Abstract 4587: Weighted Composite Endpoint Analysis of TRITON-TIMI 38: Disconnect Between Analytical Equivalence and Clinical Importance
BACKGROUND: Composite endpoints (CE) are frequently used in clinical trials for parsimonious summarization of treatment effects and for improving trial efficiency. However, a fundamental limitation of CE is the assumption that each component is statistically equivalent despite obvious distinction in clinical importance.
OBJECTIVE: To perform a “weighted” analysis of the combined efficacy plus safety “quadruple” CE utilized in TRITON-TIMI 38 trial that compared prasugrel with clopidogrel in ACS patients scheduled for PCI.
METHODS: A hierarchical (excluding double counting) CE analysis was performed using all-cause death, nonfatal MI, nonfatal stroke and TIMI Major or TIMI Major plus Minor bleeding. The results based on statistically-equivalent weights (1.0 for all endpoints) were compared with those utilizing clinically-relevant weights based on case-fatality rates (CFR) reported in TRITON: death = 1.0, MI = 0.06 (CFR of 74/1169 = 6%), bleeding = 0.05 (CFR of 27/534 = 5%); stroke = weights from 0.1 to 1.0 were used as CFR were not reported.
RESULTS (Table⇓): The data show that a statistically significant difference in the weighted CE is only observed with the use of statistically-equivalent, but not clinically-relevant weights. A sensitivity analysis demonstrated that the difference in CE achieves significance in favor of prasugrel at a MI weight of at least 0.7 using TIMI major bleeding (0.90, 95% CI 0.82–1.00; P=0.046) and 0.95 using TIMI major + minor bleeding (0.92, 95% CI 0.84 –1.00; P=0.046) in the CE analysis.
CONCLUSIONS: The benefit of prasugrel in TRITON is primarily driven by nonfatal MI under the assumption that MI is nearly as important as death, arguably an implausible clinical conjecture. Utility ranking or weighted schemes may offer a viable approach to combining endpoints of different clinical importance and provide a potential solution against interpretive errors with composite endpoint results.