Integrative Learning to Combine Individualized Treatment Rules from Multiple Randomized Trials

Using individualized treatment rules (ITRs) to tailor treatment to individual patient’s characteristics and intermediate responses holds promise to improve treatment response of chronic disorders. However, several barriers, in particular, lack of generalizability or reproducibility of ITRs derived from a single study and lack of power to detect treatment modifiers as tailoring variables, pose serious challenges for implementing ITRs in clinical practices. In this work, we propose a novel integrative learning method to combine evidence from multiple clinical trials to yield an integrative ITR that improves both precision and reproducibility. Since subject-specific covariates available in each trial may differ, ITRs learned from each study can depend on a different resolution of patient-specific characteristics. Our method does not require all studies to use the same set of covariates, and thus allows study-specific ITRs to be transferable across studies. Specifically, to transfer information we propose integrative learning to enhance a high-resolution ITR by borrowing information from coarsened ITRs or vice versa. We conduct extensive simulation studies to show that the integrative learning has improved performance compared to using single study data. We apply the developed method to multiple clinical trials of major depressive disorder (MDD) and other co-morbid mental disorders. We demonstrate that the integrative ITR yields a greater benefit and has improved precision compared to single-study ITRs or non-personalized rules for treating major depression.