Predicting self-harm within six months after initial presentation to youth mental health services: A machine learning...

Area of Science:

• Youth Mental Health
• Machine Learning in Healthcare
• Suicidology

Background:

• Reducing self-harm in young people is a key health service priority.
• Predicting self-harm is complex due to its rarity, but prediction models can aid decision-making for interventions.
• This study aimed to predict self-harm in young people within six months of their initial presentation.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting self-harm in young people.
• To assess the utility of these models in identifying individuals who may benefit from further assessment or intervention.
• To improve health service strategies for early identification and reduction of youth self-harm.

Main Methods:

• Included 1962 young people (12-30 years) from Australian youth mental health services.
• Trained and tested six machine learning algorithms using repeated ten-fold cross-validation.
• Evaluated model performance and net benefit using decision curve analysis.

Main Results:

• 16% of participants engaged in self-harm within six months.
• The top 50% of individuals ranked by predicted probability accounted for over 82% of self-harm incidents.
• Models showed fair prediction (AUROCs 0.744-0.755) and calibration, with positive net benefit.
• Key predictors included prior self-harm, age, functioning, sex, bipolar disorder, psychosis-like experiences, antipsychotic treatment, and suicide ideation history.

Conclusions:

• Self-harm prediction models can identify a significant sub-population for enhanced assessment and targeted interventions.
• These models offer a valuable tool for health services to better identify and reduce youth self-harm.
• Implementing such models can potentially decrease distress, morbidity, healthcare utilization, and mortality associated with self-harm.