|
Abstract:
|
It often happens in cancer clinical trials that a surrogate outcome rather than overall survival is used to compare two randomized treatments, for the practical purpose of shortening the duration of the trial and to avoid contamination, by subsequent nonrandomized treatment, of the treatment effect under study in the RCT. However, overall survival is nevertheless often of primary interest. An ensuing difficulty is that subsequent to the assessment of the surrogate outcome, patients enroll in a non-prescribed fashion into further trials or may be given additional treatment as part of routine practice. Thus it becomes exceedingly hard to disentangle the effect of the randomized treatment from that of the subsequent, nonrandomized treatments. We describe and apply a method of estimating optimal dynamic treatment regimes in attempt to solve this problem in an analysis of COG study P9462.
|
