Advancing Individualized Treatment Rules from Multiple Studies

This project leverages cutting-edge techniques from meta-analysis, integrative learning, and transfer learning to develop more reliable and robust individualized treatment rules (ITRs).

We develop a novel framework that integrates data from multiple randomized controlled trials (RCTs), each sharing a common treatment arm but varying in alternative treatments. The method adaptively weights information across studies by minimizing a regularized misclassification risk, demonstrating improved estimation of value and benefit functions.

Currently, we are developing methods to learn ITRs from a network of RCTs and to implement ITRs with distribution-free adverse risk control and conditional coverage guarantees. We are also working on approaches that combine meta-analysis and federated learning to further enhance ITR estimation.