstly there are data that are collected only once each year and therefore missing for the visits in between. Secondly there are data that are truly missing, for example if the patient left clinic on a particular day before providing a blood or urine sample or the sample was somehow lost. In this thesis no attempts will be made to impute missing data. Imputation of missing data is fraught with the danger of biased estimation. Furthermore, the size of the Lupus Database is such that even without imputation of missing data, there is ample data for calculation of summary measures and for regression modeling. Characteristics of data collected for the main cardiac risk factors of interest in this thesis, namely BP, TC, LDL-C and hscrp are summarized in Table 3.2. Of note, since the commencement of the lupus clinic in the 1970 s, protocol has mandated that BP and TC be measured and recorded at every visit. Indeed, to date, BP has been recorded as per protocol for 98.9% of all visits. Likewise, TC has been measured and recorded as per protocol for 92.3% of all visits. Since 1998, the lupus clinic protocol has mandated that fasting lipid sub-fractions including LDL-C and HDL-C be measured and recorded yearly for each patient. At present LDL-C is available for only 56.2% of all yearly visits. This is primarily due to the fact that LDL-C must be measured fasting and despite reminders to this effect, many patients attend clinic non-fasting. Finally, hscrp measurement at every visit has been a part of the lupus clinic protocol since To date this information has been recorded for 95.2% of all visits. Overall, for BP, TC, lipid sub-fractions and hscrp, I have used only visits where measurements for these variables are available. Therefore for these particular variables, missing data effectively mean an increased gap between study visits. Gaps between visits are discussed in the section below (section 3.5). 57

82 Gaps between visits Although most patients attending the Lupus clinic are seen every 2 to 6 months, those who live far away or are in remission may be scheduled to be seen only once a year. On occasion, a yearly review appointment may be postponed for various reasons such that in some instances the interval between planned visits may be as long as 18 months. Based on these observations, for the purpose of this thesis project, I have defined a gap between visits as greater than 18 months. Possible reasons for having gaps between visits include personal and professional commitments, intercurrent illness and hospital admission. As of August 2008, among the entire lupus cohort of 1,388 patients, 423 (30.5%) have one or more gaps exceeding 18 months between visits, corresponding to 2.4% of visit intervals (Table 3.1). In this thesis, in order to derive summary measures for TC, SBP and DBP, I have used gap-free datasets. These datasets were created by working backwards from the first coronary event (or last clinic visit in those that remained event-free) for each patient, until a gap of greater than 18 months was encountered. Only data subsequent to the gap interval were then used for analysis. In the case of TC and BP, as only 2.4% of visit intervals exceed 18 months, the gap-free dataset used for analysis likely closely resembles the crude dataset. However, in order to explore the generalisability of the findings of this study, I undertook a formal evaluation of the potential impact of gaps between TC and BP measurements on derivation of summary measures of exposure to these risk factors over time. This analysis is presented in chapter 6. For calculation of summary measures for lipid sub-fractions and hscrp, crude datasets, inclusive of gaps between measurements / visits were used. 3.6 Loss to follow-up In the U of T lupus clinic, loss to follow-up is defined as failure to attend clinic for more than 18 months prior to and including present day. As of August 2008, 486 (41.8%) patients in the U of T lupus cohort were lost to follow-up (Table 3.1). The prevalence and correlates of loss to follow-up appear to differ among lupus cohorts. In the multiethnic LUMINA cohort, cumulative loss to follow-up at five years was 36% [196]. This rate was higher for African-American patients. Patients lost to follow-up tended to be younger 58

83 and more likely to have poor social support and higher levels of helplessness. They also tended to have more renal involvement and more active disease. The investigators concluded that in their cohort, loss to follow-up does not occur at random and differs between ethnic groups, also being higher among patients with more active disease. In contrast, among patients attending the U of T lupus clinic, which constitutes the setting for this thesis, loss to follow-up is more random and does not appear to impact prognosis studies of major outcomes such as mortality. Gladman et al identified 247 (40%) patients lost to follow-up among a cohort of 621 patients in the Toronto lupus cohort [197]. Patients were contacted and encouraged to return for an evaluation or to answer a questionnaire by telephone. Of the 247 patients, 29 had died, 66 returned for a full assessment, 84 completed a questionnaire and 68 (11%) were truly lost-to-follow-up. The lost-to-follow-up patient group had 10% more Caucasians and 6% more males than the patients under regular follow-up. The estimated survival curves of the entire cohort with and without the new lost-to-follow-up data were very similar with no difference in mortality rate during the period in which patients were lost to follow-up. The investigators concluded that for major outcomes such as mortality, loss to follow-up does not appear to significantly bias prospective studies in the Toronto lupus clinic. Since the retrieved 179 lost-to-follow-up patients did not affect survival studies, it is likely that the 68 true lost to follow-up patients would also not have had an impact on prognostic studies. This study by Gladman et al. provides some reassurance that loss to follow-up is unlikely to significantly jeopardize the validity of the findings of this thesis study

84 60 Table Characteristics of patients in the University of Toronto Lupus Database as of August 2008 Characteristic Total cohort Female Postmenopausal Male Race Caucasian Black Asian Other Number (%) or Mean ± SD* (Min Max)** 1,388 1,215 (87.5%) 427 (38.9%) 173 (12.5%) 1,002 (72.2%) 136 (9.8%) 121 (8.7%) 129 (9.3%) Age at diagnosis 31.3 ± 14.0 Age at clinic entry 35.3 ± 13.7 Disease duration at clinic entry 4.0 ± 5.7 Inception patients # 650 (46.8%) Age of entire cohort 44.4 ± 15.6 Disease duration of entire cohort 13.0 ± 10.1 Visits for entire cohort 29,267 Visits per patient 21.1 ± 22.3 Interval between visits 5.4 ± 9.5 Patients with only one or two visits 172 (12.4%) Patients with one or more gaps between visits 423 (30.5%) Active clinic patients $ 691 (49.8%) Lost to follow-up (excluding deaths) 486 (41.8%) Deceased 226 (16.3%) 60

85 61 * SD = standard deviation years # patients seen in clinic within 12 months of diagnosis at patient s most recent clinic visit up to and including August 2008 months a gap is defined as greater than 18 months between visits $ active clinic patients are generally defined as those seen at least once in clinic in the previous 2 years. In this table active clinic patients are those seen at least once in clinic between August 2006 and August Loss to follow-up is generally defined as failure to attend clinic for more than 18 months prior to and including present day. In this table, loss to follow-up is as of August

86 62 Table Data collection characteristics for blood pressure, total cholesterol, lowdensity lipoprotein cholesterol and high sensitivity C-reactive protein BP TC* LDL-C # hscrp Year started regular collection Frequency of collection Every visit Every visit Yearly Every visit Number of visits since start of 29,267 28,725 13,073 6,720 collection Number of patients 1,388 1, Number of visits with data collected as per protocol % visits with data collected as per protocol Mean ± SD* time interval between collected data (months) N (%) visits with gap >18 months between actual collected data 28,958 26,500 7,351 6, % 92.3% 56.2% 95.2% 5.4 ± ± ± ± (2.4%) 671 (2.5%) 721 (9.8%) 39 (0.6%) Abbreviations: BP = blood pressure; TC = total cholesterol; LDL-C = low-density lipoprotein cholesterol; hscrp = high sensitivity C-reactive protein. * Triglyceride levels are also measured at the time of each cholesterol measurement # HDL-C, TC and TG levels are also measured at the time of each LDL-C measurement, and TC:HDL-C ratio is then calculated. 62

87 63 Chapter 4 Variability over Time and Correlates of Total Cholesterol, Systolic and Diastolic Blood Pressure in SLE Publication: Nikpour et al., Variability over time and correlates of cholesterol and blood pressure in systemic lupus erythematosus: a longitudinal cohort study. Arthritis Research and Therapy 2010; 12:R Abstract Background: Total cholesterol (TC) and blood pressure (BP) are likely to take a dynamic course over time in patients with systemic lupus erythematosus (SLE). This would have important implications in terms of using single-point-in-time measurements of these variables in order to assess coronary artery disease (CAD) risk. The objective of this study was to describe and quantify variability over time of TC and BP among patients with SLE and to determine their correlates. Method: Patients in the Toronto Lupus Cohort who had two or more serial measurements of TC, systolic and diastolic BP (SBP and DBP), were included in the analysis. Variability over time was described in terms of the proportion of patients whose TC and BP profile fluctuated between normal and elevated (TC >5.2 mmol/l; SBP 140 mmhg or DBP 90 mmhg), and also in terms of within- and between-patient variance quantified using analysis of variance modelling. Generalized estimating equations (GEE) were used to determine independent correlates of each of TC, SBP and DBP, treated as continuous outcome variables. Results: In total 1,260 patients, comprising 26,267 measurements of each of TC, SBP and DBP were included. Mean±SD number of measurements per patient was 20.8±20. Mean±SD time interval between measurements was 5.4±9.7 months. Mean±SD time interval from the start to the end of the study was 9.3±8.5 years. Over time, 64.7% of patients varied between having normal and elevated cholesterol levels, while the status of 63

88 46.4% of patients varied between normotensive and hypertensive. Using ANOVA, the within-patient percentage of total variance for each of TC, SBP and DBP was 48.2%, 51.2% and 63.9% respectively. Using GEE, independent correlates of TC and BP included age, disease activity and corticosteroids; antimalarial use was negatively correlated with TC (all p values <0.0001). Conclusions: TC and BP vary markedly over time in patients with SLE. This variability is due not only to lipid-lowering and antihypertensive medications, but also due to disease and treatment-related factors such as disease activity, corticosteroids and antimalarials. The dynamic nature of TC and BP in SLE makes a compelling case for deriving summary measures that better capture cumulative exposure to these risk factors. 4.2 Rationale and objective 64 As discussed in Chapter 1, hypercholesterolemia and hypertension are two traditional cardiac risk factors that have been shown to be independently predictive of coronary events in patients with SLE when measured at the first available visit ( baseline ) or defined as abnormal ever during follow-up [5, 12, 13]. There is evidence that in the first three years of disease, one third of patients with SLE have variable hypercholesterolemia, with cholesterol levels that fluctuate between normal and abnormal, which in this case is defined as total serum cholesterol >5.2 mmol/l [111]. Similarly, in the general population, SBP and DBP have been shown to vary over time, a phenomenon that likely also affects SLE patients in whom both disease manifestations and treatments may affect blood pressure [68, 90, 91]. To date the variability over time of TC, SBP and DBP over the course of disease in patients with SLE has not been rigorously evaluated. The work presented in this chapter relates to objective 1 of the thesis, which is to describe and quantify variability over time of TC, SBP and DBP and to determine their correlates in patients with SLE. This chapter will set the scene for the thesis objectives that follow. 64

89 Patients Among the U of T lupus cohort, patients who had two or more serial measurements of TC, SBP and DBP, were included in the analysis. 4.4 Methods TC, SBP and DBP and other variables In addition to TC, SBP and DBP, data on patients demographic profile (including age, sex, menopausal status and race), disease duration, disease activity, medications, intercurrent infections, smoking and diabetes were prospectively collected, stored and tracked in the U of T lupus database at each clinic visit for the period from study entry (first clinic visit) up to the most recent visit as of August These variables have previously been defined in sections and 3.2. To reiterate, hypercholesterolemia was defined as TC > 5.2 mmol/l. Hypertension was defined as SBP 140 mmhg or DBP 90 mmhg. Each measurement of TC, SBP and DBP was tied to a clinic visit. We only used visits wherein all three of TC, SBP and DBP had been measured and recorded Statistical analysis Characteristics of patients in the study together with the total number, frequency and values of TC, SBP and DBP measurements were described. The proportion of patients with normal or elevated TC, SBP and DBP at study entry and during follow-up was determined. Method of moments analysis of variance (ANOVA) modelling was used to quantify total, within- and between-patient variance in TC, SBP and DBP, each treated as continuous variables. Linear regression modeling with analysis of repeated measures was performed using generalized estimating equations (GEE) in order to determine the independent correlates of each of TC, SBP and DBP ( outcome variables). Predictor/independent variables ( covariates ) included sex, age, disease duration, SLEDAI-2K score, infection, diabetes, smoking and treatment with corticosteroids, antimalarials, immunosuppressives, 65

90 antihypertensives and lipid lowering medications. For each covariate the measurements used were those recorded at the time of (i.e. coincident ) with each measurement of SBP or DBP. In the model used to determine correlates of TC, hypertension was also included as a covariate, while in the model used to determine correlates of SBP and DBP, hypercholesterolemia was also included as a covariate. Modelling was repeated using only female patients. In these models, in addition to the aforementioned independent variables, menopausal status and hormone replacement therapy were also included as covariates. I performed all statistical analyses using SAS version 9.1 (SAS Institute Inc., Cary NC) Results In total 1,260 patients were included in the analysis, comprising 26,267 measurements of each of TC, SBP and DBP. The characteristics of these patients are summarized in Table 4.1. The patients were mostly female (88.3%) and Caucasian (73%). Among the female patients 224 (20.1%) were menopausal at study entry and 445 (40.0%) were menopausal either at study entry or during follow-up. Mean ± standard deviation (SD) age at first clinic visit and at entry to study were 35.0 ± 13.6 and 35.4 ± 13.7 years, respectively, indicating that for almost all patients the first clinic visit was also the entry visit into the study. Mean ± SD disease duration at first clinic visit and at entry to study were 4.0 ± 5.0 and 4.4 ± 6.0 years, respectively. Among the patients, 42% had their first study visit within 12 months of diagnosis ( inception cohort ). Among noninception patients, at the first study visit, mean ± SD disease duration was 7.3 ± 6.4 years, ranging from 1 to 52 years. Mean ± SD SLEDAI-2K score at first clinic visit and at entry to study were 9.6 ± 7.7 and 8.7 ± 7.0, respectively, indicating moderate disease activity. The total number, frequency and values of TC, SBP and DBP measurements are reported in Table 4.2. For each of TC, SBP and DBP, the mean ± SD and median number of measurements per patient were 20.8 ± 20.8 and 14, respectively. The mean ± SD and 66

91 median time interval between measurements were 5.6 ± 9.7 and 3.7 months, respectively. The mean ± SD and median time interval from the start to the end of the study were 9.3 ± 8.5 and 6.5 years, respectively. The mean ± SD level of TC at the start of study was 5.2 ± 1.7 mmol/l. The mean ± SD level of SBP at the start of study was 123 ± 19.2 mmhg. The mean ± SD level of DBP at the start of study was 77.2 ± 12.0 mmhg. 67 The proportion of patients with normal (or elevated) TC or BP at the start of the study and during follow-up is reported in Table 4.3. Of note, over time, 64.7% of patients varied between having normal and elevated cholesterol levels, with hypercholesterolemia recorded for 36% of the total number of visits. Likewise, the status of 46.4% of patients varied between normotensive and hypertensive, with hypertension recorded for 14% of the total number of visits. The total and the within- and between-patient variance in TC, SBP and DBP determined using method of moments ANOVA is reported in Table 4.4. In this analysis, each of TC, SBP and DBP were treated as continuous variables. In the case of TC, 51.8% of total variance was attributable to variance between patients, while 48.2% of total variance was seen within individuals. For SBP, 48.8% of total variance was due to variance between patients, while 51.2% of total variance was seen within patients. Similarly for DBP, between-patient variance comprised 36.1% of the total variance, whereas with-in patient variance accounted for 63.9% of the total variance. Linear regression modeling with repeated measures analysis using GEE revealed several independent correlates of TC (Table 4.5): coincident age (parameter estimate: 0.009, 95% confidence interval [CI] to 0.014, p=0.0005), coincident SLEDAI-2K score (parameter estimate: 0.04, 95% CI: 0.03 to 0.05, p<0.0001), coincident corticosteroid use (parameter estimate: 0.32, 95% CI: 0.22 to 0.42, p<0.0001), coincident use of immunosuppressives (parameter estimate: 0.17, 95% CI: 0.06 to 0.27, p=0.0017), coincident use of antihypertensives (parameter estimate: 0.19, 95% CI: 0.08 to 0.30, p=0.0009) and coincident hypertension (parameter estimate: 0.34, 95% CI: 0.22 to 0.46; p<0.0001). Coincident use of antimalarials was negatively correlated with TC (parameter estimate: -0.42, 95% CI: to -0.32, p<0.0001). When the model was run using only 67

92 female patients (Table 4.6), in addition to the variables listed above, another independent correlate of TC was coincident hormone replacement therapy (parameter estimate: 0.17, 95% CI: 0.09 to 0.25, p < ). There was also a trend toward a significant association with menopausal status (p = 0.089). Disease duration (parameter estimate: , 95% CI: to , p=0.0008) and coincident lipid-lowering therapy (parameter estimate: -0.09, 95% CI: to -0.03; p=0.004) were negatively correlated with TC. Independent correlates of SBP determined using GEE are listed in Table 4.7. Overall SBP was independently correlated with coincident age (parameter estimate: 0.41, 95% CI: 0.35 to 0.48, p<0.0001), SLEDAI-2K score (parameter estimate: 0.39, 95% CI: 0.28 to 0.50, p<0.0001), use of anti-hypertensives (parameter estimate: 6.44, 95% CI: 4.94 to 7.94, p<0.0001) and hypercholesterolemia (parameter estimate: 3.78, 95% CI: 2.50 to 5.05, p<0.0001). When the model was run using only female patients (Table 4.8), in addition to these variables, other independent correlates of SBP were diabetes (parameter estimate: 2.43, 95% CI: 1.16 to 3.70, p=0.0002) and coincident smoking (parameter estimate: 1.12, 95% CI: 0.20 to 2.04, p=0.017). A trend was noted toward a significant association with menopausal status (p=0.0927). Coincident use of antimalarials (parameter estimate: -1.32, 95% CI: to -0.69, p<0.0001), immunosuppressives (parameter estimate: -1.81, 95% CI: to -1.13, p<0.0001) and lipid lowering therapy (parameter estimate: -1.62, 95% CI: -2.52, -0.73; p=0.0004) were negatively correlated with SBP. Independent correlates of DBP determined using GEE overall mirrored those of SBP (Table 4.9). DBP was independently correlated with coincident age (parameter estimate: 0.08, 95% CI: 0.04 to 0.11, p=0.0001), SLEDAI-2K score (parameter estimate: 0.23, 95% CI: 0.16 to 0.30, p<0.0001), coincident use of anti-hypertensives (parameter estimate: 3.75, 95% CI: 2.83 to 4.66, p<0.0001) and coincident hypercholesterolemia (parameter estimate: 2.60, 95% CI: 1.83 to 3.38; p<0.0001). When the model was run using only female patients, in addition to these variables, coincident disease duration (parameter estimate: 0.03, 95% CI: 0.01 to 0.05, p=0.008) was also independently correlated with DBP. Coincident use of antimalarials (parameter estimate: -0.94, 95% CI: to -0.52, p<0.0001), immunosuppressives (parameter estimates: -0.50, 95% CI:

93 0.94 to -0.05, p=0.028) and lipid lowering therapy (parameter estimate: -1.13, 95% CI: to -0.53, p=0.0002) were negatively correlated with DBP. 4.6 Discussion 69 This study has revealed substantial change in TC, SBP and DBP level over time among patients with SLE. Multivariate regression analysis using GEE has shown an association of TC, SBP and DBP not only with lipid-lowering and antihypertensive therapy, but also with lupus activity and medications and other cardiovascular risk factors. This study of variability and correlates of TC and BP was based on numerous (on average, 20) and frequent (on average, every 5.6 months) measurements of these variables in 1260 patients with SLE, followed on average for 9.3 years. In total a large dataset of 26,267 individual data points were used in analysis of variability and correlates for each of TC, SBP and DBP. Here I chose to report variability in serial measurements taken over time in two ways. First, each of TC, SBP and DBP were dichotomized into normal and elevated values based on conventional cut-points, and over time, the proportion of patients in whom values fluctuated from one category to another was determined. Second, with TC, SBP and DBP treated as continuous variables, total variance in each variable was quantified and dissected into within- and between-patient variance using ANOVA modelling. The latter approach eliminates the need to dichotomise TC and BP values according to cutpoints, which although based on evidence, are somewhat arbitrary. Common to both methods is the assessment of change in mean or average values over time. However, it must be borne in mind that this approach does not capture the trajectory taken by each variable measured serially in each patient. In this study, over a mean and median follow-up of 9.3 and 6.5 years respectively, 8.8% of patients had persistent hypercholesterolemia while almost two thirds (64.7%) had variable hypercholesterolemia. This is even greater variability over time than previously reported in SLE patients in the first 3 years of disease, wherein one third of patients had persistent hypercholesterolemia, while one third had variable hypercholesterolemia [111]. 69

94 The greater variability, and fewer cases of persistent elevation in cholesterol may be due to fluctuations in disease activity over time and the effect of changes to therapy, including the use of corticosteroids and lipid-lowering agents. Furthermore, the longer follow-up in the present study means greater potential for recording of change over time, irrespective of cause. Certainly the variation in cholesterol over time among patients with SLE far exceeds that reported for the general population, in whom in absence of treatment, cholesterol levels tend to be relatively stable over time [70, 71]. Likewise, almost half (46.4%) of all patients in this study had varying hypertension over the duration of the study, whereas only 1.7% had persistent hypertension. Although there are no previous studies with which to compare the proportion of SLE patients who have persistent and variable hypertension, the findings of this study support my original hypothesis that BP likely takes a variable course in patients with SLE. The absolute total variance in TC and BP is reported in Table 4.4. The magnitude of total variance for TC is much smaller than SBP and DBP, reflecting the smaller range of possible values for the former. In addition, TC measurements may be inherently less variable over time due to physiological reasons and also due to the fact that TC is measured in a laboratory using standardized assays that have small interassay variation [31, 194]. Conversely, blood pressure measurements are subject to measurement error by physicians and volatility due to the phenomenon of white coat hypertension. As discussed in Chapter 1, sequential studies in the general population have shown that BP can drop by an average of 10 to 15 mmhg between clinic visits [90, 91]. Thus many patients considered to be hypertensive at initial visits to a clinic turn out to be normotensive. To date there have been no studies to directly compare blood pressure variability over time in SLE patients to healthy population controls. Previous studies have evaluated the role of TC and BP as predictors of atherosclerotic coronary events in SLE; this is the first study to look at these risk factors as outcome variables and to seek to determine their independent correlates. The importance of this approach is twofold. First, this type of analysis provides insight into the reasons for the pronounced variability over time of these cardiac risk factors in SLE. Second, identifying 70 70

95 correlates of TC and BP in SLE aids selection of covariates and interaction terms for inclusion in multivariate models later in the thesis, when the outcome of interest is atherosclerotic coronary events. In this chapter I have used GEE to allow adjustment for the expected correlation between repeated measures over time within individuals ( fixed effects ). These models have shown significant associations between increasing age and each of TC, SBP and DBP. The association between older age, and elevation in lipid levels and blood pressure is well described in the general population [198, 199]. My models have also shown that greater disease activity at the time of measurement is independently associated with higher TC, SBP and DBP. This is a very important observation. As discussed in Chapter 1, Borba et al. have previously noted a significant correlation between SLEDAI scores and all lipid subfractions including TC, as well as an active lupus pattern of dyslipidemia in times of disease activity [150]. Although I found that use of immunosuppressives was significantly and independently associated with elevated TC, it is unlikely that hypercholesterolemia is a direct effect of treatment with these agents. Rather, immunosuppressive use is likely a surrogate for persistent low-grade disease activity that may not be adequately captured by the SLEDAI-2K scoring system. Notably, coincident use of immunosuppressives was negatively associated with both SBP and DBP indicating that while greater disease activity is associated with higher BP, control of disease activity is associated with a reduction in BP. The findings of this study support the long-suspected association between hypercholesterolemia and hypertension in SLE [115]. In this study hypertension and treatment with antihypertensives were significantly associated with TC, while hypercholesterolemia and lipid lowering therapy were significantly correlated with both SBP and DBP. This association highlights the phenomenon of clustering of traditional cardiac risk factors within individuals with SLE and stresses the need for screening for additional cardiac risk factors when one or more risk factors are present

96 As shown in previous studies, concomitant use of anti-malarials was associated with lower levels of TC. Reduction in plasma cholesterol level is one of the direct pharmacologic effects of antimalarials in patients with SLE [ ]. In this study, antimalarial use was also associated with lower levels of both SBP and DBP, but a reduction in BP is not known to be a direct pharmacologic effect of this class of drugs. More likely, this association again points to the link between hypercholesterolemia and hypertension in SLE. Further support for this link was manifest in the association between lipidlowering therapy and both reduced TC and BP. This observation also suggests that lipidlowering therapy may have beneficial effects in patients with SLE, independently of a reduction in cholesterol level. However, the role of lipid-lowering therapy in prevention of atherosclerotic events in SLE can only be definitively assessed in an intervention study. Among women with SLE, other independent correlates of TC and BP were smoking and hormone replacement therapy [200]. However, our analyses were limited by lack of data on pack-years of smoking. The association between smoking and hypercholesterolemia has been well described in the general population and now, in this study has also been demonstrated in women with SLE [201]. In the general population, smoking is also associated with hypertension, in particular elevated SBP; an association that was also found in this study of patients with SLE [202]. Although among postmenopausal women, estrogen has been shown to have a beneficial effect on serum lipid concentrations, progestin contained in most standard HRT regimens partly negates this effect [200, 203, 204]. The net result of these opposing effects is dependent on the patient s age and overall cardiovascular risk profile. The association between diabetes and BP seen here has been well described in the general population [205]. The link between longer disease duration and higher TC and DBP suggests that the accrual of cardiac risk factors occurs over the course of disease and is consistent with the concept that chronic inflammation contributes to cardiac risk through association with traditional risk factors and other as yet undefined mechanisms

97 Finally, this study has confirmed the well-known association between corticosteroid use and hypercholesterolemia [149, 151]. This highlights the need for vigilant monitoring of lipid levels in times of active disease and during treatment with corticosteroids. 4.7 Conclusion 73 TC, SBP and DBP take a dynamic course in SLE, with over half of the total variance over time seen within individual patients. The variability in these risk factors over time is due to changes in disease activity, medications and the accrual of other cardiovascular risk factors. The variable nature of cholesterol and blood pressure in patients with SLE makes a compelling case for deriving summary measures that better capture cumulative exposure to these risk factors over time. 73

98 74 Table Characteristics of patients (n=1260) Characteristic N(%) or mean±sd* Female Menopausal** at entry to study Menopausal** during follow-up Race: Caucasian Black Asian Other 1113 (88.3%) 224 (20.1%) 445 (40.0%) 880 (73%) 119 (10%) 113 (9%) 96 (8%) Age at first clinic visit (years) 35.0 ± 13.6 Disease duration at first clinic visit (years) 4.0 ± 5.0 SLEDAI-2K at first clinic visit 9.6 ± 7.7 Age at entry to study (years) 35.4 ± 13.7 Disease duration at entry to study (years) 4.4 ± 6.0 SLEDAI-2K at entry to study 8.7 ± 7.0 Hypertension at entry to study 190 (15.1%) Hypercholesterolemia at entry to study 528 (41.9%) Diabetes at entry to study $ 30 / 1223 (2.5%) Smoker at entry to study ψ $ 247 / 1235 (20.0%) Corticosteroid use at entry to study $ 763 / 1257 (60.7%) Anti-malarial use at entry to study $ 462 / 1256 (36.8%) Immunosuppressive use at entry to study $ 259 / 1255 (20.6%) 74

99 75 * SD, standard deviation ** Menopause defined as a minimum of 12 months of amenorrhoea irrespective of cause Scores range from 0 to 105, with higher scores indicating more active disease Diastolic BP 90 or Systolic BP 140 mmhg $ For these variables, data were incomplete for a small number of patients. The denominator of the fractions in the second column is the total number of patients from whom the percentage was calculated. Hypercholesterolemia was defined as cholesterol > 5.2 mmol/l Diabetes was defined as fasting plasma glucose > 7.0 mmol/l or diabetes therapy ψ Smoking an average of 1 cigarettes per day over the past month Antimalarials include chloroquine and hydroxychloroquine Immunosuppressives include methotrexate, azathioprine, mycophenolate mofetil, cyclosporine and cyclophosphamide 75

100 76 Table Number, frequency and values of total cholesterol, systolic blood pressure and diastolic blood pressure measurements Mean ± SD* Min, Max** Median Number of measurements per patient 20.8 ± , Time interval between visits (months) 5.6 ± , Time from study start to end (years) 9.3 ± , TC at start of study (mmol/l) 5.2 ± , SBP at start of study (mmhg) 123 ± , DBP at start of study (mmhg) 77.2 ± , Abbreviations: TC = total cholesterol; SBP = systolic blood pressure; DBP = diastolic blood pressure. * SD, standard deviation ** Min, Max, minimum and maximum 76

101 77 Table Proportion of patients with normal and elevated total cholesterol, systolic blood pressure and diastolic blood pressure at baseline and during follow-up Variable Elevated at study start n (%) Persistently normal n (%) Persistently elevated n (%) Varying n (%) Visits elevated (%) TC # 528 (41.9%) 334 (26.5%) 111 (8.8%) 815 (64.7%) 36% SBP Ψ 153 (12.1%) 725 (58.0%) 15 (1.2%) 520 (41.3%) 12% DBP Ψ 114 (9.1%) 804 (64.0%) 7 (0.6%) 449 (35.6%) 7% BP Ψ 190 (15.1%) 654 (51.9%) 21 (1.7%) 585 (46.4%) 14% Abbreviations: TC = total cholesterol; SBP = systolic blood pressure; DBP = diastolic blood pressure. Elevated TC defined as >5.2 mmol/l. Elevated SBP defined as 140 mmhg. Elevated DBP defined as 90 mmhg. Elevated BP defined as either SBP 140 mmhg or DBP 90 mmhg. # mmol/l Ψ mmhg 77

102 78 Table Total, between and within patient variance in total cholesterol, systolic blood pressure and diastolic blood pressure during follow-up Total variance variance Betweenpatient Withinpatient variance % variance between patients % variance within patients TC # 1.9 # 0.97 # 0.91 # 51.8% 48.2% SBP Ψ Ψ Ψ Ψ 48.8% 51.2% DBP Ψ Ψ 43.1 Ψ 76.1 Ψ 36.1% 63.9% Abbreviations: TC = total cholesterol; SBP = systolic blood pressure; DBP = diastolic blood pressure. # mmol/l Ψ mmhg 78

103 79 Table Independent correlates of total cholesterol Variable # Parameter estimate 95% CI** P value Age (years) , SLEDAI-2K score , 0.05 < Corticosteroids , 0.42 < Antimalarials , < Immunosuppressives , Antihypertensives ƒ , Hypertension , 0.46 < ** CI: confidence interval # all variables measured coincident with measurement of total cholesterol SLE Disease Activity Index 2000; scores range from 0 to 105, with higher scores indicating more active disease Antimalarials include chloroquine and hydroxychloroquine Immunosuppressives include methotrexate, azathioprine, mycophenolate mofetil, cyclosporine and cyclophosphamide ƒ Antihypertensives include all classes of drugs used to lower blood pressure Hypertension defined as Systolic BP 140 mmhg or Diastolic BP 90 mmhg 79

104 80 Table Independent correlates of total cholesterol in women only Variable # Parameter estimate 95% CI** P value Age (years) , < SLEDAI-2K score , < Disease duration Ψ , Corticosteroids , 0.36 < Antimalarials , < Immunosuppressives , 0.20 < Antihypertensives ƒ , 0.24 < Hypertension , 0.32 < Lipid lowering meds , HRT $ , 0.25 < ** CI: confidence interval # all variables measured coincident with measurement of total cholesterol SLE Disease Activity Index 2000; scores range from 0 to 105, with higher scores indicating more active disease Antimalarials include chloroquine and hydroxychloroquine Immunosuppressives include methotrexate, azathioprine, mycophenolate mofetil, cyclosporine and cyclophosphamide 80

105 81 ƒ Antihypertensives include all classes of drugs used to lower blood pressure Hypertension defined as Systolic BP 140 mmhg or Diastolic BP 90 mmhg statins $ Estrogen with/without progestin hormone replacement therapy 81

106 82 Table Independent correlates of systolic blood pressure Variable # Parameter estimate 95% CI** P value Age (years) , 0.48 < SLEDAI-2K score , 0.50 < Antihypertensives ƒ , 7.94 < Hypercholesterolemia , 5.05 < ** CI: confidence interval # all variables measured coincident with measurement of total cholesterol SLE Disease Activity Index 2000; scores range from 0 to 105, with higher scores indicating more active disease ƒ Antihypertensives include all classes of drugs used to lower blood pressure Hypercholesterolemia defined as total plasma cholesterol >5.2 mmol/l 82

107 83 Table Independent correlates of systolic blood pressure in women only Variable # Parameter estimate 95% CI** P value Age (years) , 0.48 < SLEDAI-2K score , 0.44 < Antimalarials , < Immunosuppressives , < Antihypertensives ƒ , 7.53 < Diabetes , Smoking , Hypercholesterolemia $ , 3.78 < Lipid lowering meds , ** CI: confidence interval # all variables measured coincident with measurement of total cholesterol SLE Disease Activity Index 2000; scores range from 0 to 105, with higher scores indicating more active disease Antimalarials include chloroquine and hydroxychloroquine Immunosuppressives include methotrexate, azathioprine, mycophenolate mofetil, cyclosporine and cyclophosphamide ƒ Antihypertensives include all classes of drugs used to lower blood pressure 83

108 84 Diabetes defined as fasting plasma glucose > 7.0 mmol/l or diabetes therapy Smoking an average of 1 cigarettes per day over the past month $ Hypercholesterolemia defined as total plasma cholesterol >5.2 mmol/l statins 84

109 85 Table Independent correlates of diastolic blood pressure Variable # Parameter estimate 95% CI** P value Age (years) , SLEDAI-2K score , 0.30 < Antihypertensives ƒ , 4.66 < Hypercholesterolemia , 3.38 < ** CI: confidence interval # all variables measured coincident with measurement of total cholesterol SLE Disease Activity Index 2000; scores range from 0 to 105, with higher scores indicating more active disease ƒ Antihypertensives include all classes of drugs used to lower blood pressure Hypercholesterolemia defined as total plasma cholesterol >5.2 mmol/l 85

110 86 Table Independent correlates of diastolic blood pressure in women only Variable # Parameter estimate 95% CI** P value Age (years) , 0.11 < SLEDAI-2K score , 0.29 < Disease duration Ψ , Antimalarials , < Immunosuppressives , Antihypertensives ƒ , 4.52 < Hypercholesterolemia $ , 2.64 < Lipid lowering meds , ** CI: confidence interval # all variables measured coincident with measurement of total cholesterol SLE Disease Activity Index 2000; scores range from 0 to 105, with higher scores indicating more active disease Antimalarials include chloroquine and hydroxychloroquine Immunosuppressives include methotrexate, azathioprine, mycophenolate mofetil, cyclosporine and cyclophosphamide ƒ Antihypertensives include all classes of drugs used to lower blood pressure $ Hypercholesterolemia defined as total plasma cholesterol >5.2 mmol/l statins 86

111 87 Chapter 5 Approaches to Summarizing Dynamic Cardiac Risk Factors in SLE: The Examples of Cholesterol and Blood Pressure 5.1 Abstract Background: In the previous chapter, I showed that cholesterol and blood pressure take a dynamic course in SLE, fluctuating over time due to changes in disease and treatment. Objective: To calculate summary measures of total cholesterol (TC), systolic & diastolic blood pressure (SBP & DBP) in order to reflect cumulative exposure to these risk factors over time. Method: Patients in the U of T lupus cohort, who had two or more measurements of TC or BP taken before a coronary event or last visit, in whom the gap between measurements did not exceed 18 months, were included in the analysis. For each patient, for each of TC, SBP and DBP, arithmetic mean, time-adjusted mean (AM) and area-under-the-curve (AUC) were calculated for all serial measurements from the first study visit to the visit before an event (or last clinic visit). The potential impact of gaps between visits was assessed by artificially creating incremental gaps between measurements. Results: A total of 17,936 measurements taken in 956 patients were used to calculate summary measures for TC. In the TC dataset, the mean±sd number of measurements per patient were 19±19 with a mean±sd time interval between measurements of 4.3±2.3 months. Mean±SD time from study start to the visit before an event (or last clinic visit) was 6.3±6.4 years. The BP dataset was similar to the TC dataset. For each of TC, SBP and DBP, mean and AM values were similar. Gap analysis revealed that for each of TC, SBP and DBP, the difference between true and evaluated mean and AM was minimal for gaps of 3 years or less between measurements. The calculation of the AM was less susceptible to bias due to large gaps between visits. 87

112 Conclusion: In this chapter, summary measures, namely mean, AM and AUC were calculated for each of TC, SBP and DBP, based on approximately 20 serial measurements of these variables taken in over 950 patients with SLE over 6.5 years. As measurements were frequent (on average every 4 months), for each variable, mean and AM were very similar. Gaps between measurements of up to 3 years did not appear to significantly bias calculation of summary measures of TC and BP as long as the gap between the remaining measurements did not exceed 18 months. 5.2 Rationale and objective 88 Hypercholesterolemia and hypertension, although potentially reversible cardiac risk factors, only partly account for the increased risk of coronary events in SLE [104, 105, 107]. To date the hazard associated with these risk factors in SLE has not been fully quantified. In Chapter 4, under objective 1, the variability over time of total cholesterol (TC), systolic and diastolic blood pressure (SBP and DBP) was described and quantified. Due to this variability, it may be postulated that in patients with SLE, single-point-intime measurement of cholesterol and blood pressure may not adequately reflect cumulative exposure accrued over time. For the general population, NCEP/ATP III and JNC7 guidelines acknowledge the potential problem of fluctuations in lipid levels and blood pressure, respectively, and recommend that for treatment decisions, an average of two measurements be used [57, 65, 86]. One may postulate that an average of numerous measurements is even more likely to reflect the true state of the individual, especially in patients with SLE, where risk factors such as TC and BP assume an even more dynamic course than in the general population. However, an average or arithmetic mean does not convey a sense of how long the variable remains at a certain level. On the other hand, a time-adjusted mean (AM) takes into account the time interval between measurements such that the contribution of the variable to the overall mean is larger or smaller if the time interval between measurements is larger or smaller. Area-under-the-curve (AUC) is an intuitive method of summarising cumulative exposure to a risk factor; however, this approach has certain limitations, which complicate interpretation of findings. The work presented in this chapter relates to objective 2 of the thesis, which is to calculate summary measures for TC and BP in SLE, in order to reflect cumulative exposure to 88

113 these factors over time. Here I have used three approaches to calculate summary measures, namely calculation of mean, AM and AUC. 5.3 Patients 89 For calculation of summary measures of TC, patients in the U of T lupus cohort who had two or more measurements of TC taken before a coronary event (or last visit) in whom the gap between measurements did not exceed 18 months were included in the analysis. Likewise, for calculation of summary measures of SBP and DBP, patients with two or more measurements of BP taken before a coronary event (or last visit) in whom the gap between measurements did not exceed 18 months were included in the analysis. These gap-free datasets were extracted by working backwards from the event or last visit until a gap of greater than 18 months between measurements was encountered; only the data distal to this gap were then used for calculation of summary measures. Figure 5.1 is a schematic representation of examples of individual patient data used for calculation of summary measures. The rationale for use of measurements with gaps of 18 months or less is discussed in detail in section 3.5. In short, while most active patients in the U of T lupus clinic are seen every 2 to 6 months, due to various reasons, some active patients may be seen less frequently, perhaps every one or one-and-a-half years. Where the gap between visits exceeds 18 months, a patient is deemed lost to follow-up for that period of time. 5.4 Methods Measurement of TC, SBP and DBP Measurement and recording of TC, SBP and DBP in the U of T lupus clinic is discussed in detail in section 3.2. To reiterate, each measurement of TC, SBP and DBP is tied to a clinic visit. TC is measured in mmol/l, while SBP and DBP are each measured in mmhg. 89

114 Coronary events and event visits The definition and recording of coronary events, namely angina, MI and sudden cardiac death is discussed in detail in section 3.3. Irrespective of the actual event date, a coronary event that has occurred between visits V 1 and V 2 is recorded at V 2, which then becomes the event visit. For patients who have had more than one coronary event, only the first recorded event is used in analysis. Some patients may have had both angina and MI recorded for the first time at a particular visit; this is treated as only one event rather than two. In patients who have not experienced coronary events, the last visit is the most recent clinic visit recorded in the database. The analyses in this chapter are based on data collected for visits up to and including August Calculation of summary measures Mean Using the gap-free datasets, for each of TC, SBP and DBP, an arithmetic mean of all available measurements in each patient was calculated using the formula: X 1 + X 2 + +X i / n i where X 1-i are levels of TC (mmol/l), SBP or DBP (mmhg) measured at visits 1 through to i, and n i is the total number of measurements, which correspond to visits. The mean of each of TC, SBP and DBP is reported in the same unit as the variable itself Time-adjusted mean (AM) Using the same gap-free datasets, in each patient, for each of TC, SBP and DBP, using all available measurements, an AM was also calculated using the formula: n ) i= 2 # x i +x i"1 & % ( t $ 2 ' i n ) i= 2 t i 90

115 where x i is the level of the variable at visit i and t i is the time interval between visit i and i-1. In other words, using the example of TC, the time-adjusted mean is essentially a mean adjusted for the time interval between measurements and is equivalent to the area under the curve of cholesterol against time, divided by the overall time from first to last measurement. The AM of each of TC, SBP and DBP is reported in the same unit as the variable itself. Figure 5.2 is a diagrammatic representation of the calculation of a timeadjusted mean (AM) Area-under-the-curve (AUC) Using the gap-free datasets, in each patient, for each of TC, SBP and DBP, using all available measurements, AUC was calculated using integral calculus and the formula: i"1 # f ( x) dx i 91 where for the example of TC, y = ƒ (x) is the equation for the curve of TC plotted against time, and the above formula denotes the area under this curve, for the time interval between visits i and i-1. AUC is reported in mmol/l multiplied by t and mmhg multiplied by t for each of TC and BP respectively, where t is the unit of time, in this case months. For any given visit, the summary measure for each of TC, SBP and DBP is calculated up to and including the visit before (Figure 5.3) Correlations The correlations between each of TC, SBP and DBP and their associated summary measures at or up to and including each visit were determined using Pearson s correlation coefficient and presented as correlation matrices Gap analysis Patients in the U of T lupus cohort who had TC and BP measurements taken simultaneously, over at least ten visits, spanning at least five years, with no gaps greater than 18 months between measurements, were identified and included in the gap analysis. 91

116 Figure 5.4 is a diagrammatic illustration of the method used for gap analysis. For each of TC, SBP and DBP, incremental gaps between measurements were artificially created by deleting interval measurements, starting with a gap between measurements of 0.5 years, increasing in increments of 0.5 years up to a gap between measurements of 6.5 years. In creating gaps, an anchor visit was randomly selected and progressively wider gaps were initially created distal to this visit. Once all visits subsequent to the anchor visit were removed, with the exception of the last, the gap was made still wider by deleting visits proximal to the anchor visit until the only two visits left were the first and the last. Following each increment in gap size, the mean, AM and AUC were calculated using the formulae above, for the time interval from the first study visit to the last. For each gap size, the absolute difference between these evaluated summary measures (mean and AM) and the true summary measure were averaged across all patients and plotted on a graph of difference versus gap size. For each gap size, the 95% confidence intervals for the absolute difference between evaluated and true summary measures were calculated and also plotted. The gap size at which this difference first reached 0.25 mmol/l in the case of TC, and 5 mmhg in the case of SBP and DBP was noted. For AUC, the difference between evaluated and true values were expressed as a percentage of the true value and the gap size at which this percentage first exceeded 5% was noted. I performed all statistical analyses presented in this chapter using SAS version 9.1 (SAS Institute Inc., Cary NC). 5.5 Results 92 Characteristics of patients in the TC and BP datasets are presented in Table 5.1. Overall, the BP dataset contained 991 patients, while the TC dataset comprised 956 patients. In each dataset patients were mostly female (88%) and mostly Caucasian (70%). There were a total of 94 first coronary events (75 angina, 25 MI and 2 sudden cardiac deaths; 8 had both angina and MI recorded as first coronary event) in the BP dataset, while the TC dataset contained 86 coronary events (71 angina, 20 MI and 2 sudden cardiac deaths; 7 had both MI and angina recorded as a first coronary event). The mean ± SD age and disease duration at entry into the study were very similar for both datasets (37.1 ±

117 and 6.1 ± 7.9 years respectively, for the BP dataset). Likewise, mean ± SD SLEDAI-2K score and SLICC-DI at study entry were similar in the two datasets (9.2 ± 7.5 and 0.5 ± 1.2, respectively, for the BP dataset), indicating moderate disease activity and minimal disease-related damage. For each dataset, at entry into the study, over 60% of patients were taking corticosteroids, while approximately 40% were on antimalarials and 25% were taking immunosuppressives. In each dataset, approximately 22% of patients were hypertensive, 40% had hypercholesterolemia, 3% had diabetes and 19% were smokers at the start of the study. At study start, in each dataset, 25% were on antihypertensives and 5% on lipid-lowering medications. Detailed definitions of SLEDAI-2K, SLICC/ACR-DI, hypertension, hypercholesterolemia, diabetes, smoking and medications are provided in sections and 3.2, and summarised definitions are provided in the key to Table Summary measures for TC 93 The frequency and values of individual ( single-point ) measurements used for calculation of summary measures for TC are presented in Table 5.2, along with the calculated summary measures for the group as a whole. Overall, the calculation of summary measures was based on 17,936 individual measurements of TC, with a mean±sd of 19±19 (median 12) serial measurements per patient. The mean±sd (median) time interval between measurements was 4.3±2.3 (3.6) months. The mean±sd (median) time from study start to the visit before an event (or last clinic visit) was 6.3 ± 6.4 (4.2) years. The mean ± SD (median) length of follow-up from study start to event (or last visit) was 6.7 ± 6.4 (4.6) years. Among all patients, the mean TC level at the start of study was 5.3 ± 1.6 mmol/l. This was very similar to the mean of first two TC levels (5.3 ± 1.5 mmol/l), mean (5.1 ± 1.2 mmol/l) of all serial TC levels and the AM (5.1 ± 1.2 mmol/l) of all TC levels, averaged for the group. The correlation matrix (Table 5.4) for TC was consistent with these observations. Here, the Pearson correlation coefficient (r) for TC vs. mean TC at each visit was 0.80, p<0.0001, indicating a strong correlation. Likewise, the correlation between mean and AM TC at each visit was very strong (r = 0.99, p<0.0001). Overall there was poor correlation between AUC and the other 93

118 summary measures (AUC TC vs. mean TC, r = 0.12, p<0.0001; AUC TC vs. AM TC, r = 0.11, p<0.0001) Summary measures for BP 94 The frequency and values of measurements used for calculation of summary measures for SBP and DBP are presented in Table 5.3, along with the calculated summary measures for the group as a whole. Overall, the calculation of summary measures was based on 19,579 individual measurements of SBP and DBP, with a mean±sd of 20±20 (median 13) serial measurements per patient. The mean±sd (median) time interval between measurements was 4.2 ± 2.3 (3.4) months. The mean ± SD (median) time from study start to the visit before an event (or last visit) was 6.5 ± 6.7 (4.2) years. The mean ± SD (median) length of follow-up from study start to event (or last clinic visit) was 7.0 ± 6.7 (4.6) years. Among all patients, the mean SBP at the start of study was ± 19.4 mmhg. This was similar to the mean of first two SBPs (123.9 ± 17.7 mmhg), mean (123.5 ± 15.5 mmhg) of all serial SBPs and the AM (123.4 ± 15.6 mmhg) of all SBPs, averaged for the group. The correlation matrix (Table 5.5) for SBP was consistent with these observations. Here, the Pearson correlation coefficient (r) for SBP vs. mean SBP at each visit was 0.76, p<0.0001, indicating a moderately strong correlation. The correlation between mean and AM SBP at each visit was very strong (r = 0.99, p<0.0001). Overall there was poor correlation between AUC and the other summary measures (AUC SBP vs. mean SBP, r = 0.09, p<0.0001; AUC SBP vs. AM SBP, r = 0.09, p<0.0001). Among all patients, the mean DBP at the start of study was 77.6 ± 12.3 mmhg. This was similar to the mean of the first two DBPs (77.6 ± 10.6 mmhg), mean (77.3 ± 9.0 mmhg) of all serial DBPs and the AM (77.2 ± 9.0 mmhg) of all DBPs, averaged for the group. The correlation matrix (Table 5.6) for DBP was consistent with these observations. Here, the correlation coefficient (r) for DBP vs. mean DBP at each visit was 0.69, p<0.0001, indicating a moderate correlation. The correlation between mean and AM DBP at each visit was very strong (r = 0.99, p<0.0001). Overall there was poor correlation between AUC and the other summary measures (AUC DBP vs. mean DBP, r = 0.10, p<0.0001; AUC DBP vs. AM DBP, r = 0.10, p<0.0001). 94

119 Gap analysis The mean±sd number of TC and BP measurements per patient and the mean±sd length of follow-up per patient in this analysis were 32.5±20.7 and 11.5±6.3 years, respectively. The mean ± SD time interval between measurements used for this analysis was 4.4±1.2 months. Total cholesterol (TC): there was a steady and gradual increase in the absolute difference between true and evaluated mean TC with increase in gap size up to a gap of 3.0 years, beyond which there was a sharper rise up to a gap of 4 years and thereafter there was a fluctuating course up to a gap of 6.5 years (Figure 5.5). The gap at which this difference first reached 0.25 mmol/l for mean TC was approximately 3.5 years whereas this difference was not reached for AM TC until the gap exceeded 6 years (Figure 5.5 and 5.6). For AM TC, there was a steady and gradual increase in the absolute difference between true and evaluated values with increase in gap size up to a gap of 3.0 years, beyond which there was a sharper rise up to a gap of 5 years and thereafter there was a fluctuating course up to a gap of 6.5 years (Figure 5.6). For both mean and AM TC, the 95% confidence interval for the difference between true and evaluated mean and AM first began to widen at a gap size of 3 years (Figure 5.5 and 5.6). The percentage difference between true and evaluated AUC TC did not exceed 5% with gap sizes 6 years (Figure 5.7). Systolic blood pressure (SBP): there was a steady and gradual increase in the absolute difference between true and evaluated mean SBP with increase in gap size up to a gap of 4.0 years, beyond which the graph took on a more fluctuating course (Figure 5.8). The gap at which this difference first reached 5 mmhg for mean SBP was 3 years, whereas this difference was not reached for AM TC until the gap was 6 years or more (Figure 5.8 and 5.9). For AM SBP, there was a steady and gradual increase in the absolute difference between true and evaluated values with increase in gap size up to a gap of 4.0 years, beyond which the graph took on a more fluctuating course (Figure 5.9). For both mean and AM SBP, the 95% confidence interval for the difference between true and evaluated mean and AM first began to widen at a gap size of 3 years (Figure 5.8 and 95

120 5.9). The percentage difference between true and evaluated AUC SBP did not exceed 5% with gap sizes 5 years (Figure 5.10). Diastolic blood pressure (DBP): there was a steady and gradual increase in the absolute difference between true and evaluated mean DBP with increase in gap size up to a gap of 4.0 years, beyond which the graph took on a more fluctuating course (Figure 5.11). The gap at which this difference first reached 5 mmhg for mean DBP was 6 years, whereas this difference was not reached for AM DBP throughout the range of gaps (up to 6.5 years) created for this analysis (Figure 5.11 and 5.12). For AM DBP, there was a steady and gradual increase in the absolute difference between true and evaluated values with increase in gap size up to a gap of 4.0 years, beyond which the graph took on a more fluctuating course (Figure 5.12). For mean and AM SBP, the 95% confidence interval for the difference between true and evaluated mean and AM first began to widen at a gap size of 3 and 4 years respectively (Figure 5.11 and 5.12). The percentage difference between true and evaluated AUC DBP did not exceed 5% with gap sizes 5 years (Figure 5.13). 5.6 Discussion 96 In this chapter, in order to develop the concept of summary measures, I chose to work with TC and BP data, for two reasons. Firstly, these are risk factors that have repeatedly been shown to be associated with coronary events in SLE (Gladman, 1987 #75; Manzi, 1997 #29; Petri, 1992 #30; Rahman, 2000 #58; Urowitz, 2007 #51). Secondly, and perhaps more importantly, from a methodological point of view, TC and BP are two variables for which there are numerous and frequently collected data in the U of T lupus database. In addition, the TC and BP datasets used in these analyses contain a sizable number of coronary events (94 for BP and 86 for TC dataset), thus allowing enough statistical power for multivariate regression modeling of up to nine covariates, which will be performed in the next chapter. Here I chose to calculate three summary measures: an arithmetic mean, a time-adjusted mean and an AUC. An arithmetic mean or average is the most common way of 96

121 summarising a group of individual measurements. However, where these measurements are taken serially over time, a simple arithmetic mean does not take into consideration the time interval between measurements. A time-adjusted mean is an average, weighted for the length of time between measurements. As such, the contribution of each measurement to the overall mean is bigger or smaller depending on whether the size of the interval between measurements is bigger or smaller. Where the gap between measurements is small, the mean closely approximates the time-adjusted mean (AM). As illustrated in Figure 5.2, the calculation of an AM is based on the assumption that the variable of interest is increasing or decreasing at a constant rate between measurements. While in reality this may not always be the case, it is a reasonable approximation for most biologic variables, especially when the gap between measurements is not excessive. The precedence for calculation of a time-adjusted mean has previously been set for the SLE Disease Activity Index 2000 (Adjusted Mean SLEDAI-2K; AMS) (Ibanez, 2003 #73). The AMS has been shown to be a predictor of both atherosclerotic vascular events and mortality in SLE (Ibanez, 2005 #74). AUC, although an intuitive way of summarizing exposure over time, is difficult to interpret. Firstly, being intimately tied to the length of follow-up, AUC can only get progressively bigger over time. Therefore, absolute values provide no sense of the rate of change over time. Furthermore, AUC, being a product of time, is not reported in the same units as the original variable. For this reason, for each of TC, SBP and DBP, the lack of correlation between single-point measurement and AUC at each visit seen in this study is not surprising. On the other hand, both mean and AM values can theoretically get bigger or smaller over time and are reported in the same units as the original variable (mmol/l for TC and mmhg for SBP and DBP), making their interpretation easier. As shown in Figure 5.3, for any given visit, the summary measure for each of TC, SBP and DBP was calculated up to and including the visit before. In this way, cumulative exposure to each risk factor temporally precedes a particular visit. This is of importance to the analyses later presented in chapter 6, where summary measures are used as predictor variables in estimating coronary risk

122 Despite the high percentage within-patient variance in both TC and BP demonstrated in Chapter 4, the analyses in this chapter revealed that mean and AM TC, SBP and DBP, averaged for the group, were very similar to the first available ( baseline ) measurement, averaged for the group. The likely explanation for this finding is that although the percentage within-patient variance is high, the absolute within-patient variance is not very large. As expected, the standard deviation (SD) for the group mean of summary measures was smaller than the SD for the group mean of individual baseline measurements. There was strong correlation between single-point and summary measures of TC (correlation coefficients 0.8 and over) calculated up to and including each sequential visit, except for AUC. Notably, there was very strong correlation between mean and AM TC up to and including each visit, reflecting the fact that overall, among patients in the U of T lupus cohort, TC measurements were regular and frequent. Likewise, there was relatively strong correlation between single-point and summary measures of SBP (correlation coefficients approximately 0.75) calculated for the interval up to and including each visit, except for AUC. As with TC, there was very strong correlation between mean and AM SBP at each visit, reflecting the fact that overall, among patients in the U of T lupus cohort, BP measurements were regular and frequent. In the case of DBP, with the exception of AUC, the correlation between single-point and summary measures calculated at each visit was moderate (correlation coefficients approximately 0.70). Again, due to frequent and regular measurements of BP in the dataset, there was very strong correlation between mean and AM DBP at each visit. The lack of correlation between AUC and each of single-point, mean and AM values is not surprising as the AUC is measured in different units and can only increase over time, progressively assuming larger values with each subsequent visit. The analysis of the potential impact of gaps between visits relates to the generalisability of the methods used for calculation of summary measures. As the summary measures in this chapter were calculated using gap-free datasets, the validity of calculation of these summary measures in a real-life context, where patient visits may not be as regular or frequent needs to be evaluated. As outlined in Chapter 3, for TC and BP, among the U of 98 98

123 T lupus cohort, only 2.4% of visits have intervals greater than18 months and therefore, it is unlikely that the analyses presented here would have differed significantly had I used the crude gap-inclusive datasets. One strategy for assessing the potential impact of large gaps between measurements would be to apply these calculations to a crude or gap-inclusive dataset, where a larger percentage of visits have an interval greater than 18 months. Indeed, this is done later in Chapter 7 in the case of lipid and lipoprotein risk factors for CAD, for which there are less frequent measurements and larger gaps between visits among the U of T lupus cohort. Another strategy is to artificially create gaps of increasing size in the crude dataset for TC and BP. In this so-called gap analysis, I chose to include patients who had at least 10 measurements of TC and BP over at least 5 years. These selection criteria were based on a need to work with a dataset containing numerous measurements per patient, taken over at least a moderate length of follow-up in order to allow sufficient scope for the creation of progressively widening gaps between visits. As the artificially-created gap between visits became bigger, calculation of the difference between true and evaluated summary measures was based on fewer measurements per patient, resulting in wider confidence intervals. Population studies have shown that reductions as small as 1.4 mmol/l in TC level and 5 mmhg in DBP result in significant reductions in cardiovascular risk (Neal, 2000 #254; Shepherd, 1995 #281). Taking a conservative approach, based on this evidence, I defined the maximum acceptable differences between true and evaluated summary measures a priori, as 5 mmhg for SBP and DBP and 0.25 mmol/l for TC. For mean TC, the gap size at which this pre-designated difference in true and evaluated value was first reached was 3.5 years, whereas this threshold for AM TC was not reached until the gap size exceeded 6.5 years. Likewise, for SBP, the difference between true and evaluated mean first reached 5 mmhg at a gap size of 3 years, whereas this threshold for AM SBP was not reached until the gap size exceeded 6.0 years. In the case of DBP, the difference between true and evaluated mean first reached 5 mmhg at a gap size of 6 years, whereas this difference was not reached for AM DBP throughout the range of gaps (up to 6.5 years) created for this analysis. Overall, these findings mean that in the case of TC and BP, even large gaps between measurements are unlikely to significantly compromise the calculation of summary 99 99

124 100 measures (mean and AM) in SLE patients. Furthermore, it appears that by factoring in the interval between measurements in its calculation, an AM is less susceptible to the effect of large gaps between visits. In analysis of the potential impact of gaps between measurements on the validity of AUC calculations, I defined the maximum accepted difference between true and evaluated AUC as 5% of the true AUC value. I based this cut-point on the generally accepted testretest variability of 5 to 10% for most biologic variables. The gap sizes that corresponded with this cut-point for each of TC, SBP and DBP were 6.0, 5.0 and 5.0 years, respectively. As such, even gaps as large as 6 or more years are unlikely to compromise the validity of calculation of an AUC summary measure for TC and BP. Although AUC is apparently the summary measure least susceptible to the effect of gaps between measurements, for reasons discussed earlier, AUC may not be the most desirable summary measure in this context. Of course, the above conclusions drawn from the gap analysis are subject to the constraints under which the analysis was performed. Namely, despite incremental gap sizes, the remaining measurements from which the summary measures are calculated must indeed be relatively numerous and frequent (<18 months apart). In reality, the accurate calculation of summary measures of serial data is reliant not only on the maximum size of the gap between measurements but also on the total number of measurements from which summary measures are calculated, the mean length of time between measurements and the total length of follow-up from first to last measurement. In this chapter, only the effect of gaps between measurements was systematically assessed. Finally, a limitation to this type of gap analysis is that it evaluates the impact of gaps on group summary measures and does not take into consideration the impact of gaps on summary measures for the individual patient. In this thesis all analyses are based on data for groups of patients. 5.7 Conclusion In patients with SLE, summary measures such as arithmetic mean, time-adjusted mean (AM) and area-under-the curve (AUC) may be calculated in order to capture 100

125 101 cumulative exposure to such dynamic risk factors as total cholesterol (TC), systolic and diastolic blood pressure (SBP and DBP) over time. When these variables are measured frequently, the arithmetic mean approximates the AM. Mean and AM are calculated and expressed in the original units of the variable (mmol/l for TC and mmhg for SBP and DBP), and may get bigger or smaller over time. AUC is calculated and expressed as a product of the original units of the variable and time. Therefore, with each subsequent measurement, AUC may only get bigger. By factoring in the interval between measurements in its calculation, an AM is less susceptible to inaccuracy due to large gaps between visits. Gaps as large as 3 years between measurements of TC and BP do not significantly affect the accuracy of calculated summary measures for groups of patients, provided the remaining measurements are relatively frequent (less than 18 months apart) and numerous. 101

126 102 Figure Schematic representation of hypothetical examples of individual patient data used for calculation of summary measures. 102

127 103 In Patient A, from the first clinic visit (also first study visit V 1 ) to the visit before event or last study visit (V E/L-1 ), there are no gaps between visits that exceed 18 months. Therefore, all nineteen measurements corresponding to clinic visits (from V 1 to V E/L-1 ) are used in calculating summary measures. In Patient B, as in patient A, there are no gaps between visits that exceed 18 months. Therefore, all measurements from V 1 to V E/L-1 are used to calculate summary measures. However, in Patient B, there are fewer measurements (nine in total) from which to calculate summary measures. In Patient C, there is a large gap (>18 months) between visits 4 and 5, but no gaps exceeding 18 months between visit 5 and 8 (V E/L-1 ). Therefore, visit 5 is designated the first study visit (V 1 ) and measurements from four visits (V 1 to V E/L-1 ) are used to calculate summary measures. In Patient D, as in Patient C, there is a large gap (>18 months) between visits 3 and 4, but no gaps exceeding 18 months between visits 4 and 8. Therefore, visit 4 is designated the first study visit (V 1 ) and measurements from five visits (V 1 to V E/L-1 ) are used to calculate summary measures. Note: the gap between V E/L-1 and VE/L was not allowed to exceed 18 months. 103

128 104 Figure Diagrammatic representation of the calculation of a time-adjusted mean 104

129 105 The calculation of a time-adjusted mean is shown using the example of TC in a hypothetical patient. In this patient, TC was 5.0 mmol/l on January 1 st, 6.0 mmol/l on April 1 st and 4.5 mmol/l on August 1 st. In order to calculate the time-adjusted mean TC from January 1 st to August 1 st, the red and blue areas are calculated as shown, summed, and divided by the total length of follow-up, which in this example is 7 months. Hence the time-adjusted mean (AM) TC is 37.5 divided by 7, which equals 5.36 mmol/l. 105

130 106 Figure Diagrammatic representation of visits (corresponding to measurements) used to calculate summary measures for each of TC, SBP and DBP. 106

131 107 For any given outcome visit, summary measures for each of TC, SBP and DBP were calculated up to and including the visit before. For instance, in the example above, when visit 4 (V 4 ) was the outcome visit, summary measures were calculated from visit 1 (V 1 ) up to and including visit 3 (V 3 ). 107

132 108 Figure Depiction of the method used to perform gap analysis. 108

133