Interim analysis of binary outcome data in clinical trials: a comparison of five estimators

ABSTRACT

In clinical trials, where the outcome of interest is the occurrence of an event over a fixed time period, estimation of the event proportion at interim analysis can form a basis for decision-making such as early trial termination, sample size re-estimation, and/or dropping inferior treatment arms. In addition to derivation of mean squared error under an exponential time-to-event distribution, we performed a simulation study to examine the performance of five estimators of the event proportion when time to the event is assessable. The simulation results showed advantages of the Kaplan–Meier estimator over others in terms of robustness, and the bias and variability of the event proportion estimate. An example was given to illustrate how the estimators affect dropping treatment arms in a multi-arm multi-stage adaptive trial. We recommended the use of the Kaplan–Meier estimator and discourage the use of other estimators that discard the inherent time-to-event information.

KEYWORDS:

• Binary outcome data
• event proportion
• bias
• mean squared error
• Kaplan–Meier
• interim analysis

Acknowledgments

The authors would like to thank the referees for their insightful comments and suggestions.