gms | German Medical Science

Introduction: To increase the number of events and thereby to increase the power of a clinical study one may combine several events of interest within a so-called composite endpoint. However, the effect of the individual components might differ, either in magnitude or even in direction, from the effect of the composite which leads to difficulties in interpretation of results. Furthermore, in clinical application the individual event types might be of different clinical relevance leading also to interpretational difficulties. The all-cause hazard ratio is the common effect measure for composite endpoints by assessing the time to the first occurring event. This effect measure is based on the same weights for the different event types irrespective of their clinical relevance. The all-cause hazard referring to the composite thereby is the sum of the unweighted event-specific hazards. However, there also exist weighted effect measures to incorporate the different relevance levels of the individual event types. Our working group recently proposed a weighted effect measure called the weighted all-cause hazard ratio as a natural extension of the standard all-cause hazard ratio [1]. Thereby the individual hazards for each component are multiplied with a predefined relevance weighting factor. The first approach to identify the cause-specific hazards of the individual components was to use the parametric Weibull survival model which directly yields a parametric point estimator. A permutation test was proposed to test the new effect measure. However, a point estimator which does not require a pre-specification of the survival model as well as a test statistic which is not based on resampling would both be appealing to ease application in practice. In this work we present a new semi-parametric point estimator for the weighted hazard ratio and a related test to overcome the current problems.

Methods: The new estimation approach and test statistic are systematically compared to the previous approach via various Monte-Carlo based data simulations.

Results: The original estimation approach for the weighted all-cause hazard ratio is sensible against misspecification of its assumptions whereas the new easier estimation approach seems quite robust against misspecifications. Furthermore, the new test statistic is comparable to the computationally expensive permutation test in terms of power and type I error control.

Discussion: A new weighted effect measure for the analysis of a composite time-to-first event endpoint as a natural extension of the common all-cause effect is introduced. This weighted effect measure allows to ease the interpretation of a composite endpoint but does not necessarily lead to a decrease in sample size. With our new point estimator and related test statistic, the proposed weighted effect measure can be easily estimated based on the non-parametric Nelson-Aalen estimator and tested with a closed-formula in various data scenarios.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.