Objectives Screening for cardiovascular risk factors and treating hyperlipidaemia with statins are recommended to reduce the increased cardiovascular risk in individuals with rheumatoid arthritis (RA). However, poor compliance with statins may limit their therapeutic benefit. Our objective was to evaluate the impact of statin discontinuation on risk of acute myocardial infarction (AMI) among RA patients.
Methods The authors conducted a population-based cohort study of RA patients with incident statin use followed from May 1996 to March 2006 using administrative health data. Primary exposure was statin discontinuation for ≥3 months at any time during therapy course. The authors used Cox's proportional hazards models and modelled statin discontinuation as a time-dependent variable, while adjusting for age, sex, comorbidities, use of other medications influencing cardiac risk, and proxy indicators of RA severity.
Results During 15 669 person-years of follow-up in 4102 incident-statin users with RA, the authors identified 264 AMI events. Statin discontinuation was associated with 67% increased risk of AMI (adjusted HR 1.67; 95% CI 1.24 to 2.25). There was a 2% increase in risk of AMI with each 1-month increase in the duration of discontinuation (adjusted HR 1.02; 95% CI 1.01 to 1.03). These associations were not modified by timing of first statin prescription, prior AMI status, sex and age (p values for interactions >0.17).
Conclusions These population-based data indicate that RA patients who discontinue statins have increased risk of AMI. Findings emphasise the need to raise awareness, among health professionals and people with RA, of the importance of compliance with statin therapy in RA.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Rheumatoid arthritis (RA) is an inflammatory arthritis associated with systemic inflammation. Cardiovascular disease (CVD) is a significant co-morbidity, with studies reporting up to a threefold risk, independent of traditional risk factors,1,–,3 and a leading cause of mortality,4 with 50% higher risk of CVD death in RA compared with the general population.5
Care for CVD is important in RA management. Clinical guidelines recommend regular assessment of traditional cardiovascular risk factors, with RA considered as an additional risk factor.6 7 When cardiovascular risk is elevated, treatment is indicated, and statins are recommended therapy.6 7 However, deriving therapeutic effect from medications depends not only on physicians prescribing treatment, but also on patients' compliance with therapy.8 Across studies in the general population, statin discontinuation rates range from 15%9 to ≥75%10 with most ≥50%,11,–,15 indicating a problem of great magnitude in real-life patient populations. Using a population-based cohort of RA patients, we previously reported that over an 8-year mean follow-up, 38% of patients permanently discontinued statin therapy.16 Of equal importance to quantifying the problem of statin discontinuation is evaluating its impact on adverse outcomes, yet few studies have addressed this.17,–,20 These limited studies demonstrated that statin discontinuation in non-RA populations is associated with increased risk of new and recurrent acute myocardial infarction (AMI),20 and mortality in people with prior cerebrovascular accident (CVA)17 or AMI.18 19 However, relevant data are lacking for RA, a patient population with increased CVD risk in whom statins may exert a dual benefit due to lipid-lowering and anti-inflammatory effects. Given the importance of patient-specific data to guide clinicians and patients, our objective was to evaluate the magnitude of the risk of AMI associated with statin discontinuation in a population-based cohort of RA patients.
Data source and study sample
We used data from a population-based RA cohort in British Columbia (BC), Canada,21 identified using administrative billing data from the BC Ministry of Health and comprised of adults (≥18 years) who received care for RA between January 1996 and March 2006. The RA definition was the same as previously published for this cohort,21 with individuals meeting inclusion criteria if they had at least two physician visits ≥2 months apart with an RA diagnostic code (International Classification of Diseases, Ninth Revision (ICD-9), 714.x). Exclusion criteria to improve specificity were: (1) at least two visits subsequent to the second RA visit with diagnoses of other inflammatory arthritides (eg, psoriatic arthritis, systemic lupus erythematosus); (2) an RA-coded visit by a non-rheumatologist that was not confirmed on subsequent rheumatologist visit; or (3) no subsequent RA-coded visits during at least 5 years of follow-up. The number of prevalent RA cases included in the cohort in 2006 and BC census data22 yielded a prevalence rate of 0.82%, consistent with RA prevalence estimates.23
We defined a cohort of statin users by selecting individuals who had ≥1 statin prescription(s) between May 1996 and December 2003, using a 4-month period from the date of earliest available prescription data (January 1996) to ensure incident statin use. Utilisation data for all funded health services since 1990, including physician visits and hospitalisations, were obtained. We also obtained complete information on all prescriptions dispensed by pharmacists from January 1996 and mortality data from vital statistics, including date and cause of death.24 No personal identifying information was available, procedures complied with BC's Freedom of Information and Privacy Protection Act, and ethics approval was obtained from the University of British Columbia.
Using data on prescription dispensing date and days supplied, we established statin therapy course for each subject in terms of binary indicators of statin use in each month. Primary exposure was statin discontinuation, defined as ≥3 consecutive months of non-statin use at any time during follow-up. Studies have shown that it takes 2 weeks from therapy initiation for lipid-lowering effects to occur and 4–6 weeks to achieve maximum effect.25 A similar latency probably applies to the potential impact of statin discontinuation on AMI risk; moreover, this impact may be further delayed until changes in lipid levels lead to atherosclerotic plaque formation and thrombotic events. Therefore, we required at least 3 months of consecutive non-statin use prior to considering statin status as ‘discontinued’ (figure 1). This definition also avoids classifying periods of low adherence as therapy discontinuation. Switches between statin types were not considered as discontinuation. Months with resumed statin prescriptions after a period of discontinuation were classified as ‘statin-use’ months until discontinuation criteria were again met, thereby allowing capture of both intermittent and permanent discontinuation of therapy.
The primary outcome was the first AMI during follow-up, including both non-fatal and fatal AMI events. Non-fatal AMI events were ascertained using ICD-9 codes (410.x) in hospital separation data, which included up to 10 diagnoses representing either reason for admission or complications during hospitalisation. The accuracy of ICD-9 codes for AMI is well established, with positive predictive values (PPV) ranging from 89% to 96%.26,–,28 Fatal AMI included AMI deaths occurring outside of hospitals based on ICD-10 codes for AMI (I21) in vital statistics data, and deaths resulting from hospitalised AMI. The validity of ICD-10 AMI codes have also been previously demonstrated, with PPV of 93.5%.29
Assessment of covariates
Factors shown to influence statin discontinuation in RA30 and/or cardiovascular risk, available in our data, were considered as potential covariates. Fixed-in-time binary variables measured over a period of 1 year preceding the start of follow-up evaluated chronic co-morbid medical conditions and were based on physician visits (ICD-9 codes) or medication use. These included diabetes (use of insulin or oral hypoglycaemic agents), angina (411.x, 413.x or use of nitrates), use of cardiac medications – grouped as anti-hypertension medications, congestive heart failure medications and anti-arrhythmia medications, as well as other medications known to influence AMI risk, namely hormone replacement therapy and anticoagulants. We calculated a modified Charlson comorbidity score over the 1-year period preceding the beginning of follow-up.31 32 We also considered any prior AMI and CVA (434.x, 436.x) at any time between 1990 and beginning of follow-up.
The following variables, evaluated over follow-up, were considered as proxy indicators of RA severity: rate of RA-related physician visits, orthopaedic procedures for RA and disease modifying anti-rheumatic drug (DMARD) use. DMARD use was categorised as an ordinal variable, with mutually exclusive categories representing the highest rank attained during follow-up, including no DMARD use (group 1); sulphasalazine or anti-malarials (group 2); methotrexate or intramuscular gold (group 3); leflunomide, ciclosporine-A or other immunosuppressants (group 4); and biologics (group 5). Use of RA medications influencing AMI risk were also included as time-dependent covariates, namely, glucocorticosteroids, traditional non-steroidal anti-inflammatory drugs (NSAIDs), cyclooxygenase-2 (COX-2) selective NSAIDs and methotrexate.
The timing of the first statin prescription relative to RA diagnosis was used to assign the index date, which corresponded to beginning of follow-up. Specifically, the index date for individuals whose first statin prescription was dispensed on or after RA diagnosis was the date of the first statin prescription; the index date for individuals whose first statin was dispensed before RA diagnosis was the date of RA diagnosis (figure 1). Person-time of follow-up was computed from index date to AMI outcome, last healthcare service, death or end of study (31 March 2006), whichever came first.
To estimate the effect of statin discontinuation on AMI risk, we used Cox's proportional hazards models with delayed entry and time-varying covariates.33 34 Since both probability of statin discontinuation and its impact on AMI risk may depend on time since initiation of statin therapy, we ensured that individuals' AMI risks were compared at the same time. Thus, we used time since statin initiation as the time axis, with the origin (time zero) corresponding to the date of the first statin prescription (figure 1). However, since follow-up for the study outcome started at the index date, subjects who initiated statin therapy before RA diagnosis had delayed entry into the risk set. Specifically, since these subjects met our inclusion criteria only after they were diagnosed with RA, they were not considered ‘at risk’ for the study outcome during the time elapsed between their first statin prescription and their RA diagnosis. Had subjects entered the study at the time of their first statin prescription, results could be affected by survival bias.35
We modelled statin discontinuation as a binary, time-dependent covariate, updated monthly with the HR representing the risk of AMI associated with statin discontinuation in the current month. All multivariable Cox's models were adjusted for age at index date and sex. Fixed-in-time variables representing RA duration at index date (months), statin duration at index date (months), and a binary indicator of whether subjects initiated statin therapy before RA diagnosis were forced into final multivariable models. For all other aforementioned covariates, a forward selection procedure, with p<0.05 criterion for entry, was employed.
We considered the possibility that longer duration of statin discontinuation may be associated with increased AMI risk. Therefore, we repeated the analyses to model statin discontinuation as a continuous time-dependent variable that increased with every month since discontinuation. To assess whether timing of first statin prescription relative to RA diagnosis modified the risk of AMI, we added the interaction term between discontinuation status and the binary indicator of first statin prescription before or after RA diagnosis, and tested its statistical significance in the multivariable model. Similarly, we tested potential interactions between statin discontinuation and (1) prior AMI status, (2) sex and (3) age group (<65 years vs ≥65 years). Finally, we conducted sensitivity analyses of our definition of statin discontinuation by repeating the analyses on varying durations of consecutive non-statin use (1 month, 2 months) required prior to defining discontinuation status.
All hypotheses were tested using two-tailed Cox model-based Wald's tests at the 0.05 significance level. Analyses were performed using SAS (Version 9.1).
Our cohort included 4102 RA patients with incident statin use, who contributed 15 669 person-years of follow-up between May 1996 and March 2006. Subject characteristics are summarised in table 1. Women comprised 60% of the cohort and at index date, mean age was 66.6±10.4 years.
Atorvastatin was the most commonly initiated statin (47.9%), followed by simvastatin (21.9%) and pravastatin (12.2%). Overall, 1833 (44.6%) individuals met the study definition for statin discontinuation at least once during follow-up.
Over a mean follow-up of 4 years (median 3.5 years), we identified 264 AMI outcomes: 196 (74%) were first AMI events and 68 (26%) occurred in patients with a history of AMI. Incidence rates of AMI were 4.7 and 1.7 per 100 person-years in individuals with and without a prior AMI, respectively.
When we modelled statin discontinuation as a binary time-dependent variable, univariate Cox's regression analysis showed that current statin discontinuation was associated with an almost 50% increase risk of AMI (unadjusted HR 1.46; 95% CI 1.09 to 1.95). After adjusting for covariates included in the final multivariable model (table 2), the association remained highly statistically significant and became even stronger (adjusted HR 1.67; 95% CI 1.24 to 2.26, p<0.001). Other significant predictors of increased AMI risk included older age, male sex, prior AMI, diabetes, use of anti-hypertension or congestive heart failure medications, as well as current glucocorticosteroid use and rate of RA visits.
When we modelled current duration of statin discontinuation as a continuous time-dependent variable, we also found statistically significant associations. Specifically, with each 1-month increase in duration of discontinuation, the risk of AMI increased by about 2% (adjusted HR 1.020; 95% CI 1.011 to 1.029).
In subanalyses, the impact of statin discontinuation did not vary materially depending on the timing of first statin use (before versus after RA diagnosis), prior AMI status, sex or age (table 3), and none of the corresponding interactions approached statistical significance (all p values for interactions: >0.17). Finally, in sensitivity analyses evaluating our statin discontinuation definition, when discontinuation was defined as requiring 1 or 2 months of consecutive non-use, the adjusted HRs were similar to that obtained in our primary analysis (1.47; (95% CI 1.10 to 1.97 and 1.62; (95% CI 1.21 to 2.17, respectively).
In this population-based study of RA patients prescribed statins, we found a 67% increase in risk of AMI associated with discontinuation of statin therapy. With each additional month in duration of statin discontinuation, the risk of AMI was increased by approximately 2%. This increased risk was independent of age, sex, comorbidities and proxy indicators of RA severity. Findings persisted across different subgroups, stratified by sex, age, prior AMI status or timing of first statin use.
Evaluating adverse cardiovascular outcomes associated with statin discontinuation is particularly relevant in individuals with RA, given the burden of CVD in this population. Both the Nurses' Health Study2 and the Rochester Epidemiology Project3 found a twofold increase in AMI risk in individuals with RA compared with individuals without RA. Given the emphasis on management of cardiovascular risk factors in RA, as outlined in recent guidelines,6 our findings have important implications for the care of people with RA. Not only is it necessary to assess and establish CVD risk among patients and to initiate statins when appropriate,36 it is also essential to monitor and ensure compliance with the prescribed treatment. The risk associated with statin discontinuation quantified in this study will provide physicians and individuals with RA, with patient-specific information that will help them weigh risks and benefits when deciding whether or not to continue statin therapy.
Although compliance with medications has not been well studied in RA and other rheumatic conditions,37 existing data suggest that medication non-compliance is an important problem, particularly in RA. An earlier review estimated that at least 50% of patients are non-compliant with RA therapy irrespective of the intervention.38 Using the same population-based RA cohort, we previously reported a permanent statin discontinuation rate of 38%,16 and found that factors associated with reduced risk of statin discontinuation included older age, prior AMI or CVA, diabetes and anti-hypertension medication use.30 Other studies have also found that the presence of CVD risk factors was associated with better statin compliance.39 40
To our knowledge, this is the first study to evaluate the impact of statin discontinuation on cardiovascular outcomes in an RA population. Our results corroborate findings in the growing literature on adverse outcomes associated with statin discontinuation in the general population. Using administrative data in the Netherlands, Penning-van Beest et al found a reduction in AMI risk associated with continued use of statins in individuals with (RR 0.70; 95% CI 0.54 to 0.91) and without prior CVD (RR 0.70; 95% CI 0.60 to 0.81).20 A study of post-AMI patients in the US found that statin discontinuation was associated with increased risk of 1-year mortality (HR 2.86; 95% CI 1.47 to 5.55).19 A similar association between statin discontinuation and 1-year mortality risk (HR 1.88; 95% CI 1.13 to 3.07) was shown in post-AMI patients using the UK General Research Practice Database.18 Of note, the aforementioned studies evaluated statin discontinuation at a fixed time point, and both the US19 and UK18 studies had 1-year follow-up periods. Our study evaluated impact of statin discontinuation on AMI outcomes over up to 10 years of follow-up. Furthermore, application of time-dependent modelling techniques allowed us to capture the effect of both intermittent and permanent discontinuation, thereby reflecting more accurately actual patterns of statin use.
A number of postulated mechanisms may explain the association between statin discontinuation and AMI risk observed in this study. Aside from loss of cholesterol-lowering effect of statins, potential biological mechanisms based on statin anti-inflammatory effects may also play a role. For example, statin discontinuation studies have demonstrated deterioration of endothelial function41 42 and increased levels of C-reactive protein to pre-treatment levels.43 The observed associations may also be explained by a ‘healthy adherer’ effect, whereby individuals who are more compliant (less likely to discontinue therapy) are those who exhibit healthier behaviours. Observations that good compliance not only with drug therapy but even with placebo are similarly associated with mortality may suggest that compliance to any therapy may be a marker for overall healthier behaviour.44 However, a study of the association between adherence to statin and/or aspirin and CVD outcomes by Wei et al showed that in subjects taking both drugs, adherence to statins but not to aspirin, was associated with lower risk of CVD recurrence (RR 0.64; 95% CI 0.49 to 0.82), while the same was not observed for adherence to aspirin but not statins (RR 0.91; 95% CI 0.72 to 1.15).45 These findings suggest that healthy behaviour may not fully explain the adverse outcomes seen in poorly adherent patients.45
Study strengths and limitations deserve comment. The universal nature of the Canadian healthcare system has provided a population-based cohort of individuals with RA, free of selection bias, thus increasing external validity of our findings. However, observational studies using administrative data are vulnerable to diagnostic uncertainty. Because we used diagnostic codes to define RA, some misclassification of diagnosis likely occurred. However, inclusion of non-RA cases would probably bias results towards the null, given the increased risk of AMI in RA.46 Furthermore, we used the strictest published case definition for RA, which was shown to have a PPV of 0.92 in a validation study against self-reported RA diagnosis,47 48 and improved specificity with additional exclusions as described in the Methods. While use of administrative pharmacy records are well established in pharmaco-epidemiological studies49 and we had complete capture of all medications, data are limited to prescriptions dispensed and we did not have information on whether pills were actually taken or reasons for statin discontinuation. Finally, although we adjusted for all known risk factors for discontinuation and AMI that were available in our data, unmeasured or unknown confounders including lifestyle risk factors for AMI could still affect the results.
In conclusion, our population-based data indicate that patients with RA who discontinue statin therapy have increased risk of AMI. Given the established risk and burden of cardiovascular disease in RA, these findings emphasise the need to raise awareness, among health professionals and people with RA, of the importance of compliance with statin therapy.
Funding This research was supported from an operating grant from the Canadian Institutes of Health Research (CIHR grant number MOP 77605).
Competing interests MAdeV received doctoral training support from the Canadian Institutes of Health Research, the Canadian Arthritis Network/The Arthritis Society of Canada, and the Michael Smith Foundation for Health Research. MA is a James McGill Professor of Biostatistics at McGill University. VG-R's postdoctoral fellowship was supported by Consejo Nacional de Ciencia y Tecnología (CONACYT). DL holds an Investigator Award from The Arthritis Society of Canada and is the Nancy and Peter Paul Saunders Scholar.
Ethics approval This study was conducted with the approval of the University of British Columbia Behavioural Research Ethics Board.
Patient consent Obtained.
Provenance and peer review Not commissioned; externally peer reviewed.