Testing Suicide Risk Prediction Algorithms Using Phone Measurements With Patients in Acute Mental Health Settings:...

Area of Science:

• Digital health
• Machine learning in healthcare
• Mental health technology

Background:

• Digital phenotyping and machine learning are increasingly used in healthcare.
• Smartphones offer an underutilized data source for mental health risk prediction and prevention.
• Advancing global mental health through accessible technology is a key objective.

Purpose of the Study:

• To apply machine learning for suicide risk prediction in an acute mental health setting.
• To utilize smartphone-collected data as an alternative to traditional clinical data.
• To explore novel approaches in mental health risk assessment.

Main Methods:

• Developed the 'Strength Within Me' smartphone app linked to wearables and social media.
• Collected data on sleep, mood, physical activity, and phone engagement from 66 inpatients.
• Assessed mood, sleep, and suicide risk via clinical interviews and machine learning algorithm testing.

Main Results:

• K-nearest neighbors (KNN) algorithm showed 68% mean accuracy and an AUC of 0.65.
• KNN significantly outperformed baseline classifiers, including random forest and logistic regression.
• Demonstrated the first steps in prototyping a continuous suicide risk assessment system using mobile devices.

Conclusions:

• This study contributes to the under-addressed area of suicidality prediction.
• Utilizing smartphone data for suicide risk algorithms in inpatients is feasible.
• Iterative development holds potential for accurate risk prediction, but ethical and legal implications require consideration.