Modeling Short-Term Symptom Changes and Behavioral Subtypes of Depression and Anxiety in the General Population:...

Area of Science:

• Digital phenotyping
• Mental health monitoring
• Computational psychiatry

Background:

• Smartphone-based digital phenotyping offers a novel method for passive mental health monitoring using behavioral data.
• Previous research linked smartphone features to depression and anxiety, but short-term symptom changes in general populations remain understudied.
• Understanding demographic variability and behavioral patterns is crucial for interpreting passive smartphone data in mental health.

Purpose of the Study:

• To model short-term changes in depression and anxiety severity using passive smartphone data.
• To evaluate model performance across diverse demographic subgroups.
• To identify specific behavioral patterns associated with symptom fluctuations.

Main Methods:

• Collected 2 weeks of smartphone usage data from 95 general population adults.
• Assessed depression and anxiety using Hamilton Rating Scales.
• Extracted and compressed behavioral features (activity, app/screen use) via autoencoder and PCA.
• Trained random forest classifiers to predict symptom score changes, incorporating demographics and baseline scores.
• Used unsupervised clustering to identify behavioral subtypes linked to symptom changes.

Main Results:

• Models achieved moderate accuracy for predicting changes in depression (0.70) and anxiety (0.65) scores.
• Performance varied by demographics, with trends toward lower accuracy in younger adults and females.
• Excluding baseline scores significantly reduced predictive performance, highlighting its importance.
• Clustering identified 4 behavioral subtypes; structured, daytime-focused use correlated with depressive symptom improvement.

Conclusions:

• Passive smartphone data shows moderate potential for modeling short-term mental health changes in nonclinical samples.
• Baseline symptom severity significantly contributes to predictive performance, suggesting supplementary rather than stand-alone value.
• Behavioral subtypes identified through passive data can aid in understanding distinct symptom trajectories, contributing to digital phenotyping research.