Journaling Data for Daily PHQ-2 Depression Prediction and Forecasting.

Area of Science:

• Digital Health
• Mental Health Technology
• Computational Psychiatry

Background:

• Digital health applications are crucial for monitoring mental health conditions like depression.
• Current applications often rely on passive data for predicting Patient-Health-Questionnaire (PHQ) scores.
• Limited exploration of actively collected data, such as the Behavioral Activation for Depression Scale-Short Form (BADSSF), for depression monitoring.

Purpose of the Study:

• To investigate the utility of actively collected data for predicting and forecasting daily PHQ-2 scores.
• To assess the additive value of incorporating diverse, daily collected data beyond passive smartphone data.
• To develop improved predictive models for depression symptom severity using a novel longitudinal dataset.

Main Methods:

• Utilized a newly collected longitudinal dataset containing actively collected daily scores (e.g., BADSSF, CESD, PDD).
• Employed leave-one-subject-out cross-validation for robust model evaluation.
• Developed models for both daily prediction and short-term forecasting (up to 7 days) of PHQ-2 scores.

Main Results:

• Achieved a best Mean Absolute Error (MAE) of 1.417 for daily PHQ-2 score prediction.
• Obtained a best MAE of 1.914 for forecasting PHQ-2 scores using data from the preceding 7 days.
• Demonstrated significant improvement in prediction accuracy compared to baseline methods not using active data.

Conclusions:

• Actively collected data significantly enhances the accuracy of predicting and forecasting daily depression symptom severity (PHQ-2 scores).
• Integrating diverse, actively collected data streams offers additive value to digital mental health monitoring applications.
• Future digital health tools should leverage actively collected data for more precise and personalized depression management.