gms | German Medical Science

Introduction: In clinical trials for chronic heart failure (CHF), time to a composite of first hospitalisation for worsening heart failure (HFH) or death is a widely accepted primary efficacy measure. Motivated by lower event rates observed in CHF trials, there are proposals to use recurrent HFH (alone or plus death as an additional event) as the primary endpoint to improve trial efficiency [1]. However, analysis and interpretation of recurrent event endpoints may be complicated by the terminal event death. Interpretation of superiority for a recurrent event endpoint is straightforward, only, if a detrimental effect on all-cause mortality is excluded firstly.

Methods: A simulation study was conducted to compare four different primary efficacy assessment strategies (statistical analysis models): I. superiority for time to HFH or death, whichever occurs first (Cox proportional hazards model); II. superiority for recurrent HFH plus death (negative binomial (NB) or Lin-Wei-Yang-Ying (LWYY) model); III. two-stage testing procedure: non-inferiority for death with pre-specified margin in the first stage (difference in survival probability or restricted mean survival time (RMST) at a pre-specified time point) and then superiority for recurrent HFH plus death in the second stage (NB or LWYY model); IV. co-primary endpoints: non-inferiority for death with pre-specified margin and superiority for recurrent HFH (joint frailty model with pre-specified fallback strategy in case the model does not converge). Data were generated according to joint frailty models to account for intra-subject correlation. The simulation settings covered various scenarios in which the investigational treatment has a null, positive or negative effect on recurrent HFH and death as compared to the control treatment. The mean effect estimate and the one-sided rejection probability (i.e. the probability of concluding trial success) were calculated.

Results: In the scenarios where the investigational treatment has a null or negative effect on both death and recurrent HFH, the rejection probability was well-controlled at or below 2.5% for Strategy I and Strategy II using the NB model. In the scenarios where the investigational treatment has a negative effect on death but a positive effect on recurrent HFH, Strategy I had a much lower rejection probability as compared to Strategy II, and Strategy II using the LWYY model even had a moderate to high rejection probability if there was an excessive increase in mortality in the investigational treatment group. In contrast, the rejection probability was generally controlled at or below 2.5% for Strategy III and IV if a detrimental effect on mortality could not be excluded based on the pre-defined non-inferiority margin.

Discussion: Using a recurrent event endpoint as the primary endpoint may conceal a detrimental effect on mortality. A confirmatory non-inferiority assessment of mortality is a plausible gatekeeper to rule out treatments with an undesirable effect on mortality before demonstrating other advantages of a treatment and ensure a valid conclusion.

?Conclusion: In CHF trials using recurrent HFH (alone or plus death) as the primary efficacy endpoint, a confirmatory non-inferiority assessment of mortality should be incorporated into the primary efficacy assessment strategy.

The authors declare that they have no competing interests.

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