Machine Learning Models to Identify Individuals With Imminent Suicide Risk Using a Wearable Device: A Pilot Study.

Area of Science:

• Digital health
• Psychiatry
• Machine learning in healthcare

Background:

• Identifying individuals at immediate risk of suicide is a critical challenge in mental healthcare.
• Existing methods may not capture real-time physiological and behavioral changes associated with acute suicidal ideation.

Purpose of the Study:

• To evaluate the efficacy of commercially available wearable devices in identifying individuals at immediate risk of suicide.
• To compare the performance of single-level and multilevel machine learning models in predicting suicide risk using wearable device data.

Main Methods:

• Thirty-nine participants with acute depressive episodes and 20 healthy controls wore a wearable device for two months.
• Data collected included activities, sleep, heart rate, and heart rate variability; participants self-reported mood daily.
• Machine learning models were developed to predict suicide risk based on Hamilton Depression Rating Scale (HAMD-3) scores.

Main Results:

• Both single-level and multilevel models accurately predicted imminent suicide risk.
• The multilevel model achieved a higher area under the curve (0.89) compared to the single-level model (0.88).
• Key predictors in the multilevel model included HAMD total score and heart rate variability; in the single-level model, HAMD total score and diagnosis were significant.

Conclusions:

• Commercially available wearable devices represent a promising tool for real-time identification of suicide risk.
• Further research with enhanced temporal resolution is recommended to refine predictive capabilities.