Meta-analysis: The Effects of Proton Pump Inhibitors on Cardiovascular Events and Mortality in Patients Receiving Clopidogrel

C. S. Kwok; Y. K. Loke


Aliment Pharmacol Ther. 2010;31(8):810-823. 

In This Article


Eligibility Criteria

We selected randomized-controlled trials (RCTs) and controlled observational studies that reported on CV adverse events and death in patients receiving clopidogrel, with and without concomitant PPI exposure.

The specific inclusion criteria for RCTs were (i) parallel group randomized trial of any PPI for at least 30 days; (ii) placebo control arm; (iii) participants were all receiving clopidogrel; and (iv) clear reporting of CV outcomes and mortality.

For the observational studies, we selected case-control or controlled cohort (prospective or retrospective) studies that evaluated the association of CV outcomes/mortality with concomitant clopidogrel and any PPI exposure, as opposed to clopidogrel alone.

The clinical outcomes of interest were myocardial infarction/acute coronary syndrome, all cause-mortality and a composite measure of major adverse cardiovascular event (MACE) consisting of adverse events such as death or myocardial infarction or revascularization.

Search Strategy

We searched MEDLINE, EMBASE and Cochrane Central Register of Controlled Trials with no date limitations in October 2009 using the broad search terms [(proton-pump-inhibitor and clopidogrel)] without any language restriction. Additionally, we signed up with PubMed to receive automated electronic notification for any new articles containing the above search terms. To identify unpublished studies, we checked conference abstracts from the European Society of Cardiology and the American Heart Association in August and November 2009 respectively). We also checked the bibliographies of included trials and recent review articles for relevant studies.

Study Selection and Data Extraction

Two reviewers (CSK and YKL) scanned all titles and abstracts for studies that met the inclusion criteria, and excluded any articles that clearly did not fulfil the selection criteria. Full reports (where available) of potentially relevant trials and studies were retrieved and independently checked by two reviewers. We then independently collected information on study design, drug exposure, study location, characteristics of participants and relevant outcomes onto a spreadsheet. Where there was any uncertainty or discrepancies, the article was discussed between the two reviewers to determine if the studies should be included. We also contacted authors if there were any areas that required clarification.

Assessment of Risk of Bias

In accordance with the recommendations of the Cochrane Adverse Effects Methods Group, we looked at participant selection, follow-up, ascertainment of exposure, and definition and monitoring of adverse outcomes.[8]

We aimed to reduce the possibility of publication bias through searches of conference abstracts and contacting authors for any additional unreported data.

Data Analysis

We used RevMan 5.022 (Nordic Cochrane Centre, København, Denmark) to estimate pooled risk ratio (RRs) based on random effects model meta-analysis (inverse variance method). We assumed similarity between the RR and odds ratio (OR) because CV events and deaths were uncommon events.[9]

Where possible, we chose to pool adjusted RRs from the primary studies; otherwise, we used raw outcome data to yield unadjusted RRs (which may be particularly susceptible to confounding).

In view of the potential diversity of study designs, we chose to stratify the analysis on the basis of three groupings:

  1. Nonrandomized studies where we could only evaluate the unadjusted RRs for the relevant outcomes, with no correction for baseline differences or confounding.

  2. Nonrandomized studies that presented RRs adjusted for potential confounders.

  3. Studies that involved participants who had been part of a randomized trial, or studies where a propensity scoring system had been used to match patient groups. The use of a propensity score has been shown to yield estimates that are closer to the true marginal effect, which would have been found if a randomized study was possible.[10]

Statistical Heterogeneity

Statistical heterogeneity was assessed using I2 statistic, with. I2 values of 30–60% representing a moderate level of heterogeneity.[11]