The Meta-Analysis of Observational Studies in Epidemiology guidelines 6
were used as reference for all stages of design, implementation, and reporting of this systematic review and meta-analysis.
We searched for all clinical trials, or observational studies (prospective or retrospective or case-control studies) in which adults received methotrexate, the duration of follow-up was at least 3 months, and reported effect estimates on occurrence of “hard” CVD events (myocardial infarction (MI), coronary heart disease (CHD), sudden death, and/or stroke). We performed searches using multiple online databases, including MEDLINE (see Supplemental Methods
), EMBASE, AGRIS, AMED, Web of Knowledge, CINAHL, CAB abstracts, Cochrane library, conference abstracts (ZETOC), Faculty of 1000, and grey literature sources (SIGLE). We additionally reviewed related articles, hand-searched reference lists, and performed direct author contact. For each database, the years searched included the earliest available online year of indexing up to June 2010, without language restrictions. Key words were methotrexate
, and cardiovascular diseases
. We excluded a priori
studies that only had information on intermediate secondary endpoints (e.g., lipid or glucose levels) or “soft” CHD outcomes (e.g., angina, heart failure), studies in which methotrexate was administered as part of a combination therapy, and studies in which users were very different from non-users (e.g., diseased vs. non-diseased participants). We also excluded a priori ecological or cross-sectional studies; commentaries, general reviews, or case reports; duplicate publications from the same study; and studies reporting only crude risk estimates.
Of 694 identified articles, 621 were excluded based on review of the title and abstract (). The remaining 73 manuscripts were reviewed in detail independently and in duplicate to determine inclusion/exclusion (95% concordant); differences were resolved by consensus or, if necessary, group consultation among all investigators. Sixty-three studies were excluded because they were reviews (n=6), commentaries (n=1), letters (n=1), cross-sectional (n=2), or in vitro
(n=1); included methotrexate users which were very different from non users (n=1); assessed exposure other than methotrexate (n=19); did not have an appropriate control group (n=3); did not assess “hard” CVD endpoints (n=27); or reported only crude risk estimates (n=2) (see Supplemental Methods
). In the end, 10 studies were included in this meta-analysis.
Screening and selection process of studies of methotrexate use and cardiovascular disease risk.
For each study, data were extracted independently and in duplicate by 2 investigators, including year study was published and performed, study location, study design, sample size and number of events, inclusion and exclusion criteria, duration of follow-up, underlying disease (e.g., RA), duration of underlying disease, assessment of underlying disease severity, methotrexate comparison groups (e.g., initiators vs. non-initiators, ever users vs. never users), ascertainment of methotrexate use, treatment dose, disease outcome, disease incidence vs. recurrence, folate use, whether the reported analysis was primary or secondary and prespecified or posthoc in each paper, covariates adjusted for in the analysis, and adjusted risk estimates and confidence intervals (CIs). Accepted standardized quality scores are not available for observational studies. We performed quality assessment as previously described and used 7
, by evaluating and scoring 6 design criteria on an integer scale (0 or 1, with 1 being better), including review of study design and inclusion and exclusion criteria, assessment of exposure, assessment of outcome, control of confounding, assessment of underlying disease severity, and evidence of bias. These scores were summed, and quality scores from 0 to 3 were considered lower quality, and scores from 4 to 6 higher quality. Differences in data extracted and quality assessment were very unusual, and if present, were resolved by group discussion and consensus. Missing data were obtained by direct author contact for only 1 of 10 studies, as described above.
All included studies were observational and reported rate ratios or odds ratios; because of the low incidence of CVD in all studies, we collectively refer to these measures using the general term “relative risk” (RR). Between-study heterogeneity of RRs was assessed using the Dersimonian and Laird Q-statistic, the I2
statistic, and meta-regression 8, 9
. To calculate the overall pooled RR we used both a fixed effects and a random effects meta-analysis using the methods of Dersimonian and Laird 8
, and reported the former if the point estimates were virtually equal. In one study 10
, methotrexate was considered as the reference group to estimate the RRs of other RA medication on risk of MI or stroke hospitalization, including biologics, other cytotoxic agents, noncytotoxic agents, and glucocorticoids. Subsequently, to estimate the RR of methotrexate vs. other RA medication (the latter as the reference group) on CVD risk we pooled the inverse RRs of all other RA medication. Potential for publication bias was explored by visually inspecting a funnel plot of the effect size versus SE 11
, and statistically using the Begg adjusted-rank correlation test 12
. We explored potential prespecified sources of heterogeneity using stratified inverse-variance weighted fixed and random effects meta-analysis (and reported the latter if significant between-study heterogeneity was present, p < 0.1) and inverse-variance weighted meta-regression, including study design (prospective vs. retrospective cohorts), study location (America vs. Europe), years of follow-up, methotrexate comparison groups (initiators vs. non-initiators; ever vs. never users; current vs. non-current users), ascertainment of methotrexate use (physician vs. self-reported), underlying disease (RA; psoriasis; polyarthritis), disease outcome (CVD, MI, stroke), event (incident vs. recurrent), degree of covariate adjustment (most studies adjusted for sociodemographics and CVD risk factors, so further adjusting for underlying disease severity, other medication for underlying disease, or folate use), overall quality score (0-3 vs. 4-6), and whether the reported analysis was primary or secondary and prespecified or posthoc (to address potential publication bias of “positive” findings) in each publication. Analyses were performed using STATA 10.0 (StataCorp, College Station, Tex), with 2-tailed alpha<0.05.