Machine Learning Prediction of Self-Injurious Outcomes in Adolescents by Sexual and Gender Identity

Area of Science:

• Adolescent Mental Health
• Machine Learning in Psychiatry
• Sexual and Gender Minority Health

Background:

• Sexual and gender minority (SGM) adolescents experience higher rates of self-injurious thoughts and behaviors (SITBs).
• Understanding the unique risk factors for SITBs in SGM youth is crucial but challenging.
• Traditional statistical models have limitations in capturing the complexity of SITB prediction.

Purpose of the Study:

• To explore and compare machine learning models for predicting SITBs in diverse adolescents.
• To identify key predictors of suicide ideation, attempts, and non-suicidal self-injury in SGM youth.
• To assess the performance of different machine learning approaches in this population.

Main Methods:

• Utilized data from a large adolescent survey (N=2,452) with psychological and demographic features.
• Compared generalized linear models, random forest, and gradient boosting decision tree models.
• Conducted follow-up comparisons between cisgender/gender minority and heterosexual/sexual minority groups.

Main Results:

• Machine learning models demonstrated comparable prediction performance.
• Depression was a consistent top predictor across most models.
• Discrimination and loneliness emerged as significant predictors, particularly for gender minority adolescents.

Conclusions:

• Machine learning offers valuable insights into SITB prediction in SGM youth.
• Discrimination and loneliness are critical factors for SITB risk in gender minorities.
• These factors warrant further investigation as potential treatment targets for SGM adolescents.