A Feasibility Study Using a Machine Learning Suicide Risk Prediction Model Based on Open-Ended Interview Language in...

Area of Science:

• Psychiatry
• Computer Science
• Machine Learning

Background:

• Adolescent suicide rates are increasing, necessitating novel risk identification methods.
• Machine learning (ML) models can analyze language samples to detect individuals at risk of suicide.
• This study explored the feasibility of using ML in adolescent therapy sessions.

Purpose of the Study:

• To assess the performance of ML models in identifying suicide risk in adolescents.
• To evaluate the integration of voice collection technology into therapy workflows.
• To explore the potential of natural language processing (NLP) for suicide risk assessment.

Main Methods:

• Natural language processing (NLP) ML models were tested using language samples from adolescent therapy sessions.
• Data included speech samples, standardized depression/suicidality scale scores, and therapist impressions.
• Previously developed models were used for external validation and risk prediction.

Main Results:

• 267 interviews from 60 students were analyzed, with 29 identified as high-risk.
• Support vector machines (AUC: 0.75) and logistic regression (AUC: 0.76) showed good discriminative ability.
• An extreme gradient boosting model achieved the best performance (AUC: 0.78).

Conclusions:

• Voice collection technology and ML can be integrated into mental health therapists' workflows.
• Language samples analyzed by ML models demonstrated good discrimination for suicide risk.
• This approach offers a potential tool for early identification and intervention in adolescent mental health care.