B0530
Title: The underrecognized potential of logistic regression to analyze survival data from clinical trials
Authors: Paul Blanche - University of Copenhagen (Denmark) [presenting]
Thomas Scheike - Section of Biostatistics University of Copenhagen (Denmark)
Abstract: Guidelines from regulatory agencies have long emphasized the benefit of adjusting for baselines variables in the analysis of randomized clinical trials. Recently, they put forward the specific use of G-computation based on logistic regression for analyzing binary outcomes. This approach is more powerful than an unadjusted analysis and it provides a robust estimator of the average treatment effect which is consistent under arbitrary model misspecification. We argue that when the main outcome of a clinical trial is t-year survival or similar, a similar approach can be used, even when lost-of follow-up occurs and creates right-censored data. We show that under mild conditions, the inverse probability of censoring weighting provides similar robust inference as in the case of binary outcomes. As compared to alternatives based on hazards regression, the estimand of this approach is clearly defined and does not rely on arbitrary model assumptions. The approach is straightforward to implement with standard software and provides a simple and transparent approach to leverage the information in baseline variables.
