Estimating longitudinal depressive symptoms from smartphone data in a transdiagnostic cohort

Area of Science:

• Digital phenotyping in mental health research.
• Cross-diagnostic assessment of psychiatric disorders.

Background:

• Smartphone-based passive sensing shows promise for assessing depression severity.
• Previous studies have not evaluated passive measures across diverse psychiatric diagnoses.

Purpose of the Study:

• To investigate the utility of smartphone passive sensing and self-report measures in predicting depression severity across major depressive disorder, bipolar disorder, and schizophrenia spectrum disorders.
• To compare the predictive performance of passive data, self-report data, and their combination.

Main Methods:

• A cohort of 45 individuals (diagnosed with MDD, BD, SZ/SZA, or no psychiatric disorder) participated in an 8-week study.
• Participants completed weekly PHQ-8 self-reports and biweekly MADRS clinician ratings; smartphone GPS and accelerometer data were collected passively.
• Linear mixed models were used to predict MADRS scores using self-report and passive data.

Main Results:

• 84% of participants completed the study.
• The root-mean-squared error (RMSE) for predicting depression severity (MADRS score) was 4.72 for passive data alone, 4.27 for self-report alone, and 4.30 for combined data.
• Passive measures did not significantly improve prediction accuracy compared to self-report measures alone.

Conclusions:

• Passive smartphone data did not enhance depression severity prediction in this cross-disorder sample.
• Passive measures may capture unique behavioral phenotypes not amenable to objective, granular, or long-term self-report.
• Further research is needed to explore the specific value of passive sensing in understanding behavioral patterns in psychiatric populations.