Skip to main content
ABCT 2024 58th Annual Convention Main banner
Psychology Tools Ad
Final Program


Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Presentation Icons
Recorded Session
Recorded Session
Ticketed Session
Ticketed Session
Back
ViewAttendees 4

Symposia

Assessment

New Research Findings in Digital Phenotyping of Psychological Phenomena

3 - (SYM 111) Integrating Digital Phenotyping and Deep Learning to Understand Depression Symptom Variability: A Longitudinal Approach Using Passively-collected Movement and Sleep Data

Sunday, November 17, 2024
8:30 AM – 10:00 AM EST
Location: Grand Ballroom A, Level 5, Downtown Marriott
Abstract Body

Background: Major Depressive Disorder (MDD) significantly impacts global functioning and quality of life, presenting a challenge for effective management due to its symptom variability. Traditional assessments of depression severity have been complemented by recent efforts focusing on symptom variability, which is crucial for understanding clinical outcomes, including the risk of suicide attempts. This integrated study explores the predictive validity of passively-collected wearable movement and sleep data, alongside biodemographic variables, in capturing both long-term and acute fluctuations of depressive symptoms among individuals diagnosed with MDD.

Methods: In Study 1, we analyzed longitudinal data from 939 individuals over 12 months, employing an ensemble machine learning approach to detect long-term variability in depression symptoms using biodemographic, wearable movement, and sleep data. Variable importance was assessed using SHapley Additive exPlanations. Study 2 focused on the acute variability of depressive symptoms, utilizing deep learning models to analyze passively-collected accelerometry data from 179 clinically-diagnosed individuals over 90 days, resulting in 37,989 ecological momentary assessments (EMAs).

Results: Study 1 demonstrated moderate success in detecting long-term depression variability (r = 0.33 and r = 0.39 for domain-driven and exhaustive models, respectively), with significant features including both biodemographic and movement/sleep behaviors. Study 2 found a small-moderate correspondence between movement data predictions and observed depressive symptom variability (AUCTest = 0.59 ± 0.06, SensitivityTest = 0.30 ± 0.13, SpecificityTest = 0.81 ± 0.10), indicating the potential of passively-collected data to predict short-term symptom fluctuations.

Conclusions: The combined findings highlight the incremental predictive validity and clinical utility of using passively-collected wearable movement and sleep data, along with biodemographic variables, to detect both long-term and acute fluctuations in depressive symptoms among individuals with MDD. These insights underscore the potential of integrating multiple passive sensing modalities to enhance precision in monitoring and managing depression, paving the way for personalized treatment strategies aimed at stabilizing depressive symptoms over different time scales.