Differentiating adolescent suicidal and nonsuicidal self-harm with artificial intelligence: Beyond suicidal intent...

Area of Science:

• Psychiatry and Mental Health
• Adolescent Psychology
• Public Health

Background:

• Clinical differentiation between adolescent suicidal self-harm (SSH) and nonsuicidal self-harm (NSSH) presents a significant challenge.
• The feasibility of distinguishing these behaviors based solely on clinical presentation is debated.

Purpose of the Study:

• To determine if adolescent SSH can be differentiated from NSSH using a holistic multimodel domain analysis (HMDA).
• To assess HMDA's compatibility with major suicide theories without relying on intent or acquired capability.
• To explore risk and protective factors associated with SSH and NSSH.

Main Methods:

• A cross-sectional study utilizing data from a 2017-2019 Norwegian nationwide survey.
• Inclusion of 26,904 adolescents (mean age 15.01, 29.9% males) who self-reported self-harm.
• Application of a meta-learner classification model and exploratory factor analysis.

Main Results:

• The classification model achieved 77.0% AUC and 70.0% sensitivity/specificity in differentiating SSH from NSSH.
• Exploratory factor analysis identified eight key factors: disordered eating/future orientation, sleep, affect/self-orientation, risk-taking, peer relations, safety/security, family relations, and substance use.
• These factors highlight the importance of well-being domains in predicting self-harm behaviors.

Conclusions:

• Adolescent SSH and NSSH are distinct, multifaceted conditions that can be differentiated without relying on suicide intent or capability information.
• The findings support the theory of adolescent suicidality and self-destruction and partially support ideation-to-action theories.
• HMDA provides valuable insights into risk and protective indicators for adolescent self-harm.