Cardiovascular Outcomes and All-Cause Mortality

In 2008, Pezalla and colleagues reported the first study on clinical outcomes, a retrospective cohort study, as a letter. They found that patients on clopidogrel were significantly more likely to experience myocardial infarction (MI) within 1 year if they were on PPI treatment compared with those who were not. The second clinical study, a population-based nested case-control study, was published by Juurlink and colleagues in 2009. Patients on clopidogrel discharged after MI experienced more recurrent MIs (within 90 days or within 1 year) if they had claimed a prescription for a PPI within 30 days before the adverse event. The investigators performed extensive multivariate adjustment to control for baseline imbalances, although they admitted that their database did not capture data on some additional CV risk factors such as smoking status, blood pressure, and lipoproteins. Juurlink and colleagues also reported that pantoprazole was less likely than the other PPIs studied to be associated with CV events. However, this was based on the results of a secondary subgroup analysis in which pantoprazole was not associated with a statistically significant effect on CV outcomes, whereas omeprazole, lansoprazole, and rabeprazole, when grouped together, were. However, the appropriate method to evaluate for differences among subgroups is not to examine whether only 1 of the subgroups produced a statistically significant result but to assess whether the 2 subgroups were statistically different from each other. When analyzed that way, there is no statistically significant difference between pantoprazole and the other PPIs.

Given the high level of interest and potentially great clinical importance of this topic, several publications followed in rapid succession. Between early 2010 and March 2011, a further 33 studies with CV outcomes were published (35 in total). However, there has been broad disagreement among these studies regarding the results. Twenty found a statistically significant association of PPI use and worse CV outcomes in patients on clopidogrel. Of the other 15, 1 did not report the statistical significance and 14 found no evidence of an association. The magnitude of the association was reduced after adjusting for confounders in 12 out of 16 studies that performed such adjustments.

Seven systematic reviews published in 2010 have examined this association ( Table 1 ), and 5 have attempted a meta-analysis of the results. The systematic review with the most recent literature search (October 2010) found that “10 of 13 studies judged to be of low scientific quality reported a statistically positive interaction between clopidogrel and the general class of PPIs, and each concluded this was likely a clinically meaningful effect; none of the five studies judged to be of moderate or high quality reported a statistically significant association.”

Five other systematic reviews proceeded to meta-analysis despite the presence of heterogeneity; 3 have been published in full and 2 as abstracts (see Table 1 ). The number of included studies per meta-analysis ranged from 5 to 25 because of differences in literature search end dates and inclusion criteria. Overall, the most consistent finding of these meta-analyses was the highly statistically significant heterogeneity among trials for the outcomes of MI and major adverse cardiovascular events (MACE; a composite outcome measure whose exact definition has varied among studies but has always included all-cause death and MI). All 3 meta-analyses that assessed the outcome of MI agreed that PPI use was associated with statistically significantly more MI events among patients on clopidogrel. In the most recent meta-analysis, the pooled RR was 1.31 (95% CI 1.12–1.53). The association of PPIs with MACE has either approached or reached statistical significance; the pooled RR was 1.29 (95% CI 1.15–1.44) in the meta-analysis by Siller-Matula and colleagues.

However, the meta-analyses have yielded discrepant results regarding the association of PPIs with mortality. Three found no evidence of an association, whereas that of Hulot and colleagues showed statistically significantly worse outcomes for patients on PPIs. Given the position of mortality on top of the hierarchy of outcomes according to importance to patients, it is crucial to determine whether or not an association of PPIs with mortality truly exists. This association seems unlikely, and the apparently significant association found by Hulot and colleagues may have resulted from a series of flaws. Specifically, this was the only meta-analysis that pooled the unadjusted data from 2×2 tables for each included study. This happened despite all included studies except 1 being observational (either cohort or case-control studies or post hoc analyses of RCTs) and, therefore, almost all had gone to great lengths to adjust their outcomes for as many measured confounders as possible. Across the individual observational studies, the adjusted estimate of adverse outcomes on PPI cotherapy tended to be lower than the unadjusted estimate. In our review of the 16 studies that reported both unadjusted and adjusted results, only 4 did not show a reduction in the magnitude of the association after adjusting for confounders. In the study by O’Donoghue and colleagues, the unadjusted OR was not statistically significant (0.83, 95% CI 0.61–1.41). However, the adjusted HR showed a statistically significant reduction in mortality with PPI cotherapy (0.68, 95% CI 0.47–0.96). If the adjusted estimates had been used in the meta-analysis by Hulot and colleagues, the pooled result for mortality would not have been statistically significant. The other 3 meta-analyses that assessed mortality had appropriately used the adjusted estimates (with 95% CI) as provided by each individual study and pooled these according to the inverse variance method. Furthermore, even if the use of unadjusted estimates is accepted, there are several flaws in the mortality forest plot in this meta-analysis, all biasing toward a statistically significant increase in mortality related to PPIs. Most notably, although the investigators of this meta-analysis reported pooled results for all-cause mortality (and mentioned that results for CV mortality were similar), in their parallel meta-analysis regarding the effect of CYP2C19∗2 genotype they only reported “mortality” regarding the effect of PPIs. A close look at the forest plot reveals that (unadjusted) data for CV mortality were used for the study by O’Donoghue and colleagues, although (unadjusted) data on all-cause mortality were also available in the publication. It is potentially misleading to report disease-specific mortality (which is prone to classification bias) in isolation, without placing it in context with all-cause mortality. In summary, our appraisal of the published meta-analyses is that there is no evidence of an association of PPI coadministration with all-cause mortality in patients taking clopidogrel. Currently, the best available estimate of the association is the result of the most current meta-analysis by Siller-Matula and colleagues, which is an RR of 1.04 with a CI that is narrow enough to rule out a clinically meaningful association (0.93–1.16).

Although subgroup analyses should be regarded as hypothesis-generating rather than confirmatory, there was no statistically significant association between PPI use and CV outcomes among clopidogrel users in the following subgroups : (1) among RCTs or post hoc analyses of RCT data (vs observational studies) for the individual outcomes of MACE, MI, or death ; (2) among low-risk patients, defined as annual rate of MACE less than 10% in the control group (vs high-risk patients) for MACE ; (3) when nondefinite events, such as rehospitalization for cardiac symptoms or revascularization, were excluded (vs being included) regarding MACE. The differences between each subgroup pair were highly statistically significant.

Two of the meta-analyses assessed the association of individual PPIs with MACE, but reached different results. Hulot and colleagues found that, among 4 studies that provided separate outcomes for omeprazole users, omeprazole was significantly associated with a higher risk for MACE (OR 1.37, 95% CI 1.27–1.47). Silller-Matula and colleagues performed subgroup analyses for omeprazole use (5 studies) and pantoprazole use (6 studies); both analyses yielded nonsignificant results. The latter meta-analysis had the more recent literature search (April 2010); only 1 of the new studies that were published since then presented separate analyses by type of PPI and found no differences among them.

Furthermore, the systematic review by de Aquino Lima and Brophy concluded that some of the observational studies had shown an absolute increase in CV outcomes greater than 15%, which is significantly larger than the expected protective effect of clopidogrel treatment. Even if PPIs completely abolished the protective effect of clopidogrel, the consequences would have been less severe. Moreover, in some observational studies, the increased rate of CV events was observed after the 3-month window when the largest proportion of the benefit of clopidogrel is known to occur.

The timing of PPI exposure in relation to the adverse CV events in these studies is of particular interest. Most studies did not assess day-to-day use of PPI for the whole duration of the follow-up period. Instead, most studies either classified patients as users or nonusers of PPIs according to the discharge prescriptions at the start of the follow-up period, or did not specify their criteria for PPI use. Studies that took into account data on day-to-day use of PPI during the follow-up period have not found an association between PPI use and CV events in patients on clopidogrel. Misindication-protopathic bias may have contributed to the positive association found in studies that defined as PPI users those who had received these medications too close to the CV event (eg, 1–3 days in the study by Valkhoff and colleagues ; <7 days in that by van Boxel and colleagues ). Protopathic bias occurs when a pharmaceutical agent is inadvertently prescribed for an early manifestation of a disease that has not yet been diagnostically detected. For example, the rescue use of nitroglycerin within 30 days before a major CV event is significantly associated with increased risk of major CV events, although it would be unreasonable to conclude that the nitroglycerin had been responsible. Furthermore, patients with increasing angina preceding a major CV event may be more likely to be started on PPIs for misdiagnosed GERD. In addition, the timing of the occurrence of coronary thrombosis contradicts current evidence on the natural history of coronary heart disease. In the population at the highest risk for coronary thrombosis (ie, patients with recent implantation of drug-eluting stents), stopping clopidogrel (but continuing aspirin treatment) leads to stent thrombosis in a median time of 13.5 days if less than 6 months have elapsed since stent implantation, and 90 days if more than 6 months had elapsed. Because a partial reduction of the antiplatelet action of clopidogrel caused by PPIs cannot be more catastrophic than complete cessation of clopidogrel treatment, the current biologic rationale for the increased CV events in patients on PPIs becomes less plausible for studies that assessed PPI exposure too close to the CV events (eg, within 3 or 7 days).

Several observational studies have appeared since the publication of the recent meta-analyses, and the results remain contradictory. A major limitation of observational studies is that associations cannot prove causality. Adjusting for confounding in observational studies can never be proved adequate, because important unmeasured or unknown confounding factors that are not balanced among the study groups may be the true explanation for the results. Channeling bias is particularly likely to be present in the observational studies that addressed the PPI-clopidogrel interaction. This has been defined as the “tendency of clinicians to prescribe treatment based on a patient’s prognosis; as a result of the behavior, in observational studies, treated patients are more or less likely to be high-risk patients than untreated patients, leading to biased estimate of treatment effect.” In general, among the observational studies, patients on clopidogrel who were coprescribed a PPI almost invariably had higher comorbidity than those who were not given a PPI. As expected, patients were more likely to be on PPI treatment if they had risk factors for GI complications. Some of these factors (notably advanced age and use of NSAIDs) are also significantly associated with increased risk of cardiovascular events. What was less expected was that patients were also more likely to be given a PPI in response to baseline comorbid conditions apparently unrelated to the risk of GI complications but closely related to increased risk of CV complications and all-cause mortality, such as diabetes mellitus, cancer, chronic obstructive pulmonary disease (COPD), liver disease, renal failure, congestive heart failure, and cerebrovascular disease. Clinicians may prescribe PPI prophylaxis not only according to the risk of upper GI bleeding (as suggested by guidelines), but also according to the perceived risk of death in case of upper GI bleeding. Some of the risk factors for these 2 outcomes are shared, but are not identical. For example, the estimated risk of upper GI bleeding may be similar for two 70-year-old male patients on clopidogrel following an ACS, one with no comorbidity, and the other with concomitant severe COPD, moderate renal failure, and an overall fragile health status as per physician’s global impression. However, the latter patient is less likely to survive an upper GI bleed than the former and may, therefore, be more likely to receive PPI prophylaxis from his physician. Regardless of whether PPI can reduce the risk of upper GI bleeding, the latter patient is more likely to die from his additional comorbidities that the former patient. An observational study could interpret this as an association between PPI use and increased mortality.

Two other observations support the presence of channeling bias among the observational studies included in this review. First, each of the 4 studies that additionally assessed patients who were not prescribed clopidogrel found that the association between PPI use and CV outcomes was similar regardless of whether or not clopidogrel was used. One retrospective observational study on patients who had MI found no association of PPI use and CV outcomes in either of the 2 cohorts. Two other retrospective observational studies in patients following MI and 1 post hoc analysis of RCT data on patients who had coronary ischemia at the time of randomization and were undergoing (85%) or were at high likelihood of undergoing percutaneous coronary intervention found that PPI treatment was significantly associated with increased risk of CV events among both users and nonusers of clopidogrel. This finding indicates that the observed association between PPI use and CV outcomes cannot be explained by a drug-drug interaction between the PPI and clopidogrel and is more likely to be caused by residual confounding from channeling bias. A fifth study seems to contradict the concordance in findings among users and nonusers of clopidogrel: Ho and colleagues conducted a retrospective observational study among patients with ACS and reported that PPI use increased CV events among clopidogrel users but not among nonusers of clopidogrel. However, their analysis on nonusers of clopidogrel is less convincing because they did not truly assess a cohort of patients who were not prescribed clopidogrel; instead, they assessed individual 90-day time blocks during which patients who were prescribed clopidogrel did not refill their prescription. Second, 3 of the studies found that PPI treatment was significantly associated with increased risk for upper GI bleeding or PU complications. Again, the most likely explanation for this apparent paradox is channeling bias, rather than a causal association. This explanation is further supported by all 3 of these studies also finding PPI treatment to be significantly associated with increased risk for CV events.

Gastrointestinal Outcomes

As mentioned earlier, 16 studies have reported the association of PPI use with rates of GI bleeding (upper GI bleeding in most of them) and/or PU complications among patients on clopidogrel. Ten of these also reported CV outcomes and/or all-cause mortality. Two studies were RCTs ; the remaining 14 were observational. Again, the results are apparently conflicting: 7 studies did not find a significant association, 6 found a significant protective association and, as mentioned previously, 3 found a significant harmful association.

Three systematic reviews have addressed this issue. The 2 most recent, by Siller-Matulla and colleagues and by Kwok and colleagues (with search dates of April and March 2009, respectively) also included meta-analytical syntheses of the studies (see Table 1 ). Both meta-analyses found a statistically significant association of PPIs with reduced upper GI bleeding among patients on clopidogrel, with no heterogeneity among studies. The former meta-analysis (3 studies) found a protective RR of 0.50 (95% CI 0.37–0.69), whereas the latter (9 studies for upper GI bleeding) found a protective OR of 0.31 (95% CI 0.19–0.51).

Randomized Controlled Trials

It is worth elaborating further on the 2 available RCTs because they have the highest inherent validity among all studies. The results of COGENT (Clopidogrel and the Optimization of Gastrointestinal Events Trial) were fully published in 2010, although the main results had been presented in abstract form in 2009 and were, therefore, included in all of the meta-analyses mentioned earlier. COGENT was an international, randomized, double-blind, double-dummy, placebo-controlled phase 3 study of the efficacy and safety of a fixed-dose combination of once-daily clopidogrel 75 mg and omeprazole 20 mg versus clopidogrel 75 mg alone. The study population comprised 3873 patients for whom dual antiplatelet therapy was anticipated for at least 12 months; most patients had ACS or MI and/or had undergone PCI with stent implantation. All patients received low-dose aspirin. The study found that PPI treatment was associated with a statistically significant reduction in overt upper GI bleeding (HR 0.13; 95% CI 0.03–0.56; number needed to treat [NNT] 98), and had no demonstrable effect on the composite end point of CV death, nonfatal MI, coronary revascularization, or ischemic stroke (HR 0.99; 95% CI 0.68–1.44).

This study has adequate concealment of allocation and blinding, but the method of sequence generation was not reported and the follow-up was not complete. Although losses to follow-up were minimal as a proportion of the total study population (2.6%), the absolute number of lost patients (100) was large compared with the small number of events in the study (17 cases of overt upper GI bleeding and 109 CV events). Therefore, by worst case scenario, the lost cases had a potential to significantly change the ratio of events. Another interesting methodological aspect of this study is that it was terminated prematurely when the sponsor went bankrupt. The main consequence was not the smaller sample size (which was smaller than the revised sample size, but still larger than the a priori planned size), but the shorter follow-up (the planned follow-up was 1–2 years, whereas the median achieved follow-up was only 106 days), which resulted in less than the planned number of events and a wide 95% CI for the results. Notably, the upper bound of the CI for the result of the CV composite end point cannot rule out a relative increase of 43% (at worst case scenario) that is larger than the minimal important difference for most clinicians and patients. Nevertheless, despite its limitations, this study is currently the best available evidence.

The RCT by Hsu and colleagues was published in 2011 and therefore was not included in any of the available meta-analyses. One-hundred and sixty-five patients who had a history of PU (endoscopically confirmed to be healed at study entry) and who required antiplatelet therapy for coronary heart disease or stoke were randomized to receive daily treatment with either clopidogrel 75 mg at bedtime and esomeprazole 20 mg before breakfast or clopidogrel alone for 6 months. None received aspirin. PPI use resulted in a statistically significant reduction of PUs at endoscopy (scheduled at the end of the study or prompted by symptoms): 1.2% versus 11.0%, absolute difference 9.8%, 95% CI 2.6% to 17%. There was no demonstrable difference on PU bleeding (1 vs 1 event) or the composite end point of cardiovascular events, cerebrovascular events or death (4 vs 3 events), although the study was not powered to detect moderate differences in these outcomes. The study had adequate sequence generation and allocation concealment, but was not blinded. Furthermore, 5% of the patients were lost to follow-up and a further 19% did not undergo the end-of-study endoscopic evaluation; these numbers were large enough to have the potential to change the ratio of outcomes significantly.