Mentoring a Multi-Disciplinary Workforce to Leverage AI for HIV and Substance Use Research

About This Grant

Project Summary/Abstract Mortality among people who use drugs remains one of the least understood public health crises in the United States. Many deaths occur outside clinical settings, are never captured by surveillance systems, and reflect preventable causes such as overdose, infection, violence, and untreated chronic illness. This project evaluates whether verbal autopsy in the form of structured and narrative information collected from peers, family, and service providers, can be feasibly and ethically implemented in a critical area and whether artificial intelligence assisted verbal autopsy can improve classification of causes of death compared with established automated systems. As a K24 mechanism, the award will support new training for the principal investigator who will then use this training to execute the research aims and mentor early-stage investigators on artificial intelligence, substance use and HIV research. Aim 1 evaluates the feasibility and acceptability of verbal autopsy interviews among a network of people who use drugs. Aim 2 develops a reproducible workflow for linking verbal autopsy data with fragmented service records and generating deidentified factual statements suitable for artificial intelligence analysis. Aim 3 pilots multiple artificial intelligence models to classify causes of death from narratives and linked records, benchmarked against clinician adjudication and established verbal autopsy algorithms. This work will produce the first detailed mortality profiles for this population and establish an ethically governed framework for artificial intelligence that can be used in research and mentoring. Findings will inform future interventions aimed at preventing avoidable deaths related to substance use.