Application of machine learning methods in clinical trials for precision medicine

A key component for precision medicine is a good prediction algorithm for patients9 response to treatments. Machine learning methods have successfully demonstrated their superior prediction performance in many applications, but have not been applied to conduct response-adaptive randomization in clinical trials. In this article, we incorporate nine machine learning methods, such as gradient boosting machine, random forest, and artificial neural network, in clinical trial design. Realizing that no single method may fit all trials well, we also use an ensemble of these nine methods. We evaluate their performance through quantifying the benefits for trial participants, such as the percentage of patients who receive their optimal treatments and individual loss. To avoid the potential bias introduced by the adaptive scheme, we use the inverse probability of treatment weighted method to estimate the average treatment effect and the statistical power. Simulation studies show that the adoption of machine learning methods results in more personalized optimal treatment assignment and higher overall response rates among trial participants. Compared with the nine individual methods of machine learning, the ensemble approach achieves the highest response rate and assigns the largest percentage of patients to their optimal treatments. The proposed methods are applied to a real-world leukemia study.