Machine learning-based imaging biomarkers improve statistical power in clinical trials

Radiomic models, which leverage complex imaging patterns and machine learning, are increasingly accurate in predicting patient response to treatment and clinical outcome on an individual patient basis. In this work, we show that this predictive power can be utilized in clinical trials to significantly increase statistical power to detect treatment effects or reduce the sample size required to achieve a given power. Akin to the historical control paradigm, we propose to utilize a radiomic prediction model to generate a pseudo-control sample for each individual in the trial of interest. We then incorporate these pseudo-controls into the analysis of the clinical trial of interest using classical and well established statistical tools, and investigate statistical power. Effectively, this approach utilizes each individual’s radiomics-based predictor of outcome for comparison with the actual outcome, potentially increasing statistical power considerably, depending on the accuracy of the predictor. In simulations of treatment effects based on real radiomic predictive models from brain cancer and prodromal Alzheimer’s Disease, we show that this methodology can decrease the required sample sizes by as much as a half, depending on the strength of the radiomic predictor. We further find that this method is most helpful when treatment effect sizes are small and that power grows with the accuracy of radiomic prediction.