A retrospective, multicentre, observational study was conducted in patients with an admission diagnosis of ACS and who were receiving aspirin and an anti-P2Y12 drug (clopidogrel, prasugrel, or ticagrelor) at the time of hospital discharge or in-hospital death; the patients were prospectively included in the ARIAM-Andalusia registry between 2013 and 2015. The registry characteristics have been described previously.18–21 Briefly, ARIAM-Andalusia is an ongoing electronic repository that compiles data on all ACS patients admitted to coronary care units in hospitals in Andalusia; data include demographic characteristics, clinical variables, analytical results, treatments, procedures, time to reperfusion, major cardiac and cerebrovascular events, and in-hospital bleeding events. For the present study, we selected data from the 4 tertiary referral hospitals with the most registered patients and the least number of missing study variables. The procedures for data compilation and verification in the online repository are quality audited by the Andalusian School of Public Health.18–21

Antiplatelet agents were prescribed by the on-duty physician according to standard clinical practice at each center. Myocardial infarction was identified according to the third universal definition. Ischemic stroke was defined as any cerebrovascular incident causing a neurological deficit lasting for longer than 24hours, and the absence of hemorrhage was confirmed with neuroimaging tests. Stent thrombosis was defined according to Academic Research Consortium criteria.22 All clinical events were previously entered on the data acquisition form and were allocated by consensus among the investigators at each center.19–21 The primary efficacy variable was total in-hospital mortality and the secondary efficacy variables were nonfatal myocardial infarction and stroke and probable or definite stent thrombosis. The primary safety variable was the total number of bleeding events according to the TIMI (Thrombolysis in Myocardial Infarction) criteria.23

All analyses were stratified according to the type of anti-P2Y12 patients were receiving at the time of hospital discharge or in-hospital death: NAA (prasugrel or ticagrelor as a single category) vs clopidogrel. Depending on their distribution, continuous variables are expressed as mean ± standard deviation or median [interquartile range from the 25th to the 75th percentile] and were compared with the Student t test or the Mann-Whitney U test. Qualitative variables are expressed as number and percentage and were compared by the chi-square test or the Fisher exact test.

To adjust for differences between treatment groups in baseline characteristics and to attenuate possible confounding by covariates, we calculated the propensity scores (PS) for receiving an NAA,24 using a multivariate regression model that included the following variables: age, sex, body mass, smoking status, diabetes mellitus, high blood pressure, previous myocardial infarction, previous stroke, bleeding history, peripheral vascular ischemia, history of atrial fibrillation and treatment with oral anticoagulants, ST-segment elevation ACS, in-hospital percutaneous or surgical coronary revascularization, year of admission, 3-vessel disease or left main coronary artery disease, treatment with glycoprotein IIb/IIIa inhibitors, heart failure, renal failure, and total ischemia time (from symptom onset to reperfusion). Patients from each treatment group were propensity-score matched with a 1:1 greedy algorithm (caliper width, 0.05) without replacement, and the goodness-of-fit was assumed to be sufficient if the standardized differences were < 10%.24,25 To analyze the robustness of the results, we conducted additional regression analysis by IPTW (inverse probability of treatment weight), both in the full sample and with restriction to exclude extreme propensity scores.26 The contribution of each patient or weighting was calculated in the NAA group as the inverse of the propensity score (1/PS) and in the clopidogrel group as 1/1-PS. This type of analysis ensures that the contribution of covariates introduced to construct the propensity model does not differ between the members of each group. Between-group differences in baseline characteristics in the matched sample were analyzed with the McNemar test (quantitative variables) and the Student t test for paired data (quantitative variables). Between-group differences in clinical events in the matched sample were analyzed with conditional multivariate regression models, and are expressed as odds ratios (OR) and the corresponding 95% confidence intervals (95%CI). The results were internally validated by sensitivity analysis of mortality and total bleeding events in distinct patient subgroupings: GRACE and CRUSADE risk scores, age ≥ 75 years, type of ACS, and admission year. The discriminatory power of the regression models was determined with the C statistic, and the models were calibrated using the Hosmer-Lemeshow test. Differences were considered significant at a bilateral P value < .05. Statistical analysis was conducted with SPSS 19 (IBM Corporation; Somers, New York, United States) and STATA 13.1 IC (STATA Corp; College Station, Texas, United States).

During the course of the study, the rate of NAA use increased significantly (P < .001), from 31% of patients included in 2013, to 53% in 2014, and 63% in 2015; this trend was largely due to an increased use of ticagrelor from 2014 (Figure 2).

Figure 2.

Time trend in P2Y12 platelet receptor inhibitor use. The figure indicates the quarterly percentages in the use of clopidogrel, prasugrel, and ticagrelor. Q, calendar quarter.

(0.15MB).

The total mortality rate was lower in the NAA-treated group (2% vs 9%; P < .0001) (Table 2). The rate of nonfatal thrombotic events was low, with the NAA group having relatively lower rates of ischemic stroke (0.1% vs 0.5%; P = .025) and stent thrombosis (0.07% vs 0.5%; P = .025); there was no between-group difference in the rate of nonfatal myocardial infarction. There was also no significant difference in mortality between patients taking prasugrel and those taking ticagrelor (1.8% vs 1.4%; P = .701).

Regarding safety, there was no difference in bleeding rate or severity between the 2 treatment groups (Table 2) or in bleeding rate between patients taking prasugrel and those taking ticagrelor (3.1% vs 4.5%; P = .409).

Mortality rates in patient subgroups were similar to the rate in the total population; however, NAA was notably more beneficial in patients at higher ischemic risk (GRACE score > 140: 3% vs 14.5%; P for interaction = .0026) (Figure 3A, Figure 1 of the supplementary material). Moreover, this tendency was maintained after propensity-score matching to adjust for differences in baseline characteristics (Figure 2 of the supplementary material). The lower mortality with NAA was evident even for the subgroups at higher bleeding risk (CRUSADE score > 50: 9.4% vs 29%; P =.047), albeit without evidence of interaction (P =.485) (Figure 3A and Figure 1 of the supplementary material and Figure 2 of the supplementary material). After propensity-score matching, the absolute mortality reduction with NAA was around 12% in the patient subgroup with a high ischemic risk and a high or very high bleeding risk (Figure 2 of the supplementary material).

Figure 3.

Forest plot of clinical events stratified by patient subgroup. A: Total mortality. B: Total bleeding events. 95%CI, 95% confidence interval; ACS, acute coronary syndrome; NAA, new antiplatelet agents; NSTEACS, non—ST-segment elevation acute coronary syndrome; OR, odds ratio; STEACS, ST-segment elevation acute coronary syndrome.

(0.5MB).

NAA treatment was nonsignificantly associated with a higher overall bleeding rate in all subgroups, especially patients older than 75 years, although there was no evidence of interaction (Figure 3B). This tendency was maintained after propensity-score matching.

Propensity scoring matched 896 patients in each treatment group, with a good between-group balance of standardized differences (Table 1). The adjusted multivariate analysis of the propensity-score matched data showed a significant reduction in total mortality with NAA vs clopidogrel (OR, 1.07; 95%CI, 0.18-2.37; P = .094), resulting in a net clinical benefit (Table 3). The IPTW analysis confirmed the lower mortality and the nonsignificant increase in total bleeding events with NAA (Table 3). Both predictive models showed excellent discrimination and calibration. It is noteworthy that, whereas the mortality reduction with NAA was maintained after the 2 adjusted analyses, the reduction in overall bleeding rate with NAA in the total population (OR, 0.80; 95%CI, 0.54-1.20) was reversed to the opposite trend after propensity-score matching. The greater potency of NAA predicts a higher bleeding rate, and this finding therefore reinforces the validity and robustness of the model.

This study has the usual limitations of registry-based observational studies, and the detected associations should therefore not be interpreted as causal.

The study population consisted of coronary care unit patients, and the study examined only in-hospital events and did not cover the recommended 12 months of dual antiplatelet therapy for ACS. Caution should therefore be exercised in extrapolating these results to other populations. The different baseline characteristics and procedures in the 2 treatment groups reflect routine clinical practice; although several types of adjustment were applied, we cannot exclude an influence on the results from confounding covariates not included in the propensity models or not recorded in the study. As with other registries, event allocation was neither centralized nor blind. Nonetheless, events in the ARIAM-Andalusia registry are predefined and subject to periodic external audits, thus ensuring reliable allocation. The results of the subgroup analyses should be regarded as only exploratory. The study recorded treatment allocation at the time of patient discharge or in-hospital death, and therefore the results may have been influenced by antiplatelet switching during hospitalization. However, because of the retrospective nature of the study and the absence of a predefined variable, we were unable to estimate the possible influence of antiplatelet agent switching. This question is being specifically addressed in the ongoing prospective multicenter CREA registry (Spanish acronym for “Antiplatelet Therapy in Acute Coronary Syndrome: Safety and Efficacy of Switching Antiplatelet”; ClinicalTrials.gov: NCT02500290).