Sequence Modeling of Passive Sensing Data for Treatment Response Prediction in Major Depressive Disorder

Abstract

Major depressive disorder (MDD) is a prevalent mental health condition and has become a pressing societal challenge. Early prediction of treatment response may aid in the rehabilitation engineering of depression, which is of great practical significance for the relief of suffering and burden of MDD. In this paper, we present a sequence modeling approach that uses data collected by passive sensing techniques to predict patients with an outcome of treatment responded defined by the reduction in clinical administrated scales. Hundreds of patients with MDD have been recruited from outpatient clinics at 4 psychiatric sites. Each has been delivered with a self-developed app to passively record their daily phone usage and physical data with minimal human action. An unavoidable dilemma in passive sensing is missing values. To overcome that, the proposed approach combined feature extraction and sequence modeling methods to fully utilize the pattern of missing values from longitudinal data. With no treatment constraints, it enables us to predict the treatment response of MDD 8-10 weeks before the completion of the treatment course, leaving time for preventative measures. Our work explored the feasibility of treatment response prediction using longitudinal passive sensing data and sparse ground truth, and also has the potential for preventing depression by forecasting treatment outcomes weeks in advance.

From smartphone data to clinically relevant predictions: A systematic review of digital phenotyping methods in depression

“…Fujino, Tokuda, & Fujimoto (2023) investigated changes in smartphone step count surrounding MDD-related medical visits in a group-based analysis, finding that mean daily step count tended to decrease in the two weeks before the visits. Zou, et al, 2023 predicted treatment response in MDD patients 10 weeks in advance, achieving an AUROC (Area Under the Receiver Operating Characteristic Curve) of 0.65. Braund, et al (2022) investigated differences in circadian rhythm between participants with MDD and participants with bipolar disorder, not identifying a difference between groups.…”

“…Studies were grouped by analysis goal to allow for comparison of methods with similar objectives. To this end, to investigate our first research question we first compared studies that correlated individual passive smartphone features with depression symptom severity (Cao, et al, 2020; Sun, et al, 2023; Sverdlov, et al, 2021; Wasserzug, et al, 2023; Zhang, et al, 2022; Zou, et al, 2023). To address our second research question, we then shifted our focus to the different methods that have been used for various depression prediction tasks, investigating the methods used for predicting symptom severity (Braund, et al, 2022; Cao, et al, 2020; Faurholt-Jepsen, et al, 2022; Kathan, et al, 2022; Pedrelli, et al, 2020; Pellegrini, et al, 2022; Sverdlov, et al, 2021; Zhang, et al, 2021; Zhang, et al, 2022).…”

“…In terms of general phone usage, Sun, et al (2023) found that unlock duration was positively correlated with depression score, and Zou, et al (2023) found a positive correlation between latest phone usage time and depression score, and a negative correlation with night time phone usage and with afternoon entertainment app usage. However, no significant correlation between symptom scores and smartphone screen usage was found by Cao, et al, 2020, and Zou, et al, 2023 noted that only a small number of features were found to have a significant correlation.…”

“…predicted treatment response in MDD patients 10 weeks in advance, achieving an AUROC (Area Under the Receiver Operating Characteristic Curve) of 0.65.symptoms. In a later study from this consortium,Zhang, et al, 2023 also did not find a significant impact of baseline depression score on participant retention for passive data,…”

“…Fujino, Tokuda, & Fujimoto (2023) investigated changes in smartphone step count surrounding MDD-related medical visits in a group-based analysis, finding that mean daily step count tended to decrease in the two weeks before the visits. Zou, et al, 2023 however did now find differences in data availability.…”