MARGO: Machine Learning-Assisted Adaptive Randomization for Group Sequential Trials Based on Overlap Weights

Adaptive randomization is a promising approach in clinical trials that aims to optimize patient outcomes by adjusting treatment allocation probabilities based on accumulating data. However, implementing adaptive randomization in group sequential trials, which include interim analyses for early stopping decisions, poses significant challenges, such as managing type I error inflation and ensuring robust statistical validity. This paper proposes Machine learning-assisted Adaptive Randomization for Group sequential trials based on Overlap weights (MARGO) as an innovative solution to these challenges. MARGO integrates machine learning (ML) models into the adaptive randomization process, allowing dynamic updates to randomization probabilities based on real-time predictions of treatment success. To control the overall type I error rate due to the covariate imbalance in group sequential trials, MARGO utilizes overlap weights (OW), which are employed to balance covariates across treatment groups, minimizing confounding and ensuring that the comparison between treatments remains unbiased. In our implementation, various ML algorithms are evaluated for their effectiveness in predicting treatment outcomes. Through extensive simulation studies, we demonstrate that MARGO not only enhances the flexibility and efficiency of group sequential trials but also maintains statistical rigor by effectively controlling type I error rates. Our results show that MARGO provides a more ethical and data-driven approach to patient allocation, potentially improving treatment success rates while preserving the integrity of the trial.