Efficiency Comparison of Analysis Methods for Recurrent Event and Time-to-First Event Endpoints in the Presence of Terminal Events—Application to Clinical Trials in Chronic Heart Failure

Abstract

In clinical trials investigating treatments for chronic heart failure (CHF), a standard primary endpoint is the time to the first composite of hospitalization for heart failure or cardiovascular death (CVD). Since many patients experience several hospitalizations, there is interest in including recurrent hospitalizations into the primary endpoint to better capture disease burden. Several analysis methods have been proposed for recurrent event endpoints, mostly for the situation without a terminal event such as CVD. Only some methods explicitly account for terminal events, for example, the joint frailty model. We compare the power and Type I error rate of recurrent event methods with those of time-to-first event methods in the presence of terminal events. A special focus is the situation where treatment affects the risk of CVD. Our investigations are based on a simulation study, for which the scenarios are motivated by CHF trials, and based on bootstraping data from the Val-HeFT and PARADIGM-HF trials. We find that recurrent event methods can in many situations increase power considerably. But this is not always the case, for example, when there is high probability of treatment discontinuation after the first hospitalization. Also, for both recurrent and time-to-first event methods, the Type I error rate can be inflated in case of a detrimental treatment effect on CVD. Based on the simulation results we give recommendations on the choice of endpoint and analysis method for CHF trials.

Keywords:

• Chronic heart failure
• Death
• Joint frailty model
• Power comparison
• Recurrent events

Supplementary Materials

The supplementary material provides further details and simulation results for the quasi-Poisson model with Fletcher’s method as well as detailed results for the simulation study variations described in Section 3.3.

Acknowledgments

For their helpful comments on a draft qualification opinion request, the authors would like to thank: Bruce Binkowitz, Richard Cook, Dieter Häring, Björn Holzhauer, Günther Müller-Velten and Jennifer Rogers. For support with the implementation of the simulation study within the Biostatistics Innovation Center (BIC) at Bayer thanks also to Katja Brandau, Madhurima Majumder and Alexander Seipp. In addition, we thank the two referees for their helpful suggestions.