The data analyzed in this study were obtained from a merged retrospective clinical registry including all patients consecutively admitted to the cardiology service of 2 tertiary hospitals with a primary and definitive diagnosis of ACS (n = 4840) from January 1, 2012 to December 31, 2015. Diagnosis of ACS was based on the presence of new-onset symptoms suggestive of myocardial ischemia and any of the following criteria: cardiac biomarkers above the higher normal laboratory limit, ST-segment deviation on electrocardiogram, in-hospital stress testing showing ischemia, or a known history of coronary vessel disease. The patients were classified as having ST-segment elevation MI or non—ST-segment elevation ACS (non—ST-segment elevation MI or unstable angina). Diagnosis of unstable angina required the presence of suggestive symptoms together with objective evidence of myocardial ischemia on stress testing or detection of a culprit lesion of ≥ 50% on coronary angiography, in addition to cardiac biomarkers below the higher normal laboratory limit.

For the purpose of the present study, patients who died in-hospital (n = 199) were excluded. Moreover, since our objective was to study the prognostic impact of MI and bleeding separately, we excluded patients who showed both complications during follow-up (n = 23). Patients without in-hospital coronary arteriography during the index admission (n = 187) and patients lost to follow-up (n = 202) were also excluded. Therefore, the final cohort of the present study consisted of 4229 patients.

The primary endpoint of interest for this analysis was all-cause death. We aimed to evaluate the association of all-cause mortality with both MI and the presence, as well as the severity, of bleeding. A secondary endpoint was description of the cardiovascular and noncardiovascular mortality rates according to MI and bleeding status during follow-up.

Our study only includes the first bleeding event and MI that occurred during follow-up. Adjudication of MI and bleeding events was performed by personnel unaware of the endpoints of interest for this analysis.

The ascertainment of MI and bleeding status during follow-up was carried out between September and October of 2016. We reviewed the electronic medical record and all the medical attendances and hospital records. For each patient in the study, we had information on the characteristics of bleeding events, location, clinical severity, imaging tests, hemoglobin levels, and blood transfusion.

Myocardial infarction during follow-up was defined as type 1 MI (spontaneous MI), type 3 MI (fatal MI in the absence of biomarker values), or type 4b MI (MI related to stent thrombosis), in accordance with the Third Universal Definition.13

Bleeding events were classified according to the Bleeding Academic Research Consortium (BARC) criteria.14 Due to the retrospective nature of the study and the difficulty involved in the adjudication of BARC bleeding type 1, this type of bleeding was not included in the present study. Because the primary outcome of this analysis was mortality, bleeding events belonging to class 5 of BARC-defined bleeding (fatal bleeding) were not analyzed as a separate class but were distributed to other classes according to initial assessment, as previously described.15,16 Therefore, the BARC hemorrhages analyzed in this study were type 2, 3a, 3b, and 3c.

The cause of death (cardiovascular vs noncardiovascular) was classified according to the definitions of Hicks et al.17 When no information was available or there was no consensus on the cause of death, it was included in the “unknown or unclassifiable cause of death” group.

Baseline and clinical characteristics were summarized by MI and bleeding status experienced after hospital discharge. Continuous variables are presented as medians [interquartile ranges], and categorical variables as counts and proportions. Differences between continuous variables in patients who did and did not experience bleeding events or MIs were compared using the Wilcoxon rank-sum test. The chi-square test was used to compare categorical variables.

Since MI and bleeding status change over time, the adjusted effect of postdischarge MI and bleeding on subsequent all-cause mortality was investigated using separate Cox-type shared frailty models with MI and bleeding as time-varying covariates. For the prior purpose, we used the user-written command “strmcure” (without cured fraction and with nonstratification), which allows testing a direct and indirect effect of a baseline exposure on the terminal endpoint.18 In the first “strmcure” model, we investigated the effect of postdischarge MI and bleeding as a whole. In another model, on top of postdischarge MI, the severity of BARC type bleeding events was entered as a hierarchical time-varying covariate in which the highest risk state was BARC type 3c followed by BARC type 3b, BARC type 3a, and then BARC type 2 bleeding.

To obtain evidence on the role of DAPT status at the time of MI or bleeding on mortality risk, an additional analysis was performed by generating categorical time-dependent covariates for both MI and bleeding with different levels corresponding to whether a patient was on- or off-DAPT at the time of each respective event.

All regression models were adjusted using covariates that were either significantly different between groups according to MI and bleeding status, or plausibly related to mortality. Continuous variables that did not meet linearity assumptions were transformed. The following covariates were included in the regression models: age ≥ 65 years, female sex, body mass index (kg/m2), current smoking, hypertension, dyslipidemia, diabetes mellitus, prior coronary artery disease, history of congestive heart failure or left ventricular ejection fraction < 40%, prior vascular disease, history of bleeding, malignant disease, chronic obstructive pulmonary disease, ACS type, baseline serum creatinine (mg/dL) and hemoglobin values (g/dL), multivessel coronary disease, in-hospital PCI, in-hospital coronary artery bypass graft, treatment at discharge, and study center.

The bootstrap method (500 iterations) was used to calculate and estimate statistical significance and hazard ratios (HRs) and their 95% confidence intervals (95%CIs).

To estimate the number of deaths attributable to MI or bleeding, the adjusted HR for each event was applied to the actual number of deaths according to the following formula: number of deaths among patients with MI or bleeding × [adjusted HR-1]/ adjusted HR.

The differential impact on mortality of MI vs each BARC bleeding type being studied was generated from the corresponding adjusted Cox-based models using the lincom command and is expressed as relative risk ratio (RRr).

Finally, we investigated the time-pattern of the risk of death associated with postdischarge MI and bleeding as a function of the time elapsed since the event. For this purpose, the functional form of the relationship between the adjusted hazard of death, and the time elapsed since bleeding or MI was fitted using fractional polynomial, and the resulting curve along with the 95%CI of the mean was plotted.

Values were considered statistically significant when the P value was < .05. All statistical analyses were performed using Stata/MP 13.1.

The study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the local ethics committee.

During a median follow-up of 455 days [377-783] days, 204 patients experienced a MI (crude rate: 3.1 per 100 person-years; 95%CI, 2.65-3.49). The median time from discharge to the MI event was 192 [56-340] days. Most postdischarge MIs were non—ST-segment elevation MIs (83.8%). Twelve MIs were fatal, with ST-segment elevation MI accounting for 62.8% of the case-fatality.

A total of 500 patients had a bleeding complication (crude rate: 7.7 per 100 person-years; 95%CI, 7.01-8.36), with a distribution according to BARC type 2, 3a, 3b and 3c bleeding event of 359 (8.5%), 61 (1.4%), 49 (1.2%), and 31 (0.7%) patients, respectively.

The median time from discharge to BARC type 2, 3a, 3b, and 3c bleeding events was 300 [117-405] days, 149 [41-410] days, 167 [82-321] days, and 314 days [177-622] days, respectively.

Overall, 91.2% of the bleedings were spontaneous (ie, nonprocedural-related). Around half (53%) of bleeding episodes were gastrointestinal and 9 bleeding events were fatal with intracranial bleeding accounting for 55.6% of the case-fatality.

Baseline and clinical characteristics by postdischarge MI and bleeding status are presented in Table 1. In summary, patients who had a postdischarge MI were older and in general had a greater comorbidity burden (hypertension, diabetes, dyslipidemia, prior vascular disease, heart failure, and chronic obstructive pulmonary disease) compared with patients who did not have a MI. They were also more likely to have multivessel coronary artery disease, but had similar a rate of in-hospital PCI compared with patients without postdischarge MI, and they were more likely to receive clopidogrel at hospital discharge.

Patients who had a bleeding event after discharge were older, had a greater comorbidity burden (hypertension, prior vascular disease, and heart failure), and more frequently had prior bleeding than patients who did not have a bleed. They were also less likely to undergo in-hospital coronary artery bypass graft, and to more frequently receive oral anticoagulant at hospital discharge.

During follow-up, there were 335 deaths; the number of patients who died following MI or bleeding is summarized in Table 2. When considered as time-varying covariates in the multivariable-adjusted model, both MI (HR, 5.8; 95%CI, 3.7-9.8) and bleeding (as a whole) (HR, 5.1; 95%CI, 3.6-7.7) were significantly associated with mortality. The mortality attributable to MI was 10.6%, while bleeding-attributable mortality was 16.8%.

Detailed univariable and multivariable Cox analyses are shown in Table 1 of the supplementary material and Table 2 of the supplementary material.

When we repeated this analysis and included only those patients who received DAPT at discharge (n = 3588 with: 189 MIs, 428 bleeding events, and 270 deaths), the results did not substantially change: HR, 5.7; 95%CI, 3.6-9.3 for MI, and HR, 5.4; 95%CI, 3.6-8.1) for bleeding (Table 3 of the supplementary material).

The adjusted differential effect on mortality of MI and each of the 4 BARC severity categories is shown in Figure 1. There was a stepwise increase in adjusted mortality risk from the least to the most severe BARC bleeding category. The relative hazard of death was 4-fold higher in patients experiencing MI vs BARC type 2 bleeding [6.6 vs 1.7; RRr, 3.8; 95%CI, 2.6–7.8; P < .001], and almost 2-fold higher in patients experiencing MI vs BARC type 3a bleeding (6.6 vs 3.3; RRr, 1.9; 95%CI, 1.03-3.4; P = .04). In contrast, no significant difference was found comparing MI and BARC type 3b bleeding (6.6 vs 7.2; RRr, 0.9; 95%CI, 0.4–2.2; P = .88). However, the risk of mortality associated with MI was significantly lower (4-fold) than for BARC type 3c bleeding (HR, 26.5 vs 6.6; RRr, 0.25; 95%CI, 0.12–0.43; P < .001).

Figure 1.

Differential impact of postdischarge MI vs bleeding severity, on subsequent mortality. Myocardial infarction and bleeding are represented in green and blue, respectively. 95%CI, 95% confidence interval; BARC, Bleeding Academic Research Consortium; MI, myocardial infarction.

(0.17MB).

When we rerun the prior analysis and included only the 3588 patients who received DAPT at discharge, we found similar results to those observed in the entire cohort (Figure 1 of the supplementary material).

The adjusted mortality risk following both MI and bleeding peaked early after the occurrence of these events, and despite showing a substantial deceleration thereafter, remained significant for several months (Figure 2). A similar pattern was observed in patients discharged with DAPT (Figure 2 of the supplementary material).

Figure 2.

Time-association pattern of the impact of postdischarge myocardial infarction vs bleeding, on subsequent mortality throughout follow-up.

(0.18MB).

The overall rate of cardiovascular death was 45.4% (n = 152/335), and the cause of death was unknown or unclassifiable in 47 (14%) patients. The rate of cardiovascular death in patients with MI was 74.4% (n = 32/43) vs 41.4% (n = 29/70) in patients who had postdischarge bleeding (P < .001) (Figure 3). Similar rates of unknown or unclassifiable causes of death were observed after MI (11.6%) and bleeding (8.6%) during follow-up (P = .6). The specific cause of death by MI and bleeding events is shown in Figure 3 of the supplementary material.

Figure 3.

Cause of death by postdischarge MI and bleeding events during follow-up. MI, myocardial infarction.

(0.07MB).

The influence of DAPT status at the time of postdischarge MI and bleeding on mortality is shown in Table 3 of the supplementary material. Among the 3588 patients discharged with DAPT, there were 189 MIs, 428 bleeding events, and 270 deaths (crude mortality rate = 3.2 per 100 person-years; 95%CI, 2.8-3.7).

Data on DAPT status at the time of MI or bleeding were available in all except 12 patients (2 MIs and 10 bleeding events). Among patients with an MI event, 73% (n = 138/189) were on-DAPT. Of those patients who had a bleeding event, 74.8% (n = 320/428) were on-DAPT.

The mortality risk after MI was almost 3-fold higher for patients on-DAPT (HR, 2.9; 95%CI, 1.8-4.5) than for those off-DAPT (HR, 1.5; 95%CI, 0.7-3.4). In contrast, mortality risk associated with bleeding was ∼2-fold lower for patients on-DAPT (HR, 1.6; 95%CI, 1.1-2.6) than for off-DAPT patients (HR, 3.2; 95%CI, 1.7-5.8) (Table 3).

This was a retrospective study with the limitations inherent to this type of design. One important limitation was the lack of some data on follow-up in 202 patients who were excluded from the main analysis. Although this could theoretically result in selection bias, a comparison of patients with and without missing data on follow-up indicated no significant differences between the 2 groups (Table 4 of the supplementary material). Additionally, although every effort was made to collect MI and bleeding events during follow-up, we cannot completely rule out the possibility that the rate of MI was underestimated. Furthermore, we did not collect data on BARC type 1 and type 4 bleeding. However, this could represent a mild limitation, since in several studies, BARC type 1 bleeding was not associated with a significant increase in the risk of death, and postdischarge BARC type 4 is a rarity as very few patients during follow-up underwent coronary artery bypass graft.

In this study, data on MI type 2 (MI secondary to an ischemic imbalance), MI type 4a (MI related to PCI), and MI type 5 (MI related to coronary artery bypass graft) were not collected; hence our findings cannot be generalized to all types of MI. In addition, we did not perform subgroup analysis according to the P2Y12 inhibitor due to the relatively small sample size; therefore, the results cannot be generalized to the different P2Y12 inhibitor types.

Finally, we were not able to characterize the prognosis of patients with repeated MIs or bleeding events since we only had data of the first MI and bleeding complication during follow-up. Therefore, our results cannot be generalized to this particular group of patients, which, based on previous studies, could represent around 1% of the ACS population.12