Determining the most appropriate components for a composite clinical trial outcome

Background Because composite end points augment event rates, they are often thought to increase statistical power. This may not be true if the intervention has a lesser effect on some components of the composite. Consequently, treatment effect size may depend on the choice of composite. Methods To explore this issue, we performed a meto-onalysis to determine the effect of angiotensin-converting enzyme inhibitors or angiotensin receptor blockers on individual cardiovascular (CV) outcomes. We then applied these treatment effects to 2 different composite CV outcomes generated using blinded data from the ongoing Nateglinide and Valsortan in Impaired Glucose Tolerance (IGT) Outcomes Research (NAVIGATOR) trial and analyzed them on a time-to-first-event basis. The composites were CV death, nonfatal myocardial infarction, nonfatal stroke, or hospitalization for heart failure and an "extended" composite that included hospitalization for angina and coronary revascularization. Results The odds reductions due to angiotensin-converting enzyme/angiotensin receptor blocker treatment estimated from the meta-analysis were as follows: CV death: 13%, P < .0001; nonfatal myocardial infarction: 16%, P < .00001; nonfatal stroke: 14%, P = .006; heart failure: 28%, P < .00001; hospitalization for angina: 7%, P = .02; and revascularization: 5%, P = .17. For the CV composites, the projected odds reduction was larger (17.8%, 95% CI 0.452-1.189) for the narrower composite compared with the extended CV composite (11.7%, 95% CI 0.623-1.136); that is, use of the extended composite reduced power to detect a difference between treatment groups. Conclusions Although the use of CV composites augments event rates, it may not increase statistical power. Inclusion of events little influenced by an intervention may reduce the precision of the composite end point and mask treatment effects.