Can a Fitness Tracker Detect Depression?

This piece covers what peer-reviewed research says about using consumer wearable data (steps, heart rate, sleep, energy expenditure) to detect or screen for depression symptoms, including postpartum depression. It does not cover diagnosis, treatment, or any use of a tracker as a substitute for clinical evaluation.

Research has established that data from consumer wearables, fed into machine learning models, can distinguish people who screen positive for depression from those who don't with meaningfully better than chance accuracy. This has been shown separately in general working adults, student and outpatient samples, and postpartum women. It's screening-level pattern detection built on sleep, activity, and heart rate signals, not a diagnosis, and it hasn't been tested as something a tracker does on its own in the background of daily life.

The question behind the question

There's a specific moment a lot of tracker owners have: scrolling back through weeks of flat step counts, disrupted sleep, and a resting heart rate that's crept up, and wondering if the device already knows something is off before they've named it themselves. It's not a paranoid thought, it's a reasonable one, given how much of that data the app is already collecting.

The tension is this: trackers are marketed as fitness tools, not mental health tools, so it's easy to assume the two are unrelated categories entirely, one measuring your body and the other measuring your mind. What the research actually asks is narrower and more interesting than that split suggests. Whether the same behavioral and physiological signals a tracker already gathers for fitness purposes also carry a statistical signature of depressive symptoms.

What's actually being detected

None of this research claims a tracker is reading emotion directly. What gets fed into these models is the same raw material trackers were built to measure in the first place: step counts, sleep patterns, heart rate, energy expenditure, sometimes circadian rhythm markers. A scoping review of passive sensing research summarized association patterns between behavioral features gathered by wearables and smartphones and clinically diagnosed mental health conditions across 42 included studies, framing this as a growing but still-consolidating field rather than a settled clinical tool. A separate systematic review focused specifically on physical activity data and mental health outcomes processed through AI methods, and it noted that machine learning approaches here are still relatively new compared to more traditional statistical approaches to the same question.

It's worth being precise about direction here too. A systematic review of mobile sensing data for tracking depression severity over time found 9 studies, out of nearly 13,000 screened, that met criteria involving continuous data collection of 12 weeks or longer in people with a diagnosis, with sample sizes ranging from 45 to 2,200 participants. That's a much narrower evidence base than the cross-sectional screening studies above, aimed at monitoring symptom change over time in people already diagnosed. A different task than screening an undiagnosed person.

Where the accuracy claims run into limits

A model's discrimination score only means as much as the sensor data underneath it. Reviews of consumer wearable accuracy have found step counting performs reasonably well in controlled lab settings for devices like Fitbit and Apple Watch, but heart rate measurement and other outcomes showed more variable performance, and a large systematic review covering 65 studies found enough clinical heterogeneity across devices and settings that the authors couldn't pool results into a single accuracy figure. I think that variability sits underneath every depression-detection model built on the same raw signals, and questions like how a watch actually counts steps or whether tracker accuracy holds up for people with movement disorders matter more here than they might first seem.

It's also worth separating detection research from the more familiar claim that trackers get people moving more. A meta-analysis of tracker-based interventions found consistent increases in physical activity participation when people used consumer wearables with feedback, a real and separately documented effect, but it's not the same finding as a tracker detecting depression. One is about behavior change. The other is about pattern recognition in existing data.

The postpartum case, specifically

Postpartum depression gets its own mention because the underdiagnosis problem is well documented and the wearable-based approach was tested directly against it. Using a large public research dataset, researchers built individualized models per participant, comparing prepregnancy, pregnancy, postpartum without depression, and postpartum with depression periods using the same person's own Fitbit-derived heart rate, activity, and energy expenditure data as its own baseline. That within-person design is different from the cross-sectional screening studies elsewhere in this piece, and it points at something the broader field is still working out. Whether comparing someone against their own baseline data performs differently than comparing them against a population norm.

What this doesn't settle

None of these studies claim the tracker itself, as sold, tells anyone they're depressed. The models are built after data collection, in research settings, using PHQ-9 scores or clinical diagnosis as the thing being predicted. A device screen showing a stress score or readiness number is a different, less validated claim than what these studies tested. Related open questions, like how sensor calorie estimates hold up or whether wearing a tracker changes behavior on its own, sit alongside this one, reminders that a number on a screen and a validated research finding aren't automatically the same thing.

Common questions