|Home | About | Journals | Submit | Contact Us | Français|
Recent studies have raised concerns about reduced efficacy of clopidogrel when used concurrently with proton pump inhibitors (PPIs), but those studies may have overestimated the risk.
We studied the potential for increased risk of adverse cardiovascular events among users of clopidogrel with concurrent use of PPIs versus without, in three large cohorts of patients ≥ 65 years treated between 2001-2005. All patients had undergone percutaneous coronary intervention or been hospitalized for acute coronary syndrome in Pennsylvania, New Jersey, or British Columbia, and had subsequently initiated treatment with clopidogrel. We recorded myocardial infarction (MI) hospitalization, death, and revascularization among PPI users and non-users. We assessed our primary endpoint of MI or death using cohort-specific and pooled regression analyses.
18,565 clopidogrel users entered our analysis. On a pooled basis, 2.6% of those who also initiated a PPI versus 2.1% of PPI non-users had an MI hospitalization; 1.5% versus 0.9% died, and 3.4% versus 3.1% underwent revascularization. The propensity score-adjusted rate ratio for the primary endpoint of MI or death was 1.22 (95% confidence interval 0.99 to 1.51); for death 1.20 (0.84, 1.70); and for revascularization, 0.97 (0.79 to 1.21). Matched analyses generally yielded similar results.
Though point estimates indicated a slightly increased risk of MI or death in older patients initiating both clopidogrel and a PPI, we did not observe conclusive evidence of a clopidogrel/PPI interaction of major clinical relevance. Our data suggest that should this effect exist, is unlikely to exceed a 20% risk increase.
The role of clopidogrel following percutaneous coronary intervention (PCI) or acute coronary syndrome (ACS) is well established by large scale clinical trials, meta-analyses, and clinical practice guidelines.1-7 As a result, the use of clopidogrel has risen sharply since its initial marketing in 1997;8 it is now the second best-selling drug in the world.9 Given the speed at which the drug has been adopted, the full range of potential interactions between clopidogrel and other agents has yet to be fully explored.
A proton-pump inhibitor (PPI) is frequently co-prescribed with clopidogrel to reduce the risk of gastrointestinal bleeding. There has been little research evaluating the benefits of this strategy,10 and no specific guidelines are in place for the use of gastroprotective agents in patients receiving clopidogrel. Recently, driven by concerns that the antiplatelet effects of clopidogrel could be diminished by concurrent use of PPIs,11-13 questions have been raised about the safety of the drug combination.14 These concerns are attributed to the competitive inhibition by PPIs of one of the cytochrome P450 isoenzymes, CYP2C19, involved in the metabolic activation of clopidogrel. This effect is thought to be a mechanism of “clopidogrel resistance” in such patients.15-17
The clinical consequences of a potential clopidogrel-PPI interaction remain largely unclear but could be of considerable importance. Two studies have each shown an increased risk of acute cardiovascular events or death with this combination,18, 19 but may have overestimated the risk due to incomplete control of confounding and biased patient selection. We conducted a retrospective study in three large health care utilization databases with the goal of addressing study design issues and confounding. We identified cohorts of patients who had undergone PCI or survived ACS. Adjusting for confounding factors using both standard and newly-introduced approaches, we examined whether there was a higher rate of specific adverse outcomes in patients taking clopidogrel who were treated concurrently with PPI versus those given clopidogrel alone. The outcomes studied were myocardial infarction (MI) hospitalization, revascularization, and death.
We pooled data from patients enrolled in three large health insurance programs: (1) British Columbia's (BC) government-funded provincial health care system; (2) Pennsylvania's (PA) Pharmaceutical Assistance Contract for the Elderly, and (3) New Jersey's (NJ) Pharmaceutical Assistance to the Aged & Disabled. The BC database linked information from PharmaNet, which captures essentially all medications dispensed to all patients in the province,20 its Medical Services Plan and hospital discharge files, which together cover virtually all clinical encounters, and the province's vital statistics database. The PA and NJ databases were comprised of those states' drug benefit program data linked with complete claims data from Medicare Parts A and B, including hospital discharge information and all fee for service charges; and to vital status information from the Social Security Administration's Death Master File.
During the study period, the states of PA and NJ offered pharmacy assistance programs to low-income adults 65 years and older; these programs provided generous pharmaceutical benefits for a modest copayment. Owing to the low cost, unrestricted formularies, low turnover, and the fact that drug claims are recorded for purpose of payment to dispensing pharmacies, capture of out-of-hospital prescription drug use is relatively complete. BC's program is available to all residents of the province and PharmaNet records all out-of-hospital prescription drug dispensing independent of whether the drugs are paid for by the program or by the patient, or are dispensed in-person or by mail. Throughout the study period, BC had a prior authorization program in place for clopidogrel. Underreporting and misclassification is minimal because of the electronic data entry of all drug dispensings.20
All traceable personal identifiers were removed from the datasets prior to analysis in order to protect patient confidentiality. The Institutional Review Board of the Brigham and Women's Hospital approved this study. Data use agreements were in place.
We identified all patients who underwent PCI or were hospitalized for ACS in BC from January 1, 2001 to December 31, 2005; and in PA and NJ from January 1, 2001 to July 31, 2005. PCI was defined using recorded procedure codes. ACS was defined as a hospitalization with a primary discharge code for MI or unstable angina; we required that the hospitalization last at least three days and no more than 180 days. (See Appendix.) If patients had several events during this time, we included only the first as the index event. To minimize age heterogeneity among the three cohorts, we restricted BC patients to those ≥ 65 years.
Patients became eligible for this study the day after discharge for PCI or ACS, or the day after outpatient PCI (Figure 1). We limited our study cohorts to new users of clopidogrel: those who did not fill any prescriptions for clopidogrel in the 180 days before the index event and then both survived the first 7 days after the index event and initiated clopidogrel within those 7 days.21 To ensure uniform eligibility, all study subjects were required to have used at least one medical service and filled at least one prescription in each of the two six-month periods preceding the index event.
The exposure of interest was use of a PPI, defined as omeprazole, esomeprazole, lansoprazole, pantoprazole, or rabeprazole dispensed at any point 21 days before and/or within 7 days after the index event, and with sufficient supply of the drug to last through the 7 days after the index event. Follow-up began for all patients on the day following this 7-day exposure assessment run-in period. The run-in period equally allowed all patients to be on PPI therapy or be categorized as PPI non-users and the study is therefore not subject to immortal time bias.22 In the United States, omeprazole was available over-the-counter during some of the study period, but for the patients in our population, the price would have been higher than the co-payment in the state drug plans. In Canada, no PPI was available over-the-counter.
We measured the occurrence of three study endpoints over up to 180 days of follow-up: myocardial infarction hospitalization, revascularization, and death from any cause. A combined MI and death outcome was our primary endpoint. We defined MI as above and allowed for the code to appear in the secondary position.23 Revascularization was assessed as a PCI or coronary artery bypass graft (CABG) surgery at any time during follow-up. (See Appendix.) Death was assessed through vital statistics and government agencies.
With respect to patient characteristics, we recorded patients' age, sex, and calendar year of study entry. In PA and NJ, we recorded a binary variable for non-white race; race is not measured in BC. Over the 365 days prior to the index event, we assessed patients' prescription drug use, reported comorbidities, and intensity of medical service use. (See Appendix.) We computed a Charlson comorbidity score.24 We measured the length of stay during the care for the index event (1 or more days), and whether an MI occurred and/or a PCI was performed during that period.
We estimated the rate ratio comparing incidence of the study outcomes among clopidogrel users who also used PPIs versus those who used clopidogrel alone. Multivariate-adjusted rate ratios were estimated using Cox proportional hazards regression. In this “as-treated” analysis, patients were censored at the earliest of time of outcome event, end of 180 days, five days after stopping clopidogrel, or death. In addition, PPI-exposed patients were censored five days after stopping a PPI and PPI-unexposed patients were censored the day of starting a PPI. If an outcome occurred on the day of censoring, the outcome was assumed to take place before the censoring event. In all cases, drug starting and stopping were computed from pharmacy claims data using the date of fill and the reported days supplied. We analyzed the data separately within each of the three regional databases (BC, PA, NJ) and then, after pooling all cohorts, with the pooled regression models stratified by database. We computed 95% confidence intervals.
Since information content differed from database to database, we further applied a high-dimensional propensity score (hd-PS) to further improve confounding adjustment.25 The techique has been shown to provide equally valid or more valid estimates of treatment effects by adjusting for hundreds of confounders identified empirically from databases. In our study, the hd-PS algorithm examined all drugs, diagnosis codes, and procedure codes reported for a patient and identified the 400 codes that, if adjusted for, had the most potential to reduce bias. We entered these 400 codes into a propensity alongside the confounders that we identified a priori. The propensity score was ranked by deciles within each database and then entered as categorical variables into the Cox model.26
As a first sensitivity analysis, we matched the cohorts using 1:1 greedy matching on the hd-PS.27 Additionally, we performed an analysis in which we reduced the maximum duration of follow-up to 90 days. We then performed a “PPI initiator” analysis in which we excluded patients who had a current PPI prescription at the time of the index event or who had filled a PPI prescription in the 180 days prior. To assess whether individual PPIs differed with respect to an interaction with clopidogrel, we did agent-specific analyses among omeprazole and pantoprazole users. We also performed a cumulative risk, “intention-to-treat” analysis in which the PPI exposure was carried forward through a fixed 90 day follow-up period. Finally, we tested whether PPI users were at greater risk of death or MI irrespective of their use of clopidogrel. In this analysis, we allowed clopidogrel non-users to enter the cohorts, censored patients on the day a PPI non-user started PPI therapy or five days after a PPI user stopped the use of the drugs, and adjusted for clopidogrel use.
Of the 64,561 patients in our cohorts who underwent PCI or hospitalization for ACS between 2001 and 2005, 31% had any recorded use of clopidogrel in the 7 days after their hospital discharge, 40% in the 90 days after discharge, and 45% in the 180 days after discharge. 18,565 (29%) survived at least 7 days after the event or procedure, were prescribed clopidogrel during those 7 days, met the study inclusion criteria, and were therefore included in our analytic cohorts (Figure 2).
There were differences in the baseline characteristics among the three regional cohorts, although PA and NJ showed many similarities. Average age was 74 years in BC (range: 65-100) and 78 in each of PA and NJ (range: 65-104 in PA and 65-101 in NJ; Table 1). Due to the patients' age and the programs' income restrictions, in 75% of the patients in PA and 65% NJ were female versus 38% in BC. Prevalence of baseline risk factors also ranged widely: for example, history of GI bleeding and hemorrhagic strokes prior to the cohort-defining event were both more common in NJ and PA than in BC. Those who were users of PPIs generally had more co-morbidities than did their non-user counterparts.
The crude incidence rate of MI or death was 13.6% per 180 person-days in BC, 14.8% in PA, and 13.6% in NJ. Death occurred during follow-up in 0.6% of the BC cohort, 1.5% of the PA cohort, and 1.6% of the NJ cohort. Additional patients died during the 7-day run-in period (Figure 2). Across all databases and outcomes, the median length of follow-up was 29 to 30 days among both the PPI users and non-users (Table 2). 2% of patients were followed for the maximum 180 days.
Among our three cohorts of clopidogrel users, crude rates of MI or death were substantially higher for PPI users than PPI non-users. The pooled crude rate ratio (RR) was 1.74 (95% confidence interval, 1.44 to 2.01); with multivariate adjustment, the RR fell to 1.32 (1.08 to 1.61). With hd-PS adjustment, it fell again to 1.22 (0.99 to 1.51) (Table 3). For revascularization, results varied significantly among the cohorts (p-value for heterogeneity < 0.05), ranging from an RR of 0.70 (0.50 to 0.99) in NJ to 1.64 (1.15 to 2.35) in BC. Between the PPI user and non-user groups there was general balance of the propensity score and the measured covariates within the decile strata of the score.
Matching on hd-PS resulted in smaller (82% of exposed patients) but broadly balanced exposed and unexposed groups in each cohort; for example, PPI user and non-user ages were equivalent to the first decimal place. Overall, the results using the matched populations were similar to those based on the unmatched populations (Table 4). The matched RR for MI or death was 1.26 (0.97 to 1.63). The observed risk of MI or death decreased with more precise control of confounding.
The sensitivity analysis in which we limited follow-up to 90 days included 92% of the primary study population and yielded substantially similar results to the primary analysis; the pooled, hd-PS adjusted RR for MI or death was 1.23 (1.00 to 1.53). Similarly, when we included only PPI initiators, we observed this RR as 1.11 (0.81 to 1.52). The sensitivity analysis in which the exposure was carried forward through a fixed 90-day follow-up period yielded results closer to the null than did the primary analysis (Supplemental Table 1); the RR was 1.06 (0.91 to 1.23). Extending the run-in period from 7 to 15 days allowed more time for patients to become exposed, but starting follow-up on day 16 excluded events that occurred in the early days after the index event; the RR moved to 1.38 (1.12 to 1.69). The analyses in which we assessed the omeprazole and pantoprazole user groups were substantially similar both to each other and to the primary analysis: the RR in the omeprazole-only study was 1.17 (0.68 to 2.01) and 1.26 (0.93 to 1.71) in the pantoprazole-only study. Our study of risk from PPI use irrespective of clopidogrel use yielded results that were null or protective after adjustment. The pooled, hd-PS adjusted RR of MI or death associated with PPI use itself among the 61,605 patients in this analysis was 0.79 (0.73 to 0.85) and 1.02 (0.93 to 1.12) for MI.
Recent trial and non-experimental studies have reported that an interaction between PPIs and clopidogrel increases the risk for a second MI by up to 40%, which if substantiated, would be of major clinical and public health importance. In the study presented here, we applied traditional and more advanced confounding control techniques to an analysis of 18,565 North American patients in three cohorts. The patients were adults 65 years and older who had recently undergone PCI or an ACS hospitalization. To our knowledge, ours is the largest study of this question to date. While the point estimates we observed in each of the three cohorts indicated a slightly increased risk of MI or death in older patients initiating both clopidogrel and a PPI, the possible increase in risk was modest, the confidence intervals were wide, and we did not observe consistent evidence of a substantial or statistically significant clopidogrel-PPI interaction. Progressively better control of confounding reduced the apparent risk from the initial crude estimates, suggesting that some or all of the final observed risk may be the result of residual unmeasured confounding. Our data suggest that if a true interaction does exist, its effect is unlikely to exceed a 20% risk increase.
In both of the recently published studies evaluating the interaction between clopidogrel and PPIs,18, 19 patients who received PPIs had more comorbidity than those prescribed clopidogrel alone, leading to a concern about confounding by indication.28 For example, physicians may be more likely to use gastroprotective drugs in frailer patients -- such as those started on a PPI during an intensive care unit stay – or in smokers, who have a higher incidence of upper gastrointestinal disease. Conventional adjustment as employed by Ho et al. and Juurlink et al. may inadequately account for confounding in this treatment group. In our study, we performed more extensive confounding adjustment25 and in a sensitivity analysis, verified that after adjustment, the PPI users were not more likely to be at increased risk for MI or death. Further complicating the interpretation of Ho et al.'s results is their inclusion of “immortal person time” in the study's primary analysis.22 This design results in bias due to the fact that exposure status was determined during follow-up; those who were sickest and had an MI early in follow-up would have had a lower chance of being exposed such that consequently, unexposed patients would have been oversampled. Our study design avoided such biases only starting follow-up after the 7-day run-in period during which exposure status was determined for all patients.
The possibility of a modest risk posed by a PPI-clopidogrel interaction would be consistent with the findings of the randomized study by Gilard et al.,12 in which omeprazole partially reduced the inhibitory effect of clopidogrel, as assessed by tests of platelet activity. The correlation between the results of the assay used by these investigators and subacute coronary stent thrombosis has been previously observed.29 Separately, two other recent studies reported that patients with substantial loss-of-function mutations in the metabolic pathway that activates clopidogrel (CYP2C19)16, 17 have a doubling of risk of coronary events. CYP2C19 is also the putative mechanism by which clopidogrel and PPIs interact.
If PPIs do indeed reduce the beneficial effect of clopidogrel, then 20% may well be the upper bound for any increase in the relative risk of MI or death. In post-PCI or ACS patients using clopidogrel, the baseline risk of the outcome is high. If absolute risk increases are constant, relative risks are smaller in patients with high baseline risk.30 Further, the Gilard et al. study indicated only a partial reduction in clopidogrel's anti-platelet action; that is, even with a drug-drug interaction, clopidogrel would still provide some benefit. Finally, the CYP2C19 polymorphism, which is likely equally prevalent in the PPI user and non-user groups, itself reduces the protective effect of clopidogrel in a certain fraction of the population. This attenuation would yield a decrease in relative risk as compared to a population in which nobody carried the allele.
To the extent possible in our non-randomized study, we drew upon the design principles of an RCT: we limited the patient population to those who were clopidogrel-naïve, instituted a 7-day run-in period, and only started follow-up from the time that patients' exposure status had been fully determined. To minimize the effects of unmeasured confounding, in particular confounding due to the “channeling” to one treatment or another in response to disease severity and patient prognosis,28 we adjusted by a high-dimensional propensity score. Adjusting for or matching on a propensity score that includes all pertinent confounders should serve a role analogous to randomization in an RCT;26 the hd-PS method allowed us to enrich the standard propensity score with more of those pertinent confounders than are possible to specify a priori. The hd-PS we adjusted for included our investigator-identified covariates alongside 400 additional variables empirically identified from the claims codes in our databases, and therefore allowed for more confounding adjustment than did standard propensity score or multivariate analysis. One confounder that remained unmeasured was aspirin use, though in this population of patients who have all had cardiac events, aspirin would have been indicated for and likely used by almost all members of the cohort and would thus be a weak confounding factor.
In the United States, omeprazole has been available over-the-counter (OTC) since June 2003; as a result, our study may be open to some misclassification of exposure. In PA and NJ, however, the low-income patient population we studied would have little incentive to buy PPIs OTC, as their copayment for prescription PPIs was less than half of OTC prices.31 In Canada, PPIs are not available OTC -- though it is possible that BC residents purchased OTC PPIs in the United States -- yet the BC data reflected the same findings as those from NJ and PA. With respect to potential misclassification of the outcomes, the “hard” endpoint of MI or death was defined in a highly specific and non-differential way; such specificity should lead to a non-existent or negligible misclassification bias in the relative risk.32 Conversely, the discretion involved in choosing upon whom to perform a revascularization procedure likely explains at least some of the heterogeneity of the revascularization results: very sick patients are less likely to receive such treatment due to the overall risk-benefit assessment.33, 34
In conclusion, the results of this large study of community-dwelling patients 65 years and older followed up after PCI or ACS, failed to demonstrate a large or statistically significant increase in the relative risk of MI or death owing to concurrent use of clopidogrel and PPIs, as previously reported. Our data are however consistent with the possibility of an interaction with only modest clinical effect in routine care of older adults.
Association between exposure to clopidogrel with concomitant proton pump inhibitors versus clopidogrel alone after percutaneous coronary intervention or acute coronary syndrome. First exposure is carried forward over for a fixed period of 90 days of follow-up. Rate ratios and 95% confidence intervals were estimated from Cox proportional hazards regression with various methods of covariate adjustment.
Funding Sources: This study was funded from internal sources by the Division of Pharmacoepidemiology and Pharmacoeconomics. JAR is supported by a career development award from the Agency for Healthcare Research and Quality (AHRQ). SS is supported by grants from the NIH's National Institute of Aging and the National Institute of Mental Health and is principal investigator of the Brigham and Women's Hospital AHRQ-funded DEcIDE research center.
Disclosures: The authors have no conflicts of interest. SS had previously received an unrelated investigator-initiated research grant from Pfizer to study the safety of cox-2 inhibitors after the withdrawal of rofecoxib from the market. SS is a paid scientific advisory board member for HealthCore and ii4sm.
Clinical Commentary: Recent studies have raised concerns about reduced efficacy of clopidogrel when used concurrently with proton pump inhibitors (PPIs) as mediated by competitive inhibition of the CYP2C19 isoenzyme. However, many of these studies had methodological limitations. We studied a cohort of 18,565 community-dwelling elderly patients discharged from hospital after percutaneous coronary intervention or acute coronary syndrome (ACS) and who were prescribed clopidogrel with or without a PPI. Our analysis did not to demonstrate a large or statistically significant increase in the relative risk of MI or death for patients treated concurrently with clopidogrel and PPIs, but our data were consistent with the possibility of a modest clinical effect. Given the lack of compelling trial evidence documenting a meaningful gastro-protective effect from the co-prescription of a PPI with clopidogrel, coupled with the high risk of MI or death in patients after ACS or revascularization, one prudent strategy would be to co-prescribe PPIs primarily in clopidogrel-treated patients at higher risk of adverse gastrointestinal events.