Digital phenotyping correlations in larger mental health samples: analysis and replication

Area of Science:

• Digital mental health
• Mobile health (mHealth) applications
• Wearable technology and sensors

Background:

• Smartphones offer potential for patient-reported outcomes and passive data collection in mental health.
• The precise utility and correlational strength of smartphone-derived sensor data for mental health assessment remain under investigation.
• Previous research indicates variable correlations between passive sensing data and self-reported mental health scores.

Purpose of the Study:

• To investigate correlations between passive sensor data and mental health survey scores using a large dataset from the mindLAMP app.
• To evaluate the predictive capability of passive data features for mental health survey outcomes.
• To assess the impact of combining passive data with active survey responses on predictive model performance.

Main Methods:

• 147 college students provided 270 weeks of data, completing daily and weekly mental health surveys.
• Correlations were analyzed between six weekly surveys and 13 passive data metrics (GPS, call logs, sleep duration).
• Logistic regression models were trained to predict survey scores using passive data, with and without daily survey inputs.

Main Results:

• Correlations between passive data and survey scores were lower than reported in smaller prior studies.
• GPS, call frequency, and sleep duration emerged as the most informative passive data features.
• Predictive models using only passive data performed poorly; performance significantly improved when daily survey scores were incorporated.

Conclusions:

• Passive smartphone data alone has limited predictive power for mental health survey scores.
• Integrating passive data with brief, daily surveys enhances the accuracy of mental health predictions.
• The combination of active (surveys) and passive (sensor) data holds potential for refining clinical utility and improving mental health monitoring.