|
Abstract:
|
In randomized cancer clinical trials where disease progression occurs, administering second-line therapy and third-line therapy are sometimes allowed in the protocol. When analyzing survival data, intention-to-treat analysis is often conducted. On the other hand, with the aim of estimating the effect of first-line therapy on longitudinal data such as quality of life, the approach often implemented is on treatment analysis: the analysis conditions on the data measured during first-line therapy, and survival. However, such conditioning causes well-known post-treatment bias. In this presentation, we propose the general framework to estimate the survivor average causal effect of treatment regime that depends on the time-dependent covariates (i.e. dynamic treatment regime; DTR), using the principal stratification framework. Specifically, we model and predict the distribution of the time-dependent covariates, and the survival probability if the patient had followed another DTR, finally analysis with individual-specific weight is conducted. The results of simulation studies will be presented.
|
