Choice of treatment-effect measures when noninferiority margins originally defined in absolute difference translated into relative difference influenced the results of clinical trials.

OBJECTIVES

The aim of this study was to investigate the use of three treatment-effect measures in the analysis of randomized trials using a time-to-event endpoint and assess their influence on the results.

STUDY DESIGN AND SETTING

A recent equivalence trial showed discordant results with the use of different measures. Different hypotheses may explain such discordant results including a mistaken hypothesized distribution of time to failure and an overestimation of failure rates in the protocol. In a simulation study, we investigate different situations in comparing analyses based hazard ratio (HR), risk difference estimated by Kaplan-Meier curves, and difference in restricted mean survival time. We also compared these three analyses on genuine data from a recent equivalence trial.

RESULTS

In the equivalence trial, two analyses would have concluded equivalence, whereas the original analysis based on HR estimate did not declare equivalence. Results of our simulation study indicate little to moderate differences between the three analyses when the true distribution of time to failure is different to the hypothesized distribution. The main discordant results are found when failure rates have been overestimated or underestimated regardless of the distribution.

CONCLUSION

With the distributions investigated, differences between analyses based on different measures are much more driven by mistaken hypothesized failure rates than by the shape of the distribution of time to failure.