JMIR Publications

ABSTRACT

Background:

Mood disorders (MD) are among the most common mental health conditions worldwide, significantly contributing to mortality, morbidity, and disability rates. In today's society, individuals generate digital traces through their interactions with wearable and consumer-grade personal digital devices. These traces can be collected, processed, and analyzed, offering a unique opportunity to quantify and monitor individuals with mental disorders in their natural living environments. Various forms of digital traces from the use of smartphones and other personal digital devices hold the potential to reveal new behavioral markers associated with depressive symptoms in patients with mood disorders.

Objective:

We conducted an observational longitudinal study, MoMo-Mood Pilot, involving a cohort of patients with major depressive disorder (MDD) and a healthy control group without any diagnosed mental disorder to confirm the feasibility of our digital phenotyping study design. Upon confirming feasibility, we conducted a year-long, more extensive study, MoMo-Mood, with a similar design. This study comprised (1) three subcohorts of patients with a major depressive episode (MDE), including those with either MDD, bipolar disorder (BD), or concurrent borderline personality disorder (BPD), as well as (2) a healthy control group without any diagnosed mental disorder. We investigated whether differences in behavioral patterns, as quantified by passively collected digital trace data, could be observed at the group level, i.e., patients vs. healthy controls. We studied the volume and temporal patterns of smartphone screen and application usage, communication, sleep, mobility, and physical activity. Additionally, we examined which of the passively quantified behavioral features and sociodemographic factors are associated with the presence of depression in study participants at different points in time.

Methods:

MoMo-Mood Pilot and MoMo-Mood recruited 201 participants. The pilot study enrolled 14 patients with MDD and 23 healthy controls. The participants completed a 2-phase study: the first two weeks (the active phase) involved collecting data from bed sensors, actigraphy, and smartphone data passively collected through the Niima platform. Additionally, participants had to actively engage with the study by answering five sets of questions daily, prompted on their smartphones throughout the day. In the second phase (the passive phase), which lasted up to 1 year, only passive smartphone data and bi-weekly Patient Health Questionnaire (PHQ-9) assessments were collected. The MoMo-Mood study, which had a similar setting to the pilot study, enrolled 164 participants: 133 were patients diagnosed with current MDE through structured interviews (85 with MDD, 27 with BPD, and 21 with BD), and 31 were healthy controls. Survival analysis was performed to compare the adherence of healthy controls and the patient subcohorts. Statistical tests were performed to compare behavioral patterns, and a generalized linear mixed model was used to assess the association between different factors and the presence of depression.

Results:

The study design was proven feasible upon completing the pilot study. Survival analysis using the log-rank test showed no statistically significant difference in participants' adherence between subcohorts. The overall communication volume was similar in both groups; patients exhibited later peaks in communication and sleep-wake rhythms, typically waking up later. In terms of location patterns, significant differences emerged: weekday location variance showed lower values for patients (Control = -10.04 ± 2.73, Patient = -11.91 ± 2.50, MWU test P-value = .004) and normalized entropy of location was also lower among patients (Control = 2.10 ± 1.38, Patient = 1.57 ± 1.10, MWU test P-value = .05). However, no differences were observed in weekend location patterns. The temporal communication patterns for controls were found to be more diverse than that of patients (MWU test, p<0.001). In contrast, patient smartphone usage temporal patterns were more varied than the control group's. Investigating mobile-derived behaviors and their relationship with the concurrent presence of depression, we observed that the duration of outgoing calls over the past two weeks was negatively correlated with the presence of depression (beta = -7.93, 95% CI: -13.17 to- -2.69, P-value = .003). Conversely, longer durations of incoming calls (beta = 5.33, 95% CI: -0.08 to 10.75, p = 0.05) and a larger proportion of time spent at home (beta = 5.16, 95% CI: 0.78 to 9.54, P-value = .002) were positively associated with the presence of depression.

Conclusions:

This work demonstrates the design of a longitudinal digital phenotyping study harnessing data from a cohort of patients with depression. It also shows the important features and data streams for future analyses of behavioral markers of mood disorders. However, among outpatients with mild to moderate depressive disorders, their group-level differences from healthy controls in any of the modalities alone remain overall modest. Therefore, future studies need to be able to combine data from multiple domains and modalities to detect more subtle differences, identify individualized signatures, and combine passive monitoring data with clinical ratings.