|
Abstract:
|
Selecting a statistically sensitive outcome is very attractive in clinical trial design, reducing the required sample size thereby saving time and cost; however, the choice needs to be justified for testing efficiency and clinical relevance. In practice, a success or failure event associated with an individual study participant is often used to assess treatment efficacy during a follow-up period. The primary endpoint could be either binary or time-to-event and possible analysis approaches are logistic regression and a Cox proportional hazards model, respectively. Instinctively, the Cox model is perhaps more efficient by taking into account an individual's exact time of the event, which may not always be true as in the case of clinical trials with time of event recorded by intervals or truncated at a pre-defined landmark time. Logistic regression may be more appropriate for such conditional distributions of survival data. In this report, the testing efficiency is numerically assessed for both binary and survival data by simulating various data scenarios, which could help to select the more sensitive outcome in a trial design. Results are also illustrated using clinical data.
|
