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


While the overall pooled estimates suggest that concomitant clopidogrel and PPI use may be associated with adverse CV events and myocardial infarction, the available evidence shows no effect on mortality. Indeed, even when considering adverse CV outcomes alone, the presence of significant heterogeneity indicates that the evidence is at best, inconsistent, and at worst, potentially biased or confounded, as illustrated by the relative absence of any harm in the studies that were either randomized or based on propensity score-matched participants. For mortality and myocardial infarction, the pooled effect size (for harm) steadily diminished as we moved from RRs from crude raw data to observational studies that adjusted for confounders and then to participants from randomized trials or propensity-matched studies.

There are a number of potential explanations for our findings. One possibility is that the studies with the greater risk of bias or confounding tend to yield larger effect sizes for harm, whereas potentially higher quality studies (such as those with prespecified blinded outcome ascertainment, or more uniform participant selection/matching) have shown lesser evidence of harm. The gradations in magnitude of risk, which may depend on risk of bias, are apparent in the subcategories of study design within the Forest plots (Figures 2–4). Alternatively, the heterogeneity may arise from the populations studied, perhaps with varying distribution of CYP2C19 genotypes that confer different susceptibilities to the interaction. Given the considerable variations in type of PPI exposure among the studies, there is also the possibility that the extent of interaction may vary between individual PPIs and genotype of participants in particular populations, thus leading to heterogeneous results. Studies that used propensity matching, or recruited participants somewhat more uniformly based on tight inclusion criteria from RCTs may have been less susceptible to such confounding.[35]

The results of the meta-analysis may reflect differences in patient selection amongst the studies. Patients after percutaneous coronary intervention are a specific high-risk group compared with those identified in population-based studies using dispensing databases, where patients may be on long-term clopidogrel therapy for other reasons or have different CV risk factors. While the only prospective randomized trial did not show any evidence of a harmful interaction,[13] this does not exclude the possibility that certain patients in real-life clinical practice could be susceptible to such harm, as detected by some database studies.

Equally, statistical adjustment for confounding can only account for known confounders. We noted clear baseline differences in the studies between patients exposed to PPIs as compared with those not receiving PPIs and there may be other unknown confounders for which adjustment cannot be made. We are aware of two studies that carried out separate analysis of PPI exposure in patients not taking clopidogrel[15,18] and this showed significantly elevated risk with PPI (adjusted RRs of 1.55 and 1.38) compared with no PPI, even in the absence of clopidogrel therapy. This suggests either that there is residual confounding despite adjustment for known covariates or that PPIs have a deleterious effect irrespective of clopidogrel exposure.

However, outcome ascertainment in certain study settings may be more reliable, particularly regarding patients enrolled in, or selected post hoc from randomized trials who were followed up according to prespecified protocols, with adjudication of adverse events by independent safety monitoring boards. One large observational registry that used rigorous outcome ascertainment (contacting patients and their physicians, checking with registry offices), as well as blinded independent assessment of outcomes, also failed to show any significant clopidogrel–PPI interaction on death or major adverse events.[27,28] The wide variety of other observational database or registry studies may have relied on different methods of ascertaining CV events that may have been coded differently and we do not have sufficient information to judge the validity of such outcome data.

There are several limitations in this review. The quality of the studies is generally poor as many of the studies had some risk of bias, and we were only able to identify one prospective randomized trial. There was also significant heterogeneity of the pooled studies and this was particularly apparent when comparing results from different designs. In view of this heterogeneity, we have chosen not to focus on the pooled RR from the meta-analysis. As many of the studies were available only in abstract form, we did not have as much detail on the methods, actual drug exposure and results, as compared with a fully peer-reviewed journal publication. Moreover, definitions of MACE were sometimes vague and varied considerably between studies, thus contributing to weaknesses in assessing this outcome in contrast to the more robust measure of overall mortality. When sufficient details on PPI exposure and exact CV outcomes become available in future, it may then be helpful to evaluate the risk with individual PPIs and correlate this with data on bioavailability and platelet aggregation. Publication bias was not formally assessed with a funnel plot as asymmetry testing should only be carried out when there is no significant heterogeneity.[36] While we took other steps to minimize publication bias by searching through conference abstracts and trial registers for unpublished studies, we recognize that statistically significant findings are presented or published earlier, whereas findings of no effect gradually emerge later,[37] which is why we chose to evaluate abstracts at the most recent cardiology conferences about 12 months after the initial significant results had been presented.

We believe that the ideal trial would randomize patients on clopidogrel to different PPI, placebo or H2RA and evaluate the risk of MI, death and other prespecified adverse events such as gastrointestinal bleeding. While awaiting further reports from the COGENT trial[13] and any further new trials, we believe that robust pharmacokinetic and pharmacodynamic studies should be carried out on CYP2C10 genotyped patients to check the biological basis of the interaction with different PPIs and different intervals of drug administration.