Wearable-derived features can identify reproducible and clinically relevant behavioral subtypes of sleep and activity in individuals with major depressive disorder, reflecting known behavioral correlates of depression and offering a data-driven framework for reducing phenotypic heterogeneity.
Key Findings
Results
Three distinct activity subtypes were identified from Fitbit data in participants with recurrent depression.
The three subtypes were labeled high activity, light activity, and low activity.
Subtypes were identified using Gaussian mixture models (GMMs) applied to longitudinal Fitbit data.
The sample comprised 623 participants with recurrent depression enrolled in the Remote Assessment of Disease and Relapse in Major Depressive Disorder (RADAR-MDD) study.
Model selection incorporated grouped cross-validation and seed selection to ensure robustness.
These subtypes align with known associations between depression and behavioral patterns.
Results
Four distinct sleep subtypes were identified from Fitbit data in participants with recurrent depression.
The four subtypes were labeled efficient early sleepers, efficient late sleepers, disrupted sleepers, and variable late sleepers.
Subtypes were identified using Gaussian mixture models applied to objective wearable sleep data.
The disrupted sleepers subtype reflects the well-known association between depression and disturbed sleep.
Variable late sleepers represent a pattern of irregular and delayed sleep timing.
These sleep subtypes were described as 'consistently identified' across the modeling approach.
Results
Transition modeling using hidden Markov models revealed stability of subtypes within individuals over follow-up.
Hidden Markov models (HMMs) were applied to explore how participants transition between subtypes over time.
The analysis revealed 'stability within individuals over follow-up,' suggesting persistent behavioral phenotypes rather than momentary fluctuations.
This stability further supports the presence of behavioral phenotypes as opposed to transient states.
The longitudinal design allowed examination of subtype transitions across multiple time points in the RADAR-MDD study.
Methods
An unsupervised learning approach was used to identify depression subtypes without relying on predefined diagnostic labels or supervised models.
Prior efforts to identify objectively derived subtypes have relied on predefined diagnostic labels or supervised models, 'limiting discovery to existing clinical categories.'
The study applied Gaussian mixture models and hidden Markov models as unsupervised methods.
Model selection combined grouped cross-validation and seed selection to ensure robustness.
623 participants with recurrent depression contributed longitudinal Fitbit data for analysis.
The approach aimed to generate 'real-time, objective data on behavior and physiology, offering new perspectives on understanding depression phenotypes.'
Background
Digital phenotyping via wearables was proposed as a means to address symptom heterogeneity and limitations of self-report in depression research.
The study states that depression research has 'long been constrained by the disorder's vast symptom heterogeneity and by the reliance on self-report, which offers only a partial view of phenotypic expression.'
Fitbit devices were used to collect objective sleep and activity data from participants.
The RADAR-MDD study enrolled participants with recurrent major depressive disorder specifically to enable remote, longitudinal digital assessment.
The authors frame wearable-derived subtypes as potentially 'improving research stratification, and supporting personalized patient monitoring.'
Conclusions
The identified subtypes were described as clinically relevant and reflecting known behavioral correlates of depression.
Both activity and sleep subtypes were said to 'align with known associations between depression and behavioral patterns.'
The authors concluded that these subtypes 'reflect known behavioral correlates of depression.'
The subtypes were proposed to offer 'a data-driven framework for reducing phenotypic heterogeneity.'
Further work was noted as needed 'to validate these findings in independent cohorts and evaluate their potential use in reducing noise when using sleep or activity data to predict depression outcomes.'
What This Means
This research suggests that data collected from Fitbit wearable devices can reliably identify distinct behavioral subtypes among people with recurrent depression, without relying on clinician-assigned categories. By applying machine learning methods to objective sleep and physical activity data from 623 participants over time, the researchers found three types of activity patterns (high, light, and low activity) and four types of sleep patterns (efficient early sleepers, efficient late sleepers, disrupted sleepers, and variable late sleepers). Importantly, individuals tended to remain in the same subtype over time, suggesting these reflect stable behavioral traits rather than random day-to-day variation.
This matters because depression is notoriously difficult to study and treat partly due to how differently it presents across individuals — some people sleep too much, others too little; some become inactive, others remain active. Traditional research has often lumped all people with a depression diagnosis together, which can obscure these meaningful differences. This study demonstrates that wearable devices can capture these differences objectively and automatically, potentially offering a way to group patients more precisely for research or clinical purposes.
This research suggests that wearable-derived behavioral subtypes could eventually help researchers design better studies by grouping participants more meaningfully, and could support more personalized monitoring of patients with depression. However, the authors note that the findings need to be validated in other groups of people, and more work is needed to understand whether these subtypes can improve predictions about depression outcomes such as relapse or treatment response.
Check Your Own Numbers
Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.
Oetzmann C, Zhang Y, Cummins N, Carr E, Matcham F, Siddi S, et al.. (2026). Sleep and Activity Patterns in Depression From Wearable Data: Unsupervised Clustering Study.. Journal of medical Internet research. https://doi.org/10.2196/86900