|
Abstract:
|
The issue of treatment non-adherence is not uncommon in modern clinical trials. Especially, in oncology clinical trials, after disease progression, patients may be allowed to switch from the control arm to the experimental drug or to other existing treatments, which will introduce bias to the assessment of treatment efficacy of the experimental drug when the endpoint of interest is overall survival. Intention-to-treat (ITT) or per-protocol censoring at time of the switch analysis cannot appropriately adjust for this bias. There are some existing methods that can adjust for this bias. for example, Rank Preserving Structure Failure Time Model (RPSFTM) with G-estimation, Iterative Parameter Estimation (IPE) and Two-stage method. As such issue is expected to arise more often in future oncology trials, it is important to understand and appropriately apply these statistical methods. We have compared the performance of several existing methods via simulated oncology clinical trial data. We will present the concept and algorithm of each method, details of simulation design and results, and recommendations regarding how to implement these methods in real world clinical trials.
|
