Composite Endpoints

A composite of several types of events may serve as a single endpoint in a clinical trial. For example, in persons who test HIV-positive, a commonly used endpoint has been progression to AIDS, defined as the occurrence of any one of several AIDS-defining illnesses (7). By including several types of events, a composite endpoint increases the likelihood that a subject will experience an endpoint event and avoids the problem of multiplicity. Thus, the power of the study is increased, and fewer study subjects are needed to show an effect of a given magnitude.

Results on composite endpoints are most readily interpretable when all components of the composite have similar levels of clinical impact and when any differences in effects between study treatments occur in the same direction for all components. If these conditions are not met, a small effect on a high-clinical-impact component of the composite, such as mortality, may be outweighed in the analysis by a larger but less important effect on a relatively minor component of the composite, such as morbidity. Nonetheless, it is often not possible to satisfy these two conditions. Although the conditions are desirable, they are not requirements. Many composites include mortality or another event of singularly important clinical consequence together with less important events. Divergent effects on components may occur.

Weighting or ranking of components of the composite may be employed to compensate for the disparate clinical effect of the components, though neither weighting nor ranking is entirely satisfactory. Weighting outcomes involves subjectively assigning individual values according to each outcome's relative clinical effect. Ideally, the weighting method should be validated before use in an efficacy trial. Rank-ordering outcomes still requires some subjectivity, but often general consensus can be reached on a limited number of ranks (e.g., mortality, then survival with morbidity, then survival with no morbidity). Such ranking diminishes the likelihood that an effect on a less important outcome will outweigh an opposite effect on a more important outcome. However, ranking components only partially accounts for differences in impact; such ranking does not take into account differences in the degree of superiority of some ranks over other ranks or differences in the impact of outcomes within a rank.

Regardless of whether the components are weighted or ranked, an analysis should be undertaken showing how each component is affected by the intervention. This allows better understanding of the treatment effects, including an assessment of whether a benefit in one component has occurred at the expense of another component.

Another potential concern regarding the use of a composite endpoint is that the ability to communicate results may be impaired. It is relatively easy to communicate a finding of a statistically significant decrease in the incidence of a specific event, by giving the estimated effect size and confidence interval. In contrast, when a treatment causes a statistically significant decrease in the incidence of a composite of events, particularly with no clearly significant effect on the individual elements of the composite, it is harder to communicate the nature and extent of clinical benefits demonstrated.

In some settings, use of a composite endpoint offers the potential to diminish problems with missing data. In serious diseases, the ability of a therapy to affect the occurrence of a particular undesired event (e.g., end-stage renal failure, stroke, amputation), may be complicated substantially by the occurrence of a significant number of deaths before the endpoint is reached. By combining death with the undesired event in a composite endpoint, the patients who die would be counted as dying because of treatment failure, and a valid measure of clinical benefit may be able to be developed with fewer missing data. The EPIC trial comparing ReoPro® (abciximab), which is a platelet glycoprotein Ilb/IIIa inhibitor, and placebo in prevention of ischemic complications in patients undergoing a percutaneous intervention successfully used a composite endpoint. The components were death, myocardial infarction, or an unplanned urgent revascularization procedure for recurrent cardiac ischemia. Mortality was very low in the trial and not different between treatment and control arms, but treatment-associated reduction in the incidence of the other components of the composite led to a positive result (8,9).

0 0

Post a comment