Assessment of Nonfatal Bleeding Events as a Surrogate for Mortality in Coronary Artery Disease

Background Bleeding events are frequently applied as safety end points for randomized controlled trials (RCTs) investigating the effect of antithrombotic agents in patients with coronary artery disease. However, whether a bleeding event is a valid surrogate for death remain uncertain. Objectives This study aimed to assess the correlation between the treatment effect on bleeding events and mortality. Methods Multiple databases were searched to identify RCTs studying antithrombotic agents for patients with coronary artery disease through August 2022. Major and minor bleeding events were defined in included trials, mostly defined with BARC (Bleeding Academic Research Consortium) or TIMI (Thrombolysis In Myocardial Infarction) criteria. Trial-level correlations between nonfatal bleeding events and mortality were assessed. We performed subgroup analyses by the definitions of bleeding (BARC vs TIMI criteria), study year, and follow-up duration. We used a cutoff with a lower limit of 95% confidence interval of R2 >0.72 as a strong correlation and with an upper limit of 95% confidence interval of R2 <0.50 as a weak correlation. Results A total of 48 RCTs with 181,951 participants were analyzed. Overall, trial-level R2 for major and minor bleeding were 0.09 (95% CI: 0.00-0.26) and 0.09 (95% CI: 0.00-0.27) for all-cause or cardiovascular death, respectively. When confined to major bleeding, R2 were 0.03 (95% CI: 0.00-0.13) and 0.01 (95% CI: 0.00-0.05), respectively. All of the subgroup analyses did not show any significant correlations. Conclusions We demonstrated a trial-defined bleeding event may not be a valid surrogate for mortality in RCTs investigating the effect of antithrombotic agents for coronary artery disease.

A bleeding event is considered as a pri- mary safety outcome in randomized controlled trials (RCTs) investigating antithrombotic agents for patients with coronary artery disease (CAD).The significance of bleeding events has increased over the last decade, accompanied by the advancement of drug-eluting stent or percutaneous coronary intervention techniques including intravascular imaging. 1,2][5][6][7][8][9][10][11][12][13][14][15][16] In general, it is assumed that bleeding events lead to worse long-term prognosis, including increased risk of death. 17[20][21] Surrogate end points, including bleeding events, are used frequently in RCTs since they usually occur earlier than death enabling RCTs to be conducted with shorter follow-up periods, smaller sample sizes, and lower costs, which allows patients to access new therapies earlier. 22However, it remains uncertain whether a bleeding event can be a strict surrogate of death as part of a composite outcome. 14,23Indeed, a previous study demonstrated that bleeding events could lead to bleeding-related deaths, especially in the early phase after acute coronary syndrome or percutaneous coronary intervention, but not lead to non-bleeding-related death. 17,24Moreover, a recent study revealed that even myocardial infarction is not a surrogate of death in trials to treat or prevent CAD. 25 The association between the treatment effect on a bleeding end point as a surrogate end point and the treatment effect on mortality as a final end point in a meta-analysis of RCTs had not been performed. 17rein, our aim of this study was to assess the correlation between the treatment effect of an intervention such as shortened dual antiplatelet therapy (DAPT) duration on bleeding events and the treatment effect of the same intervention on all-cause or cardiovascular mortality.To clarify this correlation, we performed a meta-analysis of RCTs which investigated bleeding events and mortality in RCTs investigating the effect of antithrombotic agents for patients with CAD.

METHODS
This analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. 26The protocol was registered to the Prospective Register of Systematic Reviews (CRD 42022356035).Institutional review board/ethics exemption was granted for the design of the study.
ELIGIBILITY CRITERIA.The eligibility criteria were as follows: 1) the study was published in a peerreviewed journal; 2) the design was a RCT with at least 2 different antithrombotic agents (DAPT strategies) for patients with CAD; 3) the study investigated mortality (all-cause and/or cardiovascular death) and a trial defined major or major/minor bleeding outcome as the safety endpoint; and 4) a sample size $500 patients. 27e trial defined major/minor bleedings were defined using the definition used in each trial as the safety end point; BARC (Bleeding Academic Research Consortium) type 2, 3, or 5 bleeding, 28 BARC type 2, 3, 4, or 5 bleeding, or TIMI (Thrombolysis In Myocardial Infarction) major or minor bleeding was used. 29milarly, the trial defined major bleeding was described with the definition which was used in each trial as the safety end point; BARC type 3 or 5 bleeding, BARC type 3, 4, or 5 bleeding, 28 TIMI major bleeding, or GUSTO (Global Utilization of Streptokinase and Tissue Plasminogen Activator for Occluded Arteries) moderate or severe bleeding was used. 28,29FORMATION SOURCES AND DATA COLLECTION PROCESS.PubMed, EMBASE, and Cochrane CEN-TRAL databases were searched by a medical librarian with expertise in performing systematic reviews to identify all studies published on April 27, 2022, that investigated the effect of antithrombotic agents for patients with CAD.The extensive search strategy is shown in Supplemental Tables 1 to 3. Additional studies were added via a manual search of other sources, including references from identified articles, systematic reviews or meta-analyses, and commentaries.Two authors (A.W and T.K.) explored the search results to select studies according to the inclusion criteria.Then, these 2 authors reviewed the studies and independently judged the selection and outcomes with the Cochrane Collaboration risk of bias 2.0 tool. 30tracted variables included study subjects, year of the first patient enrollment, the total number of randomized participants, numbers of participants by intervention or control groups, median follow-up duration, and definition of trial defined major or major/minor bleeding.Extracted outcomes included the number of all-cause and cardiovascular death, trial defined major or major/minor bleeding, and odds ratio (OR) or HR of treatment effect for all-cause and cardiovascular mortality, trial defined major or major/ minor bleeding (if reported).

Bleeding Events and Mortality
M A Y 2 0 2 3 : 1 0 0 2 7 6 STATISTICAL ANALYSIS.We calculated odds ratio as a relative measure of effect from the number of events and the number of individuals in the intervention and control groups in each RCT.To confirm the correlation between all-cause or cardiovascular mortality and trial defined major/minor or major bleeding outcomes visually, we plotted the logarithm of odds ratios (log-ORs) for bleeding outcomes on the X coordinate and those for all-cause or cardiovascular mortality on the Y coordinate.We regressed log-ORs for all-cause or cardiovascular mortality on log-ORs for bleeding outcomes, weighted by the number of enrolled patients in each study, both of which were recommended for a trial-level correlation. 31We reported the slope, intercept, and the coefficient of determination (R 2 of 1 indicates that the regression predictions fit perfectly) in these models.We repeated the analysis using the only articles which reported HR.We assessed the between-study heterogeneity by the I 2 value (ranging from 0% to 100%).
To evaluate the trial-level association between the treatment effects on surrogate end point and mortality, the coefficient of determination (R 2 ) was used.
We estimated 95% confidence interval (CI) of R 2 using a formula implemented in R package Surrogate. 32,33e R 2 takes a value from 0 to 1 and the R 2 0 means the absence of surrogacy and 1 indicates perfect marker of surrogacy.We used a cut-off with a lower limit of 95% CI of R 2 >0.72 as a strong correlation and with an upper limit of 95% CI of R 2 <0.50 as a weak correlation. 34ter conducting the analysis overall, we performed the following prespecified subgroup analysis: 1) bleeding outcomes defined only with BARC bleeding criteria; 2) bleeding outcomes defined only with TIMI bleeding criteria; 3) the first year of patient enrollment (the era of the trial, before 2010 vs after 2011) according to the previous article; 25 4) follow-up duration (>12 months vs #12 months); 5) trials in East Asia or non-East-Asia; 6) patients with acute coronary syndrome; and 7) different cutoff points of the following criteria; R 2 $0.7 as strong, between 0.5 and 0.69 as moderate, and <0.5 as a weak correlation. 31 performed a meta-analysis of OR or HR extracted from enrolled article, using R software (R.4.1.2) with metafor and Surrogacy package.

RESULTS
We identified 48 trials that fulfilled inclusion criteria (Figure 1).Supplemental Table 4 provides characteristics of the 48 RCTs.Supplemental Table 5 shows the assessment of the risk of bias.I 2 of odds ratio was 20.4%, 0.5 %, 90.7%, and 75.1% for all-cause mortality, cardiovascular mortality, trial defined major bleeding, and trial defined major/minor bleeding, respectively.Funnel plots is shown in Supplemental Figures 1-4.ORs of included studies of trial defined major/minor bleeding and major bleeding, and allcause mortality is shown in Supplemental Figures 5   and 6.By the study period which we categorized by the first patient enrollment, 23  (Supplemental Table 6).
Similarly, we plotted 1) the log-ORs for trial defined major bleeding and those for all-cause mortality in 44 RCTs (53 comparisons) (Central Illustration C); and 2) the log-ORs for trial defined major bleeding and those for cardiovascular mortality in 41 RCTs (49 comparisons) (Central Illustration D).The slope of the regression was 0.07 (95% CI: À0.04 to 0.18) in the regression of log-OR for all-cause mortality on log-OR for trial defined major bleeding and 0.03 (95% CI: À0.09 to 0.15) in the regression of log-OR for cardiovascular mortality on log-OR for trial defined major bleeding (Table 1).Both of trial-level coefficients of determination R 2 did not reach the predefined threshold of 0.70: 0.03 (95% CI: 0.00-0.13)between trial defined major bleeding and all-cause mortality and 0.01 (95% CI: 0.00-0.05)between trial defined major bleeding and cardiovascular mortality.In the analysis limiting RCTs to those which reported HR, we observed similar trial-level coefficients of determination R 2 (Supplemental Table 6).
In the subgroup analyses where we performed analyses for RCTs using BARC or TIMI bleeding criteria, the results remain similar (   0.98 (95% CI: 0.92-1.00),respectively, as strong correlations although there were only 4 trials included (Table 2 and Supplemental Table 8).
When we defined the different cutoff of R 2 $0.70 as strong, between 0.50 and 0.69 as moderate, and <0.50 as a weak correlation, all of the main analyses were assessed as weak correlations.

DISCUSSION
In our meta-analysis, we demonstrated that the correlation of log-OR between trial defined major/minor or major bleeding and all-cause mortality in 48 RCTs (58 comparisons) were very low (0.09 and 0.03, respectively), implying that a bleeding outcome is not a valid surrogate for all-cause mortality in trials investigating antithrombotic agents for patients with CAD.Our results were consistent regardless of bleeding definitions, study period, follow-up period, and in patients with acute coronary syndrome.
Bleeding events are considered to be associated with death, especially with bleeding-related death that occurs within 30 days of the bleeding events. 17,24,35In comparison, bleeding events are typically not related to non-bleeding-related death. 24Our data investigating trial-level correlations between nonfatal bleeding events and mortality showed that a bleeding event cannot be a surrogate for death during the whole study period although bleeding events were associated with death shortly after the bleeding events. 24These findings should be informative for researchers who are considering using bleeding outcomes as a study end point to be a surrogate for death as well as to combine bleeding outcomes and death as NACE.
Preferably, alternative analyses such as the win ratio analysis can be used to take into account for the priorities of the composite outcomes. 36In the TRILOGY-ACS and GLOBAL-LEADERS trials, trialists mentioned that various methods including the win ratio analysis should be considered to assess patients' outcomes. 37,389][20] The TICO trial investigating short-term DAPT followed by ticagrelor vs ticagrelorbased 12-month DAPT for patients with acute coronary syndrome used NACE as the primary end point, which showed no significant difference in NACE between the 2 groups, but it was associated with higher bleeding events. 20Since NACE was even between the 2 groups as the primary end point, despite significantly different bleeding as the secondary outcome, the conclusion of the trial would be ambiguous.In general, physicians should select antithrombotic agents based on individualized ischemic and bleeding risks 39 ; however, NACE as the primary end point may be inconclusive regarding which antithrombotic agents should be used for each patient.In addition, NACE has a risk of bias toward the null since thrombotic and bleeding outcomes often tend to go in opposite directions.Moreover, a composite outcome with a different magnitude of components may result in misleading impressions of the impact of treatments. 40Therefore, in isolation, NACE may not be an optimal trial primary endpoint since our data suggest bleeding events cannot be a surrogate for death.
Our study has several limitations.First, patientlevel data were not available, and our study only included data available from the original papers and we could not assess HR in all enrolled studies.
However, measures of correlation such as R 2 or the slope of the regression the estimates of mortality on those of HF were similar among the analysis of log-ORs and log-HRs.Second, each study used a different definition of trial defined major/minor bleeding or major bleeding.Regardless, we also performed a subgroup analysis of BARC and TIMI bleeding criteria which remained similar to the analysis of trial defined bleeding outcomes.However, a subgroup of all-cause mortality and TIMI major/ minor bleeding with follow-up period >12 months tend to have a correlation, which was the only exception in our results.Longer duration of the study with the use of TIMI major/minor bleeding could be a surrogate for mortality of patients with CAD.Nonetheless, the recent DAPT trials compare short DAPT vs standard DAPT as 12 months DAPT, which may not require long duration of follow-up >12 months because 12 months may be felt to be long enough to investigate the differences between short vs standard DAPT. 5,41Thus, less than one-half of the included trials did not have follow-up period >12 months, which can be a potential limitation of our study.
Third, the relationship between the risk of bleeding and death may vary by race and region, especially in the East Asian population. 13We showed a potential correlation between TIMI major or minor bleeding and mortality among them; however, it should be interpreted cautiously because there were only 4 trials included in these analyses.Fourth, we investigated whether the correlation was linear or not.We could not exclude the possibility that the correlation is nonlinear.Finally, we showed bleeding events were not a surrogate for mortality; however, this does not mean bleeding events are not important for patients with CAD.

CONCLUSIONS
Trial defined major or major/minor bleeding events may not be a valid surrogate for mortality.Assessments such as the win ratio may be preferable to NACE as the primary outcome in trials investigating antithrombotic agents for patients with CAD.

FIGURE 1
FIGURE 1 PRISMA Flowchart CENTRAL ILLUSTRATION Correlations of Treatment Effects on Trial Defined Bleeding Outcomes and All-Cause or Cardiovascular Mortality Kuno T, et al.JACC Adv.2023;2(3):100276.(A) The association between the logarithm of the odds ratio (log OR) for trial defined major/minor bleeding (surrogate) and all-cause mortality.(B) The association between the logarithm of the odds ratio (log OR) for trial defined major/minor bleeding (surrogate) and cardiovascular mortality.(C) The association between the logarithm of the odds ratio (log OR) for trial defined major bleeding (surrogate) and all-cause mortality.(D) The association between the logarithm of the odds ratio (log OR) for trial defined major bleeding (surrogate) and cardiovascular mortality.Circle sizes are proportionate to the number of patients' number (a total of intervention group and control group in each study).

TABLE 1
Overall Analyses of the Correlation of Treatment Effects and Coefficient of Determination of Bleeding for All-Cause and Cardiovascular Mortality (OR) BARC ¼ Bleeding Academic Research Consortium; CV ¼ cardiovascular; RCT ¼ randomized controlled trial; TIMI ¼ Thrombolysis In Myocardial Infarction.cause and cardiovascular mortality, did not reach significant trial-level coefficients of determination R 2 as well as patients with acute coronary syndrome (Table2, Supplemental Tables7 to 14).One subgroup investigating the correlation of all-cause mortality and TIMI major or minor bleeding with 0.89 (95% CI: 0.67-1.00),whichdidnotmeet the criteria of lower 95% CI as 0.72 as a significant correlation (Table2, Supplemental Table11).However, all-cause and cardiovascular mortality with TIMI major or minor bleeding among trials in East Asia showed R 2 of 0.96 (95% CI: 0.81-1.00)and

TABLE 2
All-Cause Mortality Major or Minor Bleeding Subanalysis Among Subgroups (OR) BARC ¼ Bleeding Academic Research Consortium; CV ¼ cardiovascular; RCT ¼ randomized controlled trial; TIMI ¼ Thrombolysis In Myocardial Infarction.