Risk of Adverse Cardiovascular Events With Proton Pump Inhibitors Independent of Clopidogrel

Systematic Review With Meta-analysis

Riley Batchelor; Radya Kumar; Julia F. M. Gilmartin-Thomas; Ingrid Hopper; William Kemp; Danny Liew


Aliment Pharmacol Ther. 2018;48(8):780-796. 

In This Article


Search Results

The systematic search strategy identified 5285 unique articles, of which 89 were eligible for full text screening. In total, 22 studies met the inclusion criteria (including three abstracts which were excluded from meta-analysis).[24–26,35–53] The most common reasons for study exclusion were noncardiovascular outcomes (encapsulated by the broad search strategy), and the use of concomitant anti-platelet therapy. The screening process and further reasons for exclusion are documented using the PRISMA flow diagram presented in Figure 1.[27] Data from 16[24–26,36–40,42,44–48,50,52] of the 22 studies were included in the meta-analyses, which were undertaken separately for randomised controlled trials and observational studies. Three abstracts encapsulated by the search strategy were excluded from meta-analyses as risk of bias could not be accurately assessed.[35,41,49] Three full text studies were excluded as they did not have comparable methodology or quantitative data that could be accurately pooled.[43,51,53] These three excluded studies comprised a self-matched case series,[43] a case-crossover study,[51] and a cohort study.[53]

Figure 1.

PRISMA flow diagram with reasons for study exclusion

Study and Sample Characteristics

Eight of 16 studies included for the meta-analysis were clinical trials that compared PPI against placebo and recorded adverse cardiovascular events.[36,39,40,42,44,45,47,50] These involved a total of 4124 patients, with an average age of 47.5 years and 60.2% of whom were female. Although continuous anti-platelet therapy was not specifically excluded in any of the eight randomised studies, alternative exclusion criteria (ie, no history of cardiovascular disease, no anticoagulant use, no NSAID or aspirin use) yielded a study population at very low risk of concomitant clopidogrel use.

Five of the eight observational studies included for meta-analysis were cohort studies.[24,26,37,46,48] The remaining three studies employed various means of retrospective analysis, including a data mining analysis,[25] a case-control study,[52] and a post hoc analysis of a randomised controlled trial which randomised patients to clopidogrel and not to PPIs.[38] These eight studies pooled data from a total of 443 284 participants, with an average age of 54.3 years and 43.6% of whom were female. All observational studies used prescription or pharmacy dispensing record databases to identify PPI exposure, with the only exception being the single post hoc analysis of a randomised controlled trial.

Specific PPIs examined across all studies included omeprazole (n = 8 studies), esomeprazole (n = 8), lansoprazole (n = 8), pantoprazole (n = 7), rabeprazole (n = 6) and dexlansoprazole (n = 4).

The definition of major adverse cardiac event varied. Combinations of acute coronary syndrome, ischaemic stroke, cardiovascular hospitalisation, stent thrombosis and cardiovascular death were used to define the broad outcome. Eight studies reported data on fatal and nonfatal acute myocardial infarction (AMI) specifically.[25,26,37,39,40,42,47,52] The same seven studies that used prescription or pharmacy dispensing record databases to determine exposure used diagnostic codes to identify cases of major adverse cardiovascular events. The randomised controlled trials relied on investigator assessment and spontaneous reporting to ascertain adverse events. This process was not formally adjudicated.

The randomised controlled trials examined PPI use over a duration ranging from four weeks (n = 4) to 26 weeks (n = 4). A single propensity score-based cohort study reported cardiovascular outcomes after 120 days,[26] four studies reported outcomes at one year,[37,38,46,48] and the remaining three reported median follow-up times of between 521 days and 42.6 months.[24,25,52] (Table 1, Supplement 1, Supplement 4 and Supplement 5)

Risk of Bias Assessment

The 16 studies included for meta-analyses demonstrated good overall methodological quality. Shah et al's[25] data mining study that reported a positive association between PPI use and AMI was assessed as a cohort study for the purposes of assessing risk of bias. Dunn et al's[38] post hoc analysis of a randomised controlled trial was considered a randomised trial for the purposes of assessing risk of bias.

Overall, randomised controlled trials were assessed as being of overall low risk of bias as per the Cochrane Tool for Assessing Risk of Bias.[31] Dunn et al's post hoc analysis was regarded as high risk in terms of random sequence generation and blinding given that treatments were not randomised. This study was otherwise judged as low risk across the other domains.[38] All eight randomised trials were of low or unclear risk in random sequence generation and concealment of allocation, and all presented satisfactory blinding methods. One randomised study demonstrated a higher rate of discontinuation in the placebo group and was thus judged high risk of incomplete outcome data.[42] Six randomised trials had registered a protocol with a National Clinical Trial identifier and were identified as being of low risk in selective reporting.[38–40,42,44,45]

Generally, observational studies were conducted appropriately and included clear descriptions of population selection, exposure and outcome ascertainment. All observational studies used prescription and pharmacy dispensing record databases to ascertain exposure, and ICD–9 codes[54] to determine outcomes. Inconsistencies were likely to have existed between medication prescribing and actual patient adherence, in addition to misclassification of cardiovascular outcomes. A general trend observed across observational studies was that patients using PPIs were older, suffered more medical comorbidities, and used a greater number of other medications, compared to patients who were not prescribed PPI therapy. Accordingly, all included observational studies undertook adjusted analyses that accounted for key confounders: age, sex, diabetes status, hypertension and previous percutaneous intervention. However, there were significant differences between studies regarding adjustments made for medication use, smoking status, obesity and other cardiovascular risk factors. Only two observational studies published data with consideration given to the diagnosis of gastroesophageal reflux disease.[25,46] Substantial unmeasured confounding would not have been mitigated using covariate adjustments and presents a limitation when interpreting reported associations.

Domains of risk of bias for the 16 studies included in the meta-analyses are included in Supporting Information Data S3.

Meta-analysis of Risk of Major Adverse Cardiovascular Events With PPI Monotherapy

The meta-analysis of eight randomised controlled trials included 2554 patients who received PPI and 1570 who received placebo. There was little statistical heterogeneity across the studies, and random-effect modelling identified no increase in risk across these studies (RR 0.89; 95% CI 0.34–2.33, I2 = 0%, P = 0.85) (Figure 2).[36,39,40,42,44,45,47,50]

Figure 2.

Meta-analysis of randomised controlled trials reporting major adverse cardiovascular events (MACE) with PPI monotherapy vs placebo

Of the seven observational studies, Shah et al[25] presented the results from three different datasets, of which only the STRIDE dataset was included in the present meta-analysis. Similarly, Shih et al[26] presented results of both a propensity score-matched study and case-crossover study, of which only the results of the propensity score-matched study were included due to data overlap. Of the eight studies included for the meta-analysis, including the post hoc analysis of data from a trial, five reported adjusted HRs[26,37,38,46,48] and three reported adjusted ORs.[24,25,52] Seven of the eight studies demonstrated a statistically significant association.[25,26,37,38,46,48,52] Data adjusted for statistical imbalances were included preferentially over crude measures of association with the use of generic inverse variance modelling. No unadjusted effect estimates were used in the meta-analysis. Meta-analysis generated with random-effect modelling generated an overall risk ratio for major adverse cardiovascular events with PPI use of 1.25 (95% CI 1.11–1.42, P < 0.001). However, an I 2 statistic of 81% indicated substantial heterogeneity across observational studies (Figure 3).

Figure 3.

Meta-analysis of observational studies reporting major adverse cardiovascular events (MACE) with PPI monotherapy

Sensitivity Analyses

A sensitivity analysis conducted on the meta-analysis of randomised controlled trials demonstrated that no single study influenced the result significantly. Analyses stratified by duration of exposure to PPIs found point estimates on either side of the null, but neither result was statistically significant. Among the four studies which included only four weeks of exposure to PPIs, the RR was 0.66 (95% CI 0.20–2.24, P = 0.62, I2 0%),[39,40,45,47] while among the four studies in which duration of exposure to PPIs was 26 weeks, the RR was 1.44 (95% CI 0.30–6.98, P = 0.65, I2 = 0%).[36,42,44,50]

Within the pool of data from observational studies, the association between PPI therapy and major adverse cardiovascular outcomes was accentuated, and heterogeneity significantly reduced, when four studies measuring cardiovascular outcomes at a one-year endpoint were pooled (overall risk ratio 1.35; 95% CI 1.18–1.53, P = 0.19; I2 = 38%).[37,38,46,48] Because the overall meta-analysis pooled studies reporting different measures of association, sensitivity analysis pooling studies reporting HR only was also conducted (HR 1.37; 95% CI 1.24–1.52, P = 0.01, I2 = 20%).[26,37,38,46,48]

Meta-analysis of Risk of Myocardial Infarction With PPI Monotherapy

Four randomised controlled trials, involving a total population of 2356 patients, recorded adverse events that were specifically myocardial infarction.[39,40,42,47] No significant association was found between AMI and PPI use (RR 1.19, 95% CI 0.25–5.73, P = 0.83, I 2 = 0%, Figure 4).

Figure 4.

Meta-analysis of randomised controlled trials reporting myocardial infarction (MI) with PPI monotherapy vs placebo

Four observational studies described the association between PPI therapy and AMI.[25,26,37,52] Pooled results from these studies noted a statistically significant increased risk of AMI with PPI therapy, but again, the studies were heterogenous (RR 1.21; 95% CI 1.09–1.34, P = 0.08, I 2 = 56%, Figure 5).

Figure 5.

Meta-analysis of observational studies reporting myocardial infarction (MI) with PPI monotherapy

Publication Bias

A funnel plot of randomised controlled trials was generated to assess publication bias (Supporting Information Data S5). In the setting of significant heterogeneity (I2 statistic >50%), asymmetry testing for publication bias by means of funnel plot was not conducted for the observational studies.[55] In lieu of this, efforts were made to limit publication bias by including abstracts presenting eligible data for discussion within this systematic review, despite exclusion from meta-analysis.

Studies not Included in Meta-analysis

Six of the 14 studies included in the systematic review were not suitable for meta-analysis. Three of these studies were available only as abstracts.[35,41,49]

Juurlink et al's[43] self-matched case series identified 5550 hospital admissions for AMI within 12 weeks of commencing PPI therapy. The risk of AMI within the "risk interval" (weeks 0–4) against the "control interval" (weeks 8–12) increased (OR 1.8; 95% CI 1.7–1.9), suggesting an increased short-term risk of AMI with PPI therapy.[43] However, similarly raised ORs were noted for medications not known to be cardiotoxic, suggestive of significant confounding and no causal relationship.[43] Turkiewicz et al's[51] case-crossover study compared the risk of AMI with PPI use in a control period with PPI use in a hazard period (0–3 days prior to recorded AMI). An increased risk of AMI was identified with PPI prescription in patients with no previous history of AMI (OR 1.66; 95% CI 1.00–2.76), but a decrease in risk was found for the same population that had dispensed PPIs (OR 1.29; 95% CI 0.92–1.83).[51] The studies conducted by Juurlink et al[43] and Turkiewicz et al[51] were excluded from meta-analysis as both compared short-term cardiovascular risk of PPIs relative to long-term risk of PPIs, without including a cohort of patients who were not using PPI therapy. Wang et al[53] conducted a cohort study comparing concomitant PPI and clopidogrel therapy with PPI monotherapy, clopidogrel monotherapy and no therapy. In a cohort of patients with known cardiovascular disease, PPI monotherapy was associated with a reduced risk (HR 1.11; 95% CI 0.75–1.65, P–value not published) relative to no PPI use (HR 1.54; 95% CI 1.05–2.24, P–value not published), when concomitant clopidogrel and PPI use was set at a reference of HR 1.00.[53]

Of the three studies that were available in abstract form only, Ali's[35] pharmacovigilance analysis reported adverse events on behalf of Eli Lilly and Company and concluded that PPI use was not associated with ischaemic heart disease (Empirical Bayes Geometric Mean [EBGM] 0.74; EB05–EB95 0.72–0.77, where drug associations were considered significant if EBGM > 1, EBGM is a measure of association used in pharmacovigilance studies).[56] Hall et al[41] conducted a retrospective cohort study from a healthcare database that reported an increased risk of cardiovascular death and myocardial infarction with omeprazole, lansoprazole and pantoprazole use (HR 1.38; 95% CI 1.25–1.52). Syed et al's[49] large retrospective cohort study examining patients with gastro-oesophageal reflux disease found an increased risk of major adverse cardiovascular events when PPIs were used (OR 1.60 95% CI 1.58–1.61, P < 0.01). However, this abstract also presented results demonstrating that gastro-oesophageal reflux disease in the absence of PPI therapy was associated with an increased cardiovascular risk (OR 2.22; 95% CI 2.21–2.24, P < 0.01).[49]