Machine Learning Prediction of Persistent Adverse Mental Health Outcomes for Autistic Children: Leveraging Social Determinants of Health from Clinical Data

/ABSTRACT Autistic children and youth have high utilization of emergency department visits and inpatient stays for psychiatric indication. There is a need to understand who is at the greatest risk for adverse mental health outcomes in order to tailor prevention efforts. However, autism research to date has rarely incorporated social determinants of health (SDoH), which are social factors that significantly impact physical and mental health. Although SDoH account for up to 50% of health outcomes, they remain largely unstudied in autism research, and uncaptured in electronic health record (EHRs). The proposed research addresses these gaps by leveraging SDoH to predict the risk of persistent adverse mental health outcomes for autistic children and youth, ultimately improving clinician and health system responses to patients' SDoH-related needs. In this multimethod study, we will use EHR data from Children's Hospital Los Angeles (CHLA) and the University of Florida Health System (UF Health) to (1) identify ecological- and individual-level SDoH factors within EHRs, using natural language processing (NLP) to extract individual-level factors from clinical notes; and (2) build machine learning risk prediction models for persistent adverse mental health outcomes, investigating the additive effect of SDoH (compared to demographic/clinical characteristics). We will use a qualitative design to (3) explore clinician perspectives on utilizing SDoH within EHRs in clinical care, and gather clinician response to our prediction model, presented at Grand Rounds, to create action steps for both sites. We will elicit stakeholder input throughout the study from a Community Advisory Board made up of autistic adults and caregivers of autistic children and youth. By incorporating SDoH, we will be able to predict which autistic children and youth are at highest risk for persistent adverse mental health outcomes, and identify how clinicians can practically use SDoH to improve care. This will facilitate the `meaningful use' of SDoH in EHRs and ultimately improve equity in mental health service access and outcomes for this vulnerable population. Project Number: 1R01MH135867-01A1 | Fiscal Year: 2025 | NIH Institute/Center: National Institute of Mental Health (NIMH) | Principal Investigator: Amber Angell | Institution: UNIVERSITY OF SOUTHERN CALIFORNIA, Los Angeles, CA | Award Amount: $817,597 | Activity Code: R01 | Study Section: Mental Health Services Study Section[SERV] View on NIH RePORTER: https://reporter.nih.gov/project-details/11227302