|
Abstract:
|
In randomized clinical trials, adjustment for chance imbalances in prognostic baseline variables can lead to improved precision in estimating the average treatment effect. It has been shown, using asymptotic theory, that appropriate adjustment decreases the estimator variance (unconditionally) as well as bias (conditional on the observed imbalance). Our contribution is to give a concrete demonstration of these theoretical results using data from a completed trial of a new surgical intervention for Alzheimer's disease prevention. Specifically, we show how to calculate the precision gain and conditional bias reduction using data from this trial, and provide visualizations of how the conditional bias reduction due to adjustment leads directly to a gain in unconditional precision. We also show how missing data, treatment effect heterogeneity and model misspecification impact the gains from such adjustment.
|
