Using Multimodal Data Science To Gain Actionable Insights For Substance Misuse Prevention

Substance misuse, encompassing opioid misuse, non-opioid illicit use, and alcohol misuse, constitutes a heterogeneous set of conditions associated with high mortality and is a primary driver of rehospitalizations. Despite efforts to reduce harm through tertiary prevention strategies centered on fatal events, the substance misuse epidemic has continued to escalate. Recently, we constructed the Substance Misuse Data Commons (SMDC), a first-of-its-kind data platform that captures early warning signs - emergency department visits and hospitalizations - that lie on the path to fatality and that the National Drug Control Policy recommends using to improve tertiary prevention efforts. The SMDC integrates structured and unstructured longitudinal data from electronic health records, social disadvantage, medical and pharmacy claims, criminal justice, and health surveillance data to provide a comprehensive picture of regional substance misuse. The goal of this proposal is to utilize the SMDC and stakeholder engagement to develop an equitable multimodal artificial intelligence (AI) model to predict outcomes for substance misuse patients. In Aim 1, we will derive a multimodal AI model for predicting all-cause death or rehospitalization post-discharge and externally validate it on urban and rural cohorts. In Aim 2, we will use statistical algorithms to ensure that the model usage will lead to equitable treatment for all patients. Both aims will be accomplished using iterative feedback from a stakeholder group comprised of former patients, emergency providers, and addiction specialists. Completion of this proposal will lead to an accurate, equitable, and feedback-driven AI prognostication model for integration into clinical decision support for substance misuse.