|Thursday, February 18|
|PS1 Poster Session 1 & Opening Mixer sponsored by SAS||
Thu, Feb 18, 5:30 PM - 7:00 PM
Demonstration of Novel Statistical Procedures to Adjust for Baseline Variables in Estimating Average Treatment Effects with Binary Responses (303245)*Elizabeth Colantuoni, The Johns Hopkins University
Michael Rosenblum, The Johns Hopkins University
Keywords: Randomized trials, prognostic baseline variables, average treatment effect
It is well known that the analysis of covariance (ANCOVA) procedure can be used to improve precision of the estimated average treatment effect within a randomized controlled trial when the outcome is continuous and a set of prognostic baseline variables are available. In cases where the outcome is binary, statistical procedures that account for prognostic baseline variables in the estimation of the average treatment effect are less understood. The common approach to apply a logistic regression model and extract the coefficient for the fixed effect of treatment does not provide an adjusted estimate of the average treatment effect and only provides an estimate of the conditional treatment effect if the model is correctly specified. However, novel estimators for the average treatment effect are available for binary outcomes that have similar properties as ANCOVA. Details of the procedure to implement one such estimator is provided using data from a completed Phase II trial of a surgical vs. medical intervention for stroke patients. Precision gains can be substantial and guidance on how to pick a potential set of prognostic baseline variables using historical data is provided.