|
Abstract:
|
Interest in the restricted mean survival time (RMST) as an index for comparing treatment arms in a clinical trial has increased recently. RMST is an interpretable measure (average survival time from 0 to a specified time point) and group comparisons do not rely on the proportional hazards assumption (Karrison, 1987). The question addressed in this talk is the extent to which adjustment for covariates in randomized clinical trials comparing RMSTs provides improvement in the precision of the estimated treatment effect. Does comparison of RMSTs with adjustment for covariates lead to a reduction in the standard error compared to the unadjusted estimator as in ordinary multiple regression? Or does it behave in a manner similar to the logistic and Cox regression models, i.e., provides no improvement in precision. Some simulations by Chen and Tsiatis (2001) suggest that precision will be increased if covariates are judiciously chosen. We present the results of more extensive simulation studies comparing unadjusted vs. adjusted comparisons of RMST in the setting of randomized clinical trials. Both subject-specific effects and population-averaged effects will be considered.
|
