gms | German Medical Science

Introduction: Many clinical trials focus on the comparison of the treatment effect between two or more groups concerning a rarely occurring event. In this situation, showing a relevant effect with an acceptable power requires the observation of a large number of patients over a long period of time. For feasibility issues, it is therefore often considered to include several event types of interest, non-fatal or fatal, and to combine them within a composite endpoint. Commonly, a composite endpoint is analyzed with standard survival analysis techniques by assessing the time to the first occurring event. This approach neglects that an individual can experience more than one event which leads to a loss of information. As an alternative, composite endpoints could be analyzed by models for recurrent events. There exists a number of such models, e.g. regression models based on count data or Cox-based models such as the approaches of Andersen and Gill [1], Prentice, Williams and Peterson [2] or Wei, Lin and Weissfeld [3]. Although some of the methods were already systematically compared within the literature [4], [5], [6] there exists no systematic investigation for the special requirements regarding composite endpoints.

Methods: The focus lies on a comparison between the common Anderson-Gill model, the models by Prentice, Williams and Peterson and the model from Wei, Lin and Weissfeld. Different data settings with one recurrent non-fatal event and a possible dependent fatal event are investigated. The comparison is based on the statistical properties of the models’ treatment effect estimator and its correct interpretation, on the underlying model assumptions and on the robustness under deviations from these assumptions. The aim is to deduce recommendations for the choice of an appropriate analysis strategy which addresses the specific data structure of clinical trials with composite endpoints. This structure is characterized as 1. an increase in the baseline instantaneous risk for a subsequent event after a previous event, the dependence of the instantaneous risk on the time of the previous event, a change in the treatment effect (Hazard ratio) after an event occurrence, and differing treatment effects for the different event types. The performance properties of the models will be investigated using Monte-Carlo simulations based on realistic clinical trial settings.

Results: We demonstrate that all models estimate different treatment effects which can considerably deviate under commonly met data scenarios and which correspond to mixed net effects of the individual component effects [7] for our data situations because they neglect the order of events.

The approach by Prentice, Williams and Peterson delivered adequate results for all simulated data scenarios and turned out to be better in terms of interpretability, model robustness and power than the other models in the context of composite endpoints. However, the number of events per individual incorporated in the analysis should be limited.

Die Autoren geben an, dass kein Interessenkonflikt besteht.