A Co-Segmentation Algorithm to Predict Emotional Stress From Passively Sensed mHealth Data

Area of Science:

• Computational psychiatry
• Digital health
• Machine learning for healthcare

Background:

• Mood disorders and chronic pain in middle-aged and older adults are complex and often influenced by emotional stress.
• Traditional machine learning (ML) methods struggle to capture time-varying patterns in nonstationary time-series data from passive sensing and self-reported variables.
• Identifying short-term stress fluctuations is crucial for effective therapeutic interventions.

Purpose of the Study:

• To develop a data-driven cosegmentation algorithm for identifying emotionally stressful states using passively sensed and self-reported smartphone data.
• To improve the accuracy of detecting stress periods by analyzing time-varying local patterns.
• To leverage short-time window associations between different data types for predictive modeling.

Main Methods:

• Developed a novel cosegmentation algorithm that segments passive sensing variables by detecting change points.
• Examined segment-specific associations between passive sensing and active (self-reported) variables.
• Utilized identified cosegmented periods to predict future emotional stress states using standard ML methods.
• Applied the algorithm to patient data from the ALACRITY Phase I study.

Main Results:

• The data-driven segmentation algorithm accurately identifies periods of emotional stress.
• The proposed method demonstrates superior accuracy in detecting stress periods compared to traditional ML methods that do not incorporate segmentation.
• The algorithm effectively captures time-varying local patterns and short-time window associations crucial for stress detection.

Conclusions:

• Data-driven segmentation offers a more accurate approach to identifying emotional stress periods in patients with mood disorders.
• The developed algorithm enhances the analysis of complex, nonstationary time-series data from digital health tools.
• This method holds promise for improving monitoring and intervention strategies in digital mental healthcare.