All Times EDT
Keywords: randomized trial analysis
It is well known that adjustment for covariates can improve power in randomized clinical trials, but in a new disease, the best prognostic covariates may not be known. In adaptive regression, covariates are not pre-specified, but are selected based on their ability to predict the outcome in the current trial. One promising method of analyzing results from such a trial is a re-randomization test that fixes data at their observed values and re-randomizes according to the trial’s randomization scheme to generate a reference distribution for the test statistic. This talk compares different valid methods of adaptive regression analyzed using a re-randomization test. Are there extreme settings under which these methods break down? What is the proper conclusion following a statistically significant result? These are some of the questions answered in this talk.