|
Abstract:
|
In clinical studies it is common to perform a responder analysis to compare two treatment groups. In general, this analysis consists in testing treatment effects difference using proportions of subjects showing a response, where the criterion for response is mainly based on an arbitrary cut-off. Instead of a single cut-off, the response definition can be just as well defined as a continuum of cut-offs and thereby resulting in a cumulative distribution of response, also called responder curve. Methods like log-rank, Peto-Peto, Harrington-Fleming and Kolmogorov-Smirnov nonparametric tests can be utilized towards a comparative analysis of the responder curves corresponding to different treatment groups. The performance of all the aforementioned tests will be assessed by their power, estimated using simulations under various dropout models.
|
