Comprehensive Symptom Prediction in Inpatients With Acute Psychiatric Disorders Using Wearable-Based Deep Learning...

Area of Science:

• Digital psychiatry
• Machine learning in healthcare
• Wearable sensor technology

Background:

• Assessing acute psychiatric disorders is challenging, with limited research on digital tools.
• High staff workload and burnout risk in acute psychiatric wards necessitate innovative solutions.
• Wearable sensors and deep learning offer objective data to aid clinical decision-making.

Purpose of the Study:

• Develop and validate wearable-based deep learning models.
• Predict patient symptoms comprehensively across acute psychiatric wards.
• Enhance clinical decision support for psychiatric care.

Main Methods:

• Recruited patients with schizophrenia and mood disorders from 4 wards.
• Collected heart rate, accelerometer, and location data via wrist-worn wearables.
• Developed deep learning models (Single vs. Multitask learning) to predict symptom deterioration and severity.

Main Results:

• 191 participants included in the final analysis.
• Models accurately classified symptom deterioration (0.73-0.75 accuracy).
• Multitask learning models outperformed single-task models in predicting symptom severity (R² up to 0.74).

Conclusions:

• Wearable sensor data and deep learning effectively predict symptom changes and severity.
• Multitask learning is a promising, computationally efficient approach for symptom prediction.
• Variations across wards highlight the need for local validation or federated learning for generalizability.