Psychosocial challenges and substance use among suicidal autistic individuals on social media: LLM-assisted keyword generation with human-in-the-loop refinement.

Abstract

Autism Spectrum Disorder (ASD) is associated with an elevated risk of suicidal ideation due to various co-occurring psychosocial challenges. Although these challenges are recognized, their extent and prevalence-particularly autism-specific difficulties and substance use patterns among individuals with ASD-remain largely unexplored, especially within social media health discussions, where individuals frequently express their thoughts and emotions. A comprehensive understanding of these aspects is essential to develop targeted prevention strategies that mitigate suicide risk among individuals with autism. This study addresses this gap by examining the complex challenges contributing to suicidal ideation among individuals with ASD-including autism-specific factors and substance use patterns-through the lens of social media discourse. We conducted a keyword-based analysis integrating large language models (LLMs) with human-in-the-loop refinement, focusing on three prominent categories of challenges: (i) interpersonal and social struggles, (ii) sensory and emotional regulation difficulties, and (iii) co-occurring mental health conditions. Keywords were generated by an LLM, validated by humans, and applied to analyze online health discourse to identify patterns and prevalence of these key challenges. A similar LLM-based approach was employed to assess substance use prevalence. The most prevalent challenges reported were mental health conditions (52.4 %; 95 % CI: [0.487, 0.561]) and social isolation (48.9 %; 95 % CI: [0.452, 0.526]). Less common but significant challenges included bullying (23.2 %; 95 % CI: [0.201, 0.263]) and issues with identity and self-acceptance (20.3 %; 95 % CI: [0.173, 0.233]). The strongest-but still weak-correlation was observed between social isolation and bullying. Additionally, substance use-including alcohol, cannabis, and prescription drug misuse-was also prevalent in this population. This study highlights the diverse characteristics and underlying factors contributing to suicidal ideation among individuals with ASD, including social isolation, bullying, autism-specific challenges, co-occurring mental health conditions, and substance use patterns. These findings emphasize the need for tailored mental health interventions, inform preventive strategies, and demonstrate the value of social media health data in advancing research in this area.