Tracker Data Sorted Depression Scores About 80% Right
Fitbit data sorted 290 workers at about 80% accuracy, but only in the clearest cases. Across 42 reviewed studies, one was checked on new people.
Evidence rating for the proposition “Wearable-derived data detects depressive symptoms”: Uncertain
Partly. In research, wearable data can partly sort people who score as depressed on a screening questionnaire from those who do not. That is not a diagnosis. In 290 working adults wearing Fitbit Charge 2 devices, the model had limited ability across the whole sample. Among those 290 adults, it reached 80% accuracy only in subsamples of clearly depressed people against clearly healthy ones. A review of 42 studies found just one had external validation.
Can tracker data tell who has depressive symptoms?
In studies, roughly four times in five. A 2026 study used wearable data from 282 people, 187 students and 95 outpatients. Depression was defined as a score of 10 or more on the PHQ-9 questionnaire. A regularized regression model reached an area under the curve of 0.82 and accuracy of 79% on held-out data.
Three features predicted best: minimum awake heart rate, variability in sleep duration and maximum step count. Sample type was itself a strong predictor. Wearable features added value beyond it.
The model sorts questionnaire scores. It does not diagnose depression.
1 study
- In 282 participants (187 students, 95 outpatients), an elastic net model using wearable features classified PHQ-9 screen-positive status (score of 10 or more) with an area under the curve of 0.82 and accuracy of 79%. Minimum awake heart rate, variability in sleep duration and maximum step count were the strongest predictors.
How well does it work in ordinary working adults?
Less well. A study of 290 healthy working adults had participants wear a Fitbit Charge 2 for 14 days. The mean age was 33. Data included steps, heart rate, energy expenditure and sleep.
Greater symptom severity went with more variation in nighttime heart rate, lower regularity of weekday step rhythms and fewer step-based daily peaks. The authors nevertheless report limited ability to detect depression across the whole sample.
The 80% figure came from balanced subsamples. Depressed participants were compared with healthy participants who had no or minimal symptoms. There the model reached 80% accuracy, 82% sensitivity and 78% specificity. The authors conclude that current evidence shows limited predictive ability.
1 study
- In 267 analyzed working adults wearing a Fitbit Charge 2 for 14 days, digital biomarkers had limited ability to detect depression in the whole sample. In balanced subsamples of depressed and healthy participants, the model reached 80% accuracy, 82% sensitivity and 78% specificity.
Does it work for postpartum depression?
Possibly, in a very small sample. A study used All of Us data with Fitbit records. It built models for each woman, comparing her own prepregnancy, pregnancy, postpartum and postpartum-with-depression periods.
The best random forest model reached a multiclass area under the curve of 0.85 and a kappa of 0.80. Performance dropped in women without postpartum depression. Calories burned at basal metabolic rate was the most predictive marker. Prior depression history did not change performance.
The groups were small. Fewer than 20 women had postpartum depression and 39 did not. The authors call wearables a promising tool.
How strong is the whole field?
Thin. A 2025 scoping review of 42 studies of passive sensing found a median sample of 60.5 people. Seventy-six percent had fewer than 100 participants, and 45% monitored for under 7 days. Only 1 of 42 had external validation.
A second review of 49 studies of wearable activity data and machine learning found 71% had fewer than 100 participants and 47% had under 14 days of data. Studies scored an average of 6.47 out of 18 on quality.
A third review looked only at diagnosed people tracked for 12 weeks or longer. Of 12,997 articles, 9 studies qualified. Six showed mobile sensing could predict depressive episodes or severity. Four reported accuracy of 81% to 91%. Only one differentiated depressive states.
3 studies
- In 42 studies of passive sensing with machine learning, median sample size was 60.5, 76% had fewer than 100 participants, 45% monitored for under 7 days and 1 (2%) had external validation.
- Of 12,997 articles on mobile sensing in diagnosed depression tracked for 12 weeks or more, 9 qualified. Six showed prediction of episodes or severity, four reported accuracy of 81% to 91%, and one differentiated depressive states.
- In 49 studies using wearable physical activity data and machine learning for mental health, 71% had fewer than 100 participants and 47% had under 14 days of data. Mean quality score was 6.47 of 18.
Are the underlying tracker measurements accurate?
Unevenly. A review of 65 studies of wrist activity trackers could not pool results because of clinical heterogeneity. Fitbit Charge models had a mean absolute percentage error under 25% for steps across 20 studies. The Apple Watch was under 10% for heart rate in 2 studies. Energy expenditure error exceeded 30% for all brands.
Any depression model built on these signals inherits those errors. See how a watch counts steps, whether the calorie number means anything and accuracy in movement disorders. Whether wearing a tracker changes activity is in whether a fitness tracker makes people more active.
What does this not settle?
No study here shows a tracker detecting depression in the background of daily life for the general public. The models were built after data collection in research settings. Their targets were questionnaire scores or clinical diagnoses.
A stress or readiness number on a device is a different and less validated claim than these research models.
The accuracy figures come from research models built after data collection, in student, outpatient and working-adult samples of 267 to 290 people, and a postpartum sample of fewer than 60 women. None tested a tracker screening for depression on its own in everyday life, and only 1 of 42 reviewed studies had external validation.
Common questions
Can a fitness tracker detect depression?
Not as a diagnosis. In research, wearable data sorted depression screening results with about 79% to 80% accuracy. A 290-person Fitbit study reached 80% only in clearly depressed against clearly healthy subsamples, with limited ability across the whole sample.
Which tracker measurements predicted depressive symptoms?
In a 282-person study, minimum awake heart rate, variability in sleep duration and maximum step count were the strongest predictors of a PHQ-9 score of 10 or more.
Can wearables detect postpartum depression?
One study of Fitbit data from women who gave birth reached a multiclass area under the curve of 0.85 for individual models. Fewer than 20 women had postpartum depression, so it is a small preliminary result.
How strong is the evidence overall?
Weak. In a review of 42 studies, the median sample was 60.5 people, 76% had fewer than 100 participants and only one study had external validation.
Does my tracker's mood feature use these models?
Not established. The studies here built research models after data collection. None tested a consumer mood or stress score against a clinical measure.
Sources
- Using wearable data to detect depression severity across clinical and non-clinical samples.
- Digital Biomarkers for Depression Screening With Wearable Devices: Cross-sectional Study With Machine Learning Modeling.
- Harnessing Consumer Wearable Digital Biomarkers for Individualized Recognition of Postpartum Depression Using the All of Us Research Program Data Set: Cross-Sectional Study.
- Passive Sensing for Mental Health Monitoring Using Machine Learning With Wearables and Smartphones: Scoping Review.
- Use of Mobile Sensing Data for Longitudinal Monitoring and Prediction of Depression Severity: Systematic Review.
- Applying AI in the Context of the Association Between Device-Based Assessment of Physical Activity and Mental Health: Systematic Review.
- Accuracy and Acceptability of Wrist-Wearable Activity-Tracking Devices: Systematic Review of the Literature.
KnowYourPrime Evidence Registry: Finding KYP-ER-0009: Wearable Heart Rate Accuracy · Finding KYP-ER-0010: Wearable Calorie and Energy Expenditure Accuracy