Multiple wearable digital biomarkers for AI-based diagnosis of major depressive disorder defined by self-reported and clinician-rated scales

Major depressive disorder (MDD) is traditionally assessed using clinician-rated and self-reported scales. However, repeated assessments are challenging, and discrepancies between these scales may limit their suitability as ground truth for wearable-based diagnosis. This study developed machine learning models using wearable-derived digital biomarkers to diagnose the presence and severity of MDD, comparing the Beck Depression Inventory (BDI) and Hamilton Depression Rating Scale (HAMD) as ground-truth labels. A total of 111 patients with depression and 50 healthy controls were recruited. Sleep, heart rate variability (HRV), and physical activity data were collected over 6 weeks using a wrist-worn device. The BDI was assessed at weeks 0, 2, 4, and 6, while HAMD was assessed at weeks 0 and 6. Logistic regression, support vector machine, random forest, and extreme gradient boosting models were developed for binary and multiclass classification and evaluated using participant-level nested cross-validation. The HAMD-based model achieved peak area under the curve (AUC) values of 0.84 for diagnosis and 0.72 (macro-averaged) for severity classification, outperforming the BDI-based model (AUCs: 0.71 and 0.64); the advantage of HAMD persisted when both scales were compared on identical participants and time points (weeks 0 and 6). Key features in the HAMD-based models included low-frequency power, entropy, root mean square of successive differences (RMSSD), Rapid Eye Movement (REM) sleep latency, deep sleep duration, total active minutes, and weekday distance. While RMSSD, entropy, deep sleep duration, and activity were negatively associated with the predicted probability of depression, REM sleep latency showed a positive association. Wearable-derived multimodal biomarkers are effective for MDD diagnosis, and the clinician-rated HAMD may provide a more suitable ground truth than the BDI. This approach supports continuous monitoring and early detection in digital mental health management.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-19
DOI
https://doi.org/10.1038/s41598-026-80489-y
Primary Topic
Emotion and Mood Recognition
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Multiple wearable digital biomarkers for AI-based diagnosis of major depressive disorder defined by self-reported and clinician-rated scales

Seon Min Lee, Kwang Gi Kim, Seung-Gul Kang, Seon Heui Lee
Scientific Reports
Emotion and Mood Recognition
article

Multiple wearable digital biomarkers for AI-based diagnosis of major depressive disorder defined by self-reported and clinician-rated scales

Seon Min Lee, Kwang Gi Kim, Seung-Gul Kang, Seon Heui Lee
article en

Abstract

Major depressive disorder (MDD) is traditionally assessed using clinician-rated and self-reported scales. However, repeated assessments are challenging, and discrepancies between these scales may limit their suitability as ground truth for wearable-based diagnosis. This study developed machine learning models using wearable-derived digital biomarkers to diagnose the presence and severity of MDD, comparing the Beck Depression Inventory (BDI) and Hamilton Depression Rating Scale (HAMD) as ground-truth labels. A total of 111 patients with depression and 50 healthy controls were recruited. Sleep, heart rate variability (HRV), and physical activity data were collected over 6 weeks using a wrist-worn device. The BDI was assessed at weeks 0, 2, 4, and 6, while HAMD was assessed at weeks 0 and 6. Logistic regression, support vector machine, random forest, and extreme gradient boosting models were developed for binary and multiclass classification and evaluated using participant-level nested cross-validation. The HAMD-based model achieved peak area under the curve (AUC) values of 0.84 for diagnosis and 0.72 (macro-averaged) for severity classification, outperforming the BDI-based model (AUCs: 0.71 and 0.64); the advantage of HAMD persisted when both scales were compared on identical participants and time points (weeks 0 and 6). Key features in the HAMD-based models included low-frequency power, entropy, root mean square of successive differences (RMSSD), Rapid Eye Movement (REM) sleep latency, deep sleep duration, total active minutes, and weekday distance. While RMSSD, entropy, deep sleep duration, and activity were negatively associated with the predicted probability of depression, REM sleep latency showed a positive association. Wearable-derived multimodal biomarkers are effective for MDD diagnosis, and the clinician-rated HAMD may provide a more suitable ground truth than the BDI. This approach supports continuous monitoring and early detection in digital mental health management.

Scientific Reports
Gachon University (KR), Gachon University Gil Medical Center (KR)
Emotion and Mood Recognition
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon