Using Machine Learning to Predict Suicide Attempts in Military Personnel

Area of Science:

• Psychiatry
• Psychology
• Military Health

Background:

• Identifying suicide attempt predictors is crucial for effective intervention and prevention.
• Low base rates and statistical limitations complicate predictor identification.
• High-risk suicidal soldiers in outpatient mental health services were studied.

Purpose of the Study:

• To utilize machine learning to identify predictors of suicidal behaviors.
• To examine predictors among soldiers receiving Brief Cognitive Behavioral Therapy for Suicide Prevention (BCBT) versus treatment as usual (TAU).

Main Methods:

• Machine learning analysis of self-report clinical and demographic variables.
• Data collected pre-treatment from 152 participants with recent suicidal ideation/behaviors.
• Prediction of suicide attempts during or after a two-year follow-up period.

Main Results:

• A combination of variables correctly classified 30.8% of patients who attempted suicide.
• Key predictors identified include worst-point suicidal ideation, history of multiple suicide attempts, treatment group (BCBT/TAU), suicidogenic cognitions, and male sex.
• The identified combination demonstrated higher sensitivity than many previous models.

Conclusions:

• The study provides a combination of assessable clinical variables for identifying high-risk suicidal individuals.
• Machine learning offers a powerful approach to uncovering complex patterns in suicidal behavior prediction.
• Findings can inform targeted suicide prevention strategies in military populations.