Mobile Communication Log Time Series to Detect Depressive Symptoms.

Area of Science:

• Digital mental health
• Computational psychiatry
• Machine learning in healthcare

Background:

• Major Depressive Disorder (MDD) is a prevalent mental health condition with complex, heterogeneous symptoms.
• Current MDD assessment relies heavily on self-report, limiting objective measurement.
• Understanding individual symptom variations is crucial for personalized MDD treatment.

Purpose of the Study:

• To investigate the predictive power of machine learning models for individual depressive symptoms using mobile phone communication data.
• To explore the relationship between objective behavioral data from call and text logs and subjective self-reported symptoms of MDD.
• To identify which communication attributes are most informative for predicting specific depressive symptoms.

Main Methods:

• Collected call and text logs from over 300 participants diagnosed with MDD.
• Created time-series data by aggregating communication attributes (e.g., length, count, contacts) at 4, 6, 12, and 24-hour intervals.
• Applied machine learning models to predict nine distinct self-reported depressive symptoms.

Main Results:

• Machine learning models achieved the highest accuracy in predicting movement irregularities (balanced accuracy 0.70).
• Suicidal ideation was predicted with a balanced accuracy of 0.67.
• Outgoing text message features were identified as the most valuable data type for symptom prediction.

Conclusions:

• Objective data from mobile phone logs can predict individual depressive symptoms with notable accuracy.
• This research supports the potential of mobile health technologies for personalized MDD assessment and intervention.
• Future studies can leverage these findings to develop more objective and tailored approaches to managing MDD.