Objective To synthesise, quantify and compare risks for incident myocardial infarction (MI) across five major types of arthritis in population-based studies.
Methods A systematic search was performed in MEDLINE, EMBASE and CINAHL databases with additional manual/hand searches for population-based cohort or case-control studies published in English of French between January 1980 and January 2015 with a measure of effect and variance for associations between incident MI and five major types of arthritis: rheumatoid arthritis (RA), psoriatic arthritis (PsA), ankylosing spondylitis (AS), gout or osteoarthritis (OA), adjusted for at least age and sex. All search screening, data abstraction quality appraisals were performed independently by two reviewers. Where appropriate, random-effects meta-analysis was used to pool results from studies with a minimum of 10 events.
Results We identified a total of 4, 285 articles; 27 met review criteria and 25 criteria for meta-analyses. In studies adjusting for age and sex, MI risk was significantly increased in RA (pooled relative risk (RR): 1.69, 95% CI 1.50 to 1.90), gout (pooled RR: 1.47, 95% CI 1.24 to 1.73), PsA (pooled RR: 1.41, 95% CI 1.17 to 1.69), OA (pooled RR: 1.31, 95% CI 1.01 to 1.71) and tended towards increased risk in AS (pooled RR: 1.24, 95% CI 0.93 to 1.65). Traditional risk factors were more prevalent in all types of arthritis. MI risk was attenuated for each type of arthritis in studies adjusting for traditional risk factors and remained significantly increased in RA, PsA and gout.
Conclusions MI risk was consistently increased in multiple types of arthritis in population-based studies, and was partially explained by a higher prevalence of traditional risk factors in all types of arthritis. Findings support more integrated cardiovascular (CV) prevention strategies for arthritis populations that target both reducing inflammation and enhancing management of traditional CV risk factors.
- Cardiovascular Disease
Statistics from Altmetric.com
Despite reductions in mortality over time, ischaemic heart disease (IHD) is a major contributor to the burden from chronic diseases and the leading cause of death worldwide.1 A number of modifiable lifestyle factors including smoking, physical inactivity, overweight/obesity, diabetes, hypertension and hyperlipidaemia have long been shown to increase the risk of developing heart disease and are common targets for prevention strategies.2 More recently, inflammation has also been shown to play an integral role in the development, instability and rupture of atherosclerotic plaques leading to acute ischaemic events.3 ,4 This has prompted extensive research examining heart disease risk in inflammatory conditions.
Arthritis is a prevalent inflammatory joint disorder affecting the ageing population characterised by joint pain, stiffness and frequent disability.5 ,6 Several systematic reviews have examined whether individual types of arthritis (rheumatoid arthritis (RA),7 ankylosing spondylitis (AS),8 psoriatic arthritis (PsA),9 gout10 and osteoarthritis (OA)11) are independently associated with various cardiovascular (CV) outcomes, with the strongest most consistent evidence pointing towards relationships between RA and myocardial infarction (MI).7 Previous reviews however included studies with variable study populations, sources for controls, designs, outcomes and definitions and had minimal study inclusion criteria for control of confounding. Furthermore, focus on individual types of arthritis in each review did not allow for comparisons of IHD risk across different types of arthritis. These limitations make it difficult for clinicians, patients and policy makers to get a clear sense of whether and to what extent IHD risk is increased across different arthritis populations and may result in missed opportunities for preventing or lowering IHD risk, particularly in non-RA types of arthritis.
The present systematic review and meta-analysis was undertaken to synthesise best available evidence from population-based studies to quantify and compare risks for incident MI in five major types of arthritis (RA, AS, PsA, gout and OA).
Patients and methods
Data sources and searches
A systematic search developed by an information specialist for population-based studies estimating associations between arthritis and incident MI was performed in MEDLINE, EMBASE and CINAHL databases. Keyword and major subject headings were specified for arthritis and for major types of arthritis (RA, AS, PsA, gout and OA), MI or acute coronary syndrome (ACS) and cohort or case-control study designs. Search results were limited to studies published through January 2015, adults (19+) and English and French languages (sample MEDLINE search strategy provided as online supplementary material). In addition to the database search, we performed manual web searches (eg, Google Scholar) and hand searched bibliographies of retrieved studies, review articles and screened conference abstracts from 2013 to 2014 American and European rheumatology meetings (attempted to contact potential abstract authors by email twice 2 weeks apart) to identify potentially relevant studies that were accepted or in press in a peer-reviewed journal by January 2015.
We included cohort and case-control studies that reported a measure of association (ie, OR, risk ratio, HR) and variability (SE or 95% CI) for effects of arthritis on incident MI or ACS defined according to established clinical criteria, physician diagnosis, validated administrative billing codes or self-reported physician diagnosis and adjusted for at least age and sex. MI was selected as a common objective IHD outcome across studies for meta-analysis. Studies were excluded if the outcome was a composite cardiovascular disease end point or if the study did not include a population-based comparison group. In the event of multiple publications from the same data source, we included studies if extractable estimates were reported for different types of arthritis, but only included the most recent publication if estimates were reported for the same type of arthritis.
Data extraction and quality assessment
The following information was abstracted from each study: country, funding, study design, data source, sample size, follow-up period, sample age range, percent female, exposure and outcome ascertainment and measures of association with 95% CIs, using a predetermined data abstraction template. Study quality was evaluated using the Newcastle-Ottawa Scale (NOS).14 The NOS is an 8-item instrument (total score range 0—lowest to 9—highest quality) evaluating risk of bias in observational studies in relation to three domains: selection of study groups (range 0–4), comparability of study groups (range 0–2) and exposure/outcome ascertainment (range 0–3). NOS quality scores are presented as part of descriptive summaries for each study and did not influence decisions to pool studies in meta-analysis.
All search screening, data abstraction and quality appraisals was performed independently by two reviewers (OS, CT), with discrepancies resolved by consensus (n=3).
Data synthesis and analysis
In order to minimise bias that may result from combining small study effects with high variability, only studies with at least 10 events were eligible to be pooled in meta-analyses. Furthermore, final decisions on whether or not to pool studies were based on quantitative assessments of heterogeneity described below and qualitative assessments based on the number of studies, and the consistency in direction/magnitude of effect estimates across studies so that pooled results would be more interpretable.15
The relative risk (RR) was selected as a common measure of association across studies. Random-effects meta-analysis with inverse variance weighting16 was used to obtain pooled RRs and 95% CIs by type of arthritis, first in studies adjusting for age and sex only, and then for studies also adjusting for at least one of the following traditional risk factors (RF): smoking, obesity/body mass index (BMI), physical activity, hyperlipidaemia, diabetes and high blood pressure. Heterogeneity among studies was assessed with Cochrane's Q statistic and I2 statistic representing the percentage of heterogeneity across studies attributable to between-study differences.17
We used univariate random-effects meta-regression to compare associations with MI across arthritis disease types as well as examined how much arthritis disease type potentially contributed to overall heterogeneity between studies. Additional, planned subgroup analyses to examine other potential sources of heterogeneity between studies included comparisons by age, sex, inception versus prevalent arthritis cohorts, calendar period and geographic region. Changes in pooled effect sizes after excluding each individual study were performed to assess if any single study was strongly influencing pooled results. We assessed publication bias graphically using a funnel plot where the natural log of the ratio of the RR was plotted against its SE, and statistically with the Egger test.18 All analyses were performed using Stata V.12 (StataCorp, College Station, Texas, USA).
Literature search and study election
Figure 1 summarises study screening and selection results. The search strategy identified 4, 285 articles, of which 27 met all criteria to be included in the systematic review. Two articles reported results based on fewer than 10 outcomes,19 ,20 leaving 25 articles eligible for meta-analysis (RA: 13,21–33 AS: 3,34–36 PsA: 1,28 gout: 6,37–42 OA: 243 ,44).
Characteristics of included studies published between 1988 and 2015 are presented in table 1: 26 were cohort studies (9 prospective; 17 retrospective) and 1 was a case-control study. Study populations were from the USA (8), UK (7), Sweden (5), Canada (3), China (3) and Denmark (1), with participant follow-up ranging from 1 to 46 years. Thirteen studies included inception cohorts and 14 prevalent cohorts with arthritis. Of the 24 studies that reported age entry criteria, 16 had no adult age limits and 8 were restricted to middle-aged and/or older adult samples. Higher female-to-male sample ratios were reported in studies of RA and OA, higher male-to-female sample ratios in studies of gout and variable sex ratios in studies of AS and PsA. Traditional risk factors were more prevalent in all types of arthritis in studies with internal population-based comparisons.
Associations between incident MI and all major types of arthritis
Overall, 22/27 (82%) studies reported higher risks for MI in arthritis groups relative to the general population. Cumulative incidence of MI based on 6466 MIs in a combined arthritis sample of 226 962 from 25 studies with available data was 2.85% (95% CI 2.78% to 2.92%). Combined, all five types of arthritis were associated with an average 50% increased risk for MI based on 23 studies adjusted for age and sex only (figure 2A), and an average 30% increased risk for MI based on 17 studies adjusted for at least one traditional risk factor (figure 2B), although quantitative estimates of between-study heterogeneity were high and significant.
Subgroup analyses of incident MI by type of arthritis
We performed subgroup analyses by arthritis disease type. In studies adjusted for age and sex only, risk of incident MI was significantly increased in RA, gout, PsA, OA and tended towards increased risk in AS (figure 2A). Associations with MI were attenuated for all types of arthritis in studies that adjusted for traditional risk factors, and remained significant for RA, gout and PsA (figure 2B). Between-study heterogeneity was high and significant in RA, gout and AS in studies adjusting for age and sex only, however, heterogeneity between studies was much lower and no longer significant for RA and gout in studies with adjustment for traditional RF. Only two studies in OA and AS, respectively, included adjustment for traditional RF and the direction or magnitude of results was highly inconsistent between studies so pooling was not performed for either type of arthritis.
In meta-regression based on studies adjusting for age and sex only, estimated risks of MI in PsA, gout, AS and OA did not differ significantly from RA, and arthritis disease type was not a significant factor driving heterogeneity between studies (p=0.198) (table 2). In meta-regression analyses of studies adjusting for traditional RF limiting to RA, gout and PsA only (OA and AS omitted due to small number of studies and inconsistency across studies), type of arthritis was a significant factor driving heterogeneity between studies (p=0.0035) and associated risks for MI in gout were on average 20% lower than in RA (table 2).
Other subgroup and sensitivity analyses
Results of planned subgroup analyses examining other potential sources of between-study variance including age, sex, arthritis duration, calendar period and geographic region are presented in table 3. Results of subgroup analyses from studies providing age-stratified estimates that could be harmonised across studies showed that while arthritis was associated with increased risks for incident MI across the adult age span, relative risks tended to be highest in young, then middle-aged and older adults. Among studies that provided sex-stratified estimates, pooled relative risks for MI were significantly increased in both women and men. Point estimates were consistently higher in women than in men, but confidence limits between studies overlapped and differences were inconclusive. There were no differences in effects of arthritis by duration, calendar period or region.
Review results showed that risk of incident MI risk was consistently increased across multiple major types of arthritis in population-based studies. Traditional CV RF were also more prevalent in all types of arthritis under study and explained part of the added risk for MI associated with each type of arthritis. Included studies were all of moderate to high quality based on NOS quality scores. Pooled estimates in RA and gout were based on strong and consistent evidence from multiple studies, and for PsA based on one large high-quality study. Evidence in OA and AS was relatively sparse and there were too few studies with inconsistent results with adjustment for RF to be combined in meta-analysis. Visual examination of the funnel plot did not reveal any notable asymmetry (provided as online supplementary material) and the statistical Egger test was not significant (p=0.255) suggesting a low likelihood for publication bias.
Review results have important implications for clinical practice and for planning health services given current and projected increases in arthritis prevalence, and rising trends in obesity and other associated heart disease RF in the general population. First, contrary to RA where CV risk has been extensively studied and increased risk is generally well accepted,45 risk in other types of arthritis have received far less attention and are likely under-recognised. The present study reported consistent increased risk of MI across multiple common arthritis disease types. The general public and healthcare providers should be made aware of more generalised risks for MI associated with arthritis. Second, results that traditional RF were more prevalent in all five types of arthritis under study and consistently explained part of the added risk for MI in each type, point to better management of traditional CV RF as an indirect pathway to prevent or lower MI risk in arthritis populations. This is important for patients with inflammatory arthritis as clinicians may focus on controlling inflammation as a sole means of reducing risk. Similarly, in other types of arthritis with a lower inflammatory burden, patients may not be perceived as higher risk despite having several traditional RF, some of which may evolve over time secondary to arthritis. For example, disability may lead to lower physical activity or weight gain,46 ,47 and certain arthritis medications may improve symptoms of inflammation but have adverse effects on weight and blood pressure.48–50 Lastly, together results of meta-analyses of studies with and without adjustment for traditional RF support common mechanisms leading to increased risk across different types of arthritis (ie, inflammation, traditional RF). However, the relative contribution of these mechanisms may vary by type of arthritis. For example, there was a greater difference in estimated risk for MI in gout than in RA between studies that did and did not adjust for RF. The direct contribution of systemic inflammation may be higher in RA than in gout, and conversely the role of traditional RF may be greater in gout.
Many management guidelines have been developed for individual types of arthritis though few, predominantly those for RA, include recommendations for CV prevention.51 Recently, there has been a shift towards developing more integrated CV prevention strategies for inflammatory conditions. Joint recommendations for managing CV comorbidity in RA, AS and PsA were first developed by the European League Against Rheumatism in 2010.52 Recommendations for managing common comorbidities in RA, PsA and psoriasis were also made by Canadian rheumatology and dermatology expert panels in 2015.53 However, gout and OA were not included in either sets of guidance. Current results combined with increasing prevalence, high levels of OA-attributable disability, associations with obesity and physical inactivity and prolonged treatment with non-steroidal anti-inflammatory drugs/coxibs,54–57 would suggest that prevention strategies recommended for other types of arthritis may help reduce risk in gout and OA as well.
Targeted education and more effective implementation of CV risk management will be important to help mitigate heart disease risk in arthritis populations. It is now recommended that rheumatologists screen/manage traditional heart disease RF given their expertise with complex arthritis treatment regiments and emerging evidence that traditional RF can affect treatment response to certain therapies.53 However, rheumatologists may have limited time in a standard visit to incorporate additional comprehensive CV assessments and continued monitoring, as well as keep up with evolving evidence-based standards for primary and secondary CV prevention. Furthermore, given current shortages of rheumatology specialists in several geographic regions, many types of arthritis, particularly OA and gout are often managed outside of rheumatology.58 ,59 These issues point to the need for coordinated efforts between rheumatologists, internists, primary care and other allied healthcare professionals to ensure that excess CV risk in people with arthritis is appropriately managed.
Strengths of the present review include systematic identification and synthesis of best-available evidence from population-based studies used to estimate effects of five prevalent types of arthritis on MI. This review is the first to compare MI risk across different types of arthritis in studies with adjustment for age and sex only, and with adjustment for traditional RF, respectively.
Limitations should also be addressed. Relatively few studies were identified for PsA, AS and OA, and pooled risks of MI for AS and OA adjusted for traditional RF could not be estimated due to the small number of studies and inconsistency in results between studies. Limiting to studies published in English or French could have led to some bias in pooled effect estimates, although tests for publication bias were not significant. While we excluded clear duplicate publications, there could have been some overlap in study populations in more than one study from the same country. However, pooled estimates were robust to sensitivity analyses that eliminated individual studies (available as online supplementary material). Differences in the types and proportions of patients being treated were not examined and could have contributed to heterogeneity. Despite differences in baseline risks and characteristics of arthritis and heart disease by age and sex,60 less than half of included studies tested for these interactions or reported stratified results. Over a third of studies did not adjust for any traditional RF and studies that did were often limited to baseline measures or lacked information on smoking, BMI/obesity and physical activity, increasing potential for residual confounding in pooled estimates. Given the common mechanisms linking arthritis and MI discussed above, risks may also extend to other less prevalent types of arthritis (eg, systemic lupus erythematosus), inflammatory conditions other than arthritis and to other vascular end points but were beyond the scope of the present review and should be explored in future studies. Large longitudinal population-based studies with repeated measures examining effects of different types of arthritis, particularly OA, on IHD and other vascular end points are needed. Stratified analyses by age and sex and mediation analyses examining direct and indirect pathways (and potential variations by type of arthritis) would help inform primary and secondary prevention strategies.
In conclusion, the present study estimated risk of incident MI associated with five major types of arthritis from population-based studies. Results showed that MI risk was consistently increased in multiple types of arthritis, and was partially explained by a higher prevalence of traditional RF in all types of arthritis under study. Study findings support more integrated CV prevention strategies for arthritis populations that target both reducing inflammation and enhancing management of traditional CV RF.
Handling editor Tore K Kvien
Contributors OS designed and prepared the study protocol, developed the search strategy in conjunction with an information specialist, performed search screening, data abstraction, quality appraisal of studies, carried out the analysis and interpretation of results and drafted the manuscript. She is guarantor. CT performed search screening, data abstraction, quality appraisal of studies, revised and approved the manuscript. SH-J, RHG and EMB contributed to the study design, protocol, analysis plan, interpretation of the results and revised and approved the manuscript.
Funding OS received a doctoral training award from the Fonds de la Recherche du Québec—Santé. RHG is supported as a Clinician Scientist in the Department of Family and Community Medicine at the University of Toronto and at St. Michael's Hospital.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Add data provided within manuscript or as online supplementary material.
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.