On primary composite endpoints in cardiovascular clinical trials
*Nancy L. Geller, NHLBI 

Keywords: composite endpoints, time to first of several events, primary endpoint selection in clinical trials

A composite endpoint is commonly used as the primary endpoint in cardiovascular clinical trials. This is driven by a belief is that the intervention will affect the likelihood of several negative events, often including cause-specific or overall survival. For concreteness, consider a trial of an intervention compared with placebo. In cardiovascular outcome trials in diabetes, the primary endpoint is frequently time from randomization to non-fatal myocardial infarction, non-fatal stroke or cardiovascular death. In a drug (versus placebo) trial in patients who already have an implanted cardiac defibriIlator to “shock” irregular electrical activity of the heart (ventricular tachycardia, VT), the primary outcome could be time to death or shock for VT. There is an advantage in using a composite when the belief is that the intervention will show an improvement on all of the events included in the composite; that is, for fixed power, the sample size needed to detect a given difference in the proportion of composite events in the two arms will be smaller than for a trial whose primary endpoint is any single component of the composite (assuming that hypothesis testing is undertaken at the same significance level). The usual procedure is to report further analyses of the individual components of the composite. Thus those who experienced a non-fatal event will continue to be followed so that their survival can be assessed. Of course power is consequently compromised for the components. Nonetheless, when the primary analysis is statistically significant and all of its components go in the same direction as the primary endpoint (although not necessarily statistically significant), the conclusion is that there is a treatment difference. However a disadvantage of a composite endpoint is that the trial’s primary results will be dominated by the most common of the outcomes comprising the composite. Frequently that is the least serious of the events, e.g., shock for VT as compared to death. A consequence is that even with a statistically significant difference in the composite, at least one of its components may go the “wrong” direction. Trial interpretation is then more difficult. We examine examples of trials with composite endpoints and their advantages and disadvantages as well as a method from the literature (Finkelstein and Schoenfeld, Statistics in Medicine, 1999) which we recommend, perhaps as a secondary analysis. This method orders the events of the composite (worst to best) and combines them to test whether there is an effect in any of the events in the composite.