Leveraging harmonized data to improve external validity and efficiency of clinical trials for treating opioid use disorder

Randomized clinical trials for the treatment of psychiatric and substance use disorders can be difficult, time- consuming, and expensive to conduct. Partially as a consequence, their sample sizes are typically underpow- ered for: 1) detecting moderately sized average treatment effects (ATEs) that may nonetheless be important for health at the population level, and 2) learning optimal individualized treatment rules (i.e., rules that match treatments to individuals based on demographic and clinical characteristics to optimize outcomes of interest), which are the cornerstone of personalized medicine. These are major barriers to 1) improving population drug use disorder outcomes in terms of broad implementation of treatments with moderate effects and 2) improving individual drug use disorder outcomes in terms of refining treatment strategies away from a “one- size-fits-all” approach to one incorporating a patient’s clinical characteristics. Data fusion typically involves combining individual patient data from similar studies to improve statistical power and answer questions that cannot be addressed by a single study alone. However, the statistical theory underlying data fusion is rel- atively new, and numerous open problems remain. One commonly occurring challenge is that studies will often measure different outcomes at follow-up such that a given outcome of interest will only be observed in a subset of studies. In practice, trials are typically combined when they have a common set of covariates, treatment, and outcome. This means that statistical efficiency and power gains from incorporating data from additional studies that do not share a common outcome, but do measure other related outcomes, are left on the table. Consequently, there is a critical need for principled, flexible, and efficient estimators that can be applied to multi-study, multi-outcome fused data sets to maximize statistical power and efficiency. Doing so is a prerequisite for learning individualized treatment rules that optimize the health outcomes relevant to each treatment and for detecting more moderately sized, yet meaningful treatment effects. The objectives of this project are: 1) to develop both simple parameteric and semiparametric efficient estimators of the ATE, conditional ATE (CATE) and optimal individualized treatment rules in multi-study, multi-outcome data-fusion settings (Aims 1a and 2a); 2) to develop sensitivity analyses for ATE and CATE estimation if the untestable transport identification assumption is not met (Aims 1b and 2b); 3) to apply these estimators to harmonized medication for the treatment of opioid use disorder (MOUD) trials with multiple outcomes to increase power to detect effects (Aims 1c and 2c); and 4) to develop and disseminate user-friendly software that implements all the above methods (Aim 3:). This proposal is expected to make a significant contribution to increasing the power and information gained from data fusion, and consequently, to improve the individualized treatment of psychiatric and substance use disorders.