A machine learning analysis of suicidal ideation and suicide attempt among U.S. youth and young adults from multilevel, longitudinal survey data

Front Psychiatry. 2025 Feb 24;16:1511966. doi: 10.3389/fpsyt.2025.1511966. eCollection 2025.

ABSTRACT

OBJECTIVES: To investigate individual, interpersonal, health system, and community factors associated with suicidal ideation (SI) and attempts (SA).

METHODS: Utilizing nationally representative data from the National Longitudinal Study of Adolescent to Adult Health (7^th-12^th graders in 1994-95 followed >20 years until 2016-18, N=18,375), least absolute shrinkage selector operator (LASSO) regression determined multilevel predictors of SA and SI. Models comprised full and diagnosis subgroups (ADD/ADHD, depression, PTSD, anxiety, learning disabilities [LD]).

RESULTS: Approximately 2.48% and 8.97% reported SA and SI, respectively. Over 25% had depression, and 20.98% anxiety, 6.42% PTSD, 4.55% ADD/ADHD, and 2.50% LD. LASSO regression identified 20 and 21 factors associated with SA and SI. Individual-level factors associated with SI and SA included educational attainment, substance use, ADD/ADHD, depression, anxiety, and PTSD. Interpersonal-level factors included social support, household size, and parental education, while health system-level factors comprised health care receipt, health insurance, and counseling. The strongest associations were among individual-level factors followed by interpersonal and health system factors.

CONCLUSIONS: The distinct factors associated with SI and SA across diagnostic subgroups highlight the importance of targeted, subgroup-specific suicide prevention interventions. These findings emphasize the value of precise, data-driven approaches for suicide prevention among diverse populations and individuals with disabilities across the life-course.

PMID:40066138 | PMC:PMC11891230 | DOI:10.3389/fpsyt.2025.1511966