P2Y12 Inhibitors in Acute Coronary Syndromes: A Real-World, Community-Based Comparison of Ischemic and Bleeding Outcomes

Background Randomized trials have shown superiority of the novel P2Y12 inhibitors over clopidogrel in patients with acute coronary syndrome (ACS), but clinical benefit in the community remains controversial. Our objective was to compare the safety and efficacy of clopidogrel to ticagrelor and prasugrel in patients with ACS undergoing percutaneous coronary intervention (PCI) in a real-world population. Methods We conducted a retrospective cohort study of patients with ACS who underwent PCI and were discharged with clopidogrel, ticagrelor, or prasugrel from 2012 to 2018 within Kaiser Permanente Northern California. We used Cox proportional hazard models with propensity-score matching to evaluate the association of the P2Y12 agent with the primary outcomes of all-cause mortality, myocardial infarction (MI), stroke, and bleeding events. Results The study included 15,476 patients (93.1% on clopidogrel, 3.6% on ticagrelor and 3.2% on prasugrel). Compared to the clopidogrel group, ticagrelorand prasugrel patients were younger with less comorbidities. In multivariable models with propensity-score matching, we found a lower risk of all-cause mortality in the ticagrelor vs the clopidogrel group (HR (95% CI) 0.43 (0.20–0.92)), but no differences in the other endpoints, and no difference between prasugrel and clopidogrel among any endpoints. A larger proportion of patients on ticagrelor or prasugrel switched to an alternative P2Y12 agent vs. clopidogrel (p < 0.01), and a higher level of persistence was seen among patients on clopidogrel vs. ticagrelor (p = 0.03) or prasugrel (p < 0.01). Conclusion Among patients with ACS who underwent PCI, we observed a lower risk of all-cause mortality in patients treated with ticagrelor vs clopidogrel, but no difference in other clinical endpoints nor any differences in endpoints between prasugrel vs. clopidogrel users. These results suggest that further study is needed to identify an optimal P2Y12 inhibitor in a real-world population.


Introduction
Randomized trial data have demonstrated an approximately 2% reduction in ischemic events and a 0.5-1% increase in bleeding events with ticagrelor or prasugrel versus clopidogrel in those presenting with acute coronary syndrome (ACS) [1,2]. Tese data have not been uniformly replicated in real-world, nonclinical trial populations [3][4][5]. Te largest of these studies evaluated ticagrelor versus clopidogrel and showed that 1 year risk of net adverse clinical events was not diferent between the groups after propensity score matching, with higher rates of bleeding and dyspnea in the ticagrelor group [3]. In this study, we aimed to determine the comparative efcacy and safety of the novel P2Y12 inhibitors versus clopidogrel in a large integrated healthcare delivery system, among those presenting with ACS undergoing percutaneous coronary intervention (PCI).

2.2.
Exposure. P2Y12 antagonist use was obtained electronically, using previously validated methods [8]. Briefy, patients were determined to be taking one of the three P2Y12 antagonists if they had dispense records of ticagrelor, prasugrel, or clopidogrel within 30-days post index PCI discharge.
Adherence, persistence, and switching of P2Y12 inhibitors were assessed using previously defned methods [4]. Adherence, or medication refll adherence (MRA), was defned as the total days of medication supply in one year divided by 365 and was expressed as a percentage. Patients with an MRA >80% were defned as adherent. We considered patients "nonpersistent" if the gap between reflls exceeded the days of supply plus a 15-day grace period. Drug switching was defned as more than one P2Y12 inhibitor within the frst 365 days. For subjects who switched P2Y12 inhibitors during the study, all P2Y12 fll information was considered for the calculation of adherence and persistence (i.e., patients were considered adherent if they flled the second P2Y12 inhibitor at appropriate intervals).

Outcome.
Te primary study outcomes included allcause mortality, hospitalized myocardial infarction, hospitalized stroke, and hospitalized bleeding events examined within the frst year after the index ACS event. All hospitalized events were determined using ICD codes with principal, primary, or secondary diagnoses during the hospitalization. Patients were followed from the index event until they died, had an outcome of interest, or for at most one year, whichever occurred frst. All hospitalized outcomes were collected independently (e.g., patients who had hospitalized stroke continued to be followed for other outcomes).

Data Collection.
Baseline demographic, laboratory data, procedural data, and medication use after index PCI were extracted from various KPNC electronic databases including variables submitted to the ACC/NCDR Cath PCI Quality Registry in accordance with defnitions specifed in both version 4.4 and 5 [7]. Baseline comorbidities were identifed using ICD-9 and 10 code defnitions within one-year prior to the index event. Laboratory data were obtained at the time of cardiac catheterization or the most recent value within one day before the procedure. Te use of other cardiac medications was identifed within 30-days post index event. Additionally, the PRECISE DAPT score, a validated score for predicting bleeding risk after stent implantation, was calculated for each patient [9].

Statistical Analysis.
Descriptive statistics were used to describe the demographic and clinical characteristics of the cohort by P2Y12 inhibitor use. We used clopidogrel as our reference group and ticagrelor or prasugrel was compared to clopidogrel, respectively. Diferences in characteristics were assessed using the two-sample t-tests for continuous variables and the chi-square tests for categorical variables. Tese tests were also used to assess diferences in adherence, persistence, and drug switch between novel P2Y12 inhibitors and clopidogrel from index event to one-year follow-up. Trends in P2Y12 inhibitor use during the study period were evaluated using the Cochran-Mantel-Haenszel test.
We used the Kaplan-Meier (KM) method to analyze outcomes in patients across one-year of follow-up. Comparisons between patients on diferent P2Y12 inhibitors were conducted using log-rank tests. We calculated the hazard ratio (HR) and 95% confdence interval (CI) for the associations between P2Y12 inhibitor use and each outcome using bivariate and multivariable Cox proportional hazards (PH) regression models. Covariates for the multivariable models were selected a priori based on previously published studies, clinical relevance, or a p value <0.1 from bivariate analyses assessing demographic and clinical characteristics and P2Y12 inhibitor use. We performed two steps of adjustment by including all potential covariates in the multivariable models frst, and subsequently dropped those covariates that did not have signifcant associations (i.e., p > 0.05) with the outcomes except demographic characteristics and medication adherence rate, based upon a priori relevance to the research question.
Since the choice of P2Y12 inhibitor may be associated with other prognostic factors, we also performed propensity score (PS) matching using the optimal variable ratio matching method to balance the groups at baseline. Each patient in the ticagrelor or prasugrel group was matched with patients in the clopidogrel group with minimal 1 : 1 match and maximal 1 : 5 match. Caliper width was 0.25. Baseline variables selected for PSM included age, gender, and all clinical risk factors demonstrated p < 0.1 in Table 1 (i.e., all cardiovascular history and risk factors, smoking status, chronic kidney disease, chronic lung disease, and PCI indication). Tese variables are also controlled in the subsequent multivariable models, with additional variables that are deemed statistically signifcant in unadjusted analysis or clinically important. Tis study was presented as an abstract at the Society for Cardiovascular Angiography and Interventions (SCAI) conference in May, 2022 [10]. During the preparation of the manuscript, it was realized that hypertension was inadvertently omitted from the statistical models (both multivariate and propensity matching) for the abstract. Given that hypertension is an important risk factor that should be adjusted for, we made sure it was added into the models and the analysis was rerun-this explains the results on the manuscript that are discordant from the abstract.
All data extraction and analyses were performed using SAS 9.4 (Cary, North Carolina).
Te mean age at index PCI was 66.3 (standard deviation [SD] ± 12.0) years and 27.8% of the cohort were female (Table 1). Compared to the ticagrelor or prasugrel groups, the clopidogrel group was older, had a higher proportion of female patients, was more likely to have a signifcant cardiovascular history such as prior CABG, cerebrovascular disease, and heart failure, and had more comorbidities such as chronic kidney disease and hypertension (all p < 0.05). Additionally, the clopidogrel group had lower hemoglobin and platelet values compared to either the ticagrelor or   prasugrel group (p < 0.001). A larger proportion of patients in the prasugrel (40.7%) or ticagrelor (49.8%) group had STelevation myocardial infarction (STEMI) as their PCI indication versus clopidogrel (20%), and a higher proportion of these patients were classifed as emergency cases and having cardiogenic shock. Clopidogrel patients also had more veins graft PCI, as well as higher PRECISE DAPT scores, and higher use of concomitant oral anticoagulation. Table 2 shows the baseline characteristics of the propensity-matched population, and Supplementary Tables 1a and 1b demonstrate the characteristics of the clopidogrel patients that were excluded in propensity matching with ticagrelor and prasugrel, respectively. Adherence to P2Y12 inhibitor therapy was similar across the three groups (Table 3). However, a higher proportion of patients on novel P2Y12 inhibitors switched drugs during follow-up and most of these (92%) switched to clopidogrel. Additionally, persistence to treatment was lower in the prasugrel and ticagrelor group, compared to the clopidogrel group (p < 0.03 for ticagrelor vs clopidogrel; p < 0.001 for prasugrel vs clopidogrel).
At one-year follow-up, the clopidogrel group had higher mortality rates than the ticagrelor group (p < 0.01) ( Table 4).
Te incidence of myocardial infarction, stroke, and bleeding events were similar across the three groups. Te Kaplan-Meier curves for the outcomes are included in Supplementary Figure 1. Table 5 shows the incidence of adverse events in the propensity-matched populations.
After combined propensity and multivariable adjustment, ticagrelor was associated with a lower risk of all-cause mortality, compared to clopidogrel (adjusted HR 0.43, 95% CI, 0.20-0.92) ( Table 6). Tere were no diferences in myocardial infarction, stroke, or bleeding events between the ticagrelor and clopidogrel groups. Similarly, after adjustment, there were no diferences between clopidogrel and prasugrel groups in risks of death, myocardial infarction, stroke, or bleeding events.

Discussion
After combined multivariable and propensity adjustment in patients undergoing PCI for ACS, the use of ticagrelor compared to clopidogrel was associated with a lower risk of all-cause mortality, but similar rates of hospitalized myocardial infarction, stroke, and bleeding. We found no    Journal of Interventional Cardiology diferences between prasugrel and clopidogrel for any of the adverse outcomes.
In the PLATO trial, the pivotal randomized trial that compared ticagrelor with clopidogrel in ACS, investigators found a similar lower hazard for all-cause death with ticagrelor use, but this was also accompanied by lower rates of myocardial infarction and death from vascular causes [1]. In our study, without any diferences in the rates of myocardial infarction or stroke, it is difcult to ascertain a mechanism for associated diferences in all-cause    show an association between novel P2Y12 inhibitors and myocardial infarction despite its demonstration in randomized trials, while others have replicated some of the trial fndings [3][4][5][11][12][13][14]. Randomized trials, the gold standard for isolating a treatment efect while minimizing bias, create rarefed environments that are not necessarily refective of daily patient care. For example, the PLATO trial excluded patients on oral anticoagulation, dialysis patients, clinically important thrombocytopenia and anemia, as well as "any other condition that may put the patient at risk or infuence study results in the investigators' opinion (e.g., cardiogenic shock, severe hemodynamic instability, active cancer)." In addition, any condition that increases the risk of medication noncompliance or being lost to follow-up was excluded.
Te TRITON TIMI 38 trial, [2] which led to the approval of prasugrel in ACS patients undergoing PCI, similarly has a number of exclusion criteria, including, but not limited to, cardiogenic shock, NYHA class IV congestive heart failure, "clinical fndings, in the judgement of the investigator, associated with an increased risk of bleeding," history of hemorrhagic stroke, ischemic stroke within 3 months, platelet count less than 100,000, anemia (hemoglobin <10 g/dL) at the time of screening, presence of concomitant oral anticoagulation, known severe hepatic dysfunction, and "concomitant medical illness that in the opinion of the investigator is associated with reduced survival over the expected treatment period." Tese protocols exclude many important patient phenotypes. For example, in our study, there were 218 patients with cardiogenic shock, 3201 patients with heart failure, 1136 with liver disease, 420 on dialysis, and 1563 on oral anticoagulation. Tere also tends to be less diversity in ethnic representation in randomized trials, with >90% of the participants in both PLATO and TRITON TIMI 38 being white, compared to over one-third of our patient population being nonwhite.
A higher proportion of patients in the ticagrelor and prasugrel groups switched drugs compared to the clopidogrel group, and most of them switched to clopidogrel.
Both ticagrelor and prasugrel have higher cost [15], and thus that may have played a role, and there is a potential risk of stent thrombosis if switching is not done properly early after stent placement [16]. Also, a higher proportion of patients on clopidogrel were persistent to treatment compared with ticagrelor and prasugrel, which is an observation that has been made before [17] and which also has implications for treatment efect, as premature antiplatelet discontinuation has been identifed as the single most important predictor of stent thrombosis [18].
Our study has a number of limitations. It was an observational study, and P2Y12 inhibitor treatment was not randomly assigned. Any diferences or lack of diferences shown in our study could still be infuenced by unmeasured confounders. Te larger sample size of clopidogrel patients compared to ticagrelor and prasugrel may limit the statistical power of the analysis. Also, aspirin use was not available through the pharmacy dispense database.
Ascertainment of exposure, namely, P2Y12 inhibitor, was based on pharmacy dispense records. Pill counts may have provided a more precise estimate of thienopyridine use, but the use of flled prescriptions has been shown to refect actual medication use by patients with a high degree of accuracy [19].

Conclusion
In patients undergoing PCI for acute coronary syndrome, after both multivariable and propensity adjustment, ticagrelor compared to clopidogrel was associated with lower all-cause mortality, and yet similar rates of myocardial infarction, stroke, and bleeding, while prasugrel compared to clopidogrel was associated with similar rates of all studied, clinically important outcomes. In addition, there was more switching of therapy in patients initially receiving novel P2Y12 inhibitors and these patients tended to be less persistent to treatment. Additional randomized studies, and ones that are more inclusive of patients in real-world practice, are needed to clarify these efects. For all outcomes, we controlled for demographics (age, gender, and race/ethnicity), precise DAPT score, medication adherence rate, creatinine, and hemoglobin. Additionally, for mortality, we controlled for prior myocardial infraction (MI), peripheral vascular disease, heart failure, diabetes, and chronic lung disease; for hospitalized MI, we controlled for prior MI, heart failure, diabetes, PCI indication, and anticoagulation use; for hospitalized stroke, we controlled for cerebrovascular disease, peripheral vascular disease, heart failure, diabetes, and anticoagulation use; for bleeding events, we controlled for peripheral vascular disease, heart failure, and anticoagulation. * * Baseline variables in propensity score (PS) matching include demographics (age, gender, and race/ethnicity), cardiovascular history (prior MI, prior CABG, cerebrovascular disease, peripheral vascular disease, heart failure, atrial fbrillation, hyperlipidemia, hypertension, and diabetes), smoking status, chronic kidney disease, chronic lung disease, and PCI indication. For the subsequent multivariable models, all variables in the PS models were included, with additional variables including race/ethnicity, creatinine, hemoglobin, and medication adherence rate for all outcomes. We also adjusted anticoagulation for all hospitalized events, and precise DAPT score for mortality, MI, and bleeding events. 8 Journal of Interventional Cardiology

Data Availability
Te data used to support the fndings of this study are available from the corresponding author upon request.

Conflicts of Interest
Te authors declare that they have no conficts of interest.