An artificial intelligence approach to support adolescent suicide prevention initiatives in the United States.

Abstract

Adolescent suicide remains a critical public health issue in the United States, with complex, interrelated risk and protective factors operating across multiple levels of the social ecology. This study presents the design of a novel agent-based model (ABM) created to simulate the progression of suicidal behaviors among U.S. adolescents ages 12-19 years and evaluate the effectiveness of suicide prevention strategies. The model integrates data from eight nationally representative datasets and is grounded in a comprehensive conceptual framework developed by CDC's National Center for Injury Prevention and Control. Our ABM represents a first-of-its-kind effort in the U.S. to model the adolescent population with this level of granularity, supporting scenario analyses across multiple domains of suicide prevention, including facilitating access to care, teaching coping skills, and creating protective environments. Calibration against national suicide mortality data demonstrates strong alignment for major demographic groups, while extreme condition testing confirms model validity across policy-relevant scenarios. Findings indicate that interventions promoting protective environments that minimize bullying or teaching coping skills, such as social-emotional learning programs, yielded some of the largest reductions in suicidal ideation and attempts. This simulation model provides decision makers with a robust tool to inform public health strategies, with the potential to extend to other injury and violence prevention efforts. The transparent, data-driven design of the model supports future adaptation for cost-effectiveness and evaluations of prevention strategies.