Over 20 years of empirical data document the individual and societal costs of co-occurring substance use disorders (SUD) and posttraumatic stress disorder (PTSD). Although evidence-based treatments are available for SUD+PTSD, the lack of methods or guidelines for personalized treatment recommendations represents a critical barrier to progress in the field. Secondary analyses of existing treatment research datasets have the promise to identify cost-effective and noninvasive variables that represent predictors, moderators, mediators, mechanisms of action, and secondary effects in a way that can be easily replicated in research and expanded to the real world. Importantly, extant datasets provide an opportunity to develop and test innovative analytic approaches that leverage technological advances in computing power to: 1) develop precision medicine algorithms that can prognosticate and prescribe at the patient-level, 2) generate new hypotheses to guide future research on treatment development and dissemination, and 3) innovate existing methods that can be applied to solve related problems in the field. Accordingly, the proposed project will apply machine learning methodologies to the largest existing NIDA-funded multi-site clinical trial for SUD+PTSD in order to:
Aim 1 : Develop algorithms that can identify the probability of treatment response on both SUD and PTSD outcomes based on routinely collected intake data.
Aim 2 : Identify key variables associated with a significant increase or decrease in the treatment response probability.
Aim 3 : Assess the feasibility of providing data-driven, personalized prescriptions that fit patients to the treatment most likely to benefit them based on their individual profile. The proposed machine learning analyses will be developed by the applicant at the Anxiety and Health Behaviors Laboratory of The University of Texas at Austin, and evaluated at the Texas Advanced Computing Center (TACC), which houses the fastest academic supercomputer in the United States of America, and one of the most powerful systems in the world. Importantly, the algorithms (once developed) can be easily disseminated and do not require powerful hardware to execute. Therefore, the combination of advanced machine learning analyses and high-performance computing has the promise to provide methods and findings that can be expanded across extant datasets and future research projects to facilitate scientific discovery in the field of personalized medicine for addiction that: 1) goes beyond comparing mean response across treatments, and 2) that is otherwise not possible with traditional statistical methods. If successful, the project will support a strategy that represents a paradigm shift from one-size-fits-all recommendations derived from treatment comparisons in clinical trials, to a data-driven, precision medicine approach that prognosticates and prescribes at the patient level.
The proposed project will apply machine learning methodologies to the largest existing NIDA-funded multi-site clinical trial for substance use disorders (SUD) and posttraumatic stress disorder (PTSD). If the project is successful, it will be the first to: 1) develop algorithms that can identify the probability of treatment response on both SUD and PTSD outcomes based on routinely collected intake data, 2) identify key variables associated with a significant increase or decrease in the probability for treatment response, and 3) assess the feasibility of providing data-driven prescriptions that fit patients to the treatment most likely to benefit them based on their individual profile.
