Researchers combined something inherited and something behavioral in the same statistical model, then asked which adults later received a depression diagnosis in their medical records.
The behavioral half came from Fitbits worn for months on end. The inherited half came from genome data. Several ordinary daily patterns, including the number of steps people took and the consistency of their sleep duration, were associated with who was later diagnosed.
The analysis, led by Yuezhou Zhang and colleagues at King's College London, integrated genomic data, electronic health records, and longitudinal wearable data from 3,030 adults in the National Institutes of Health's All of Us Research Program. It is currently posted as a preprint and has not undergone peer review, which affects how much weight it carries.
What the Wearable Data Was Associated With
The researchers followed participants after a 180-day baseline period, during which 284 of them went on to develop major depressive disorder, as recorded in their health records.
Using time-varying Cox models, which allow a person's measured behavior to change over the follow-up period rather than fixing it at a single baseline value, the team examined monthly wearable-derived activity and sleep features alongside a polygenic risk score for major depression.
Several factors were associated with a higher risk of a subsequent recorded diagnosis: a higher polygenic risk score, lower daily step counts, lower light physical activity, lower vigorous physical activity, lower sleep efficiency, and greater sleep duration variability.
The variability finding is the one that stands somewhat apart. Most of these measures are about doing less. Sleep duration variability is about inconsistency rather than amount, meaning a person who sleeps a steady seven hours and a person who swings between five and ten look very different on this measure, even if their weekly totals match.
Two Kinds of Risk in One Model
Putting inherited liability and current behavior into the same analysis is the methodological point of the study.
Genetic risk for depression is stable across a lifetime and can be measured before anything happens. It is also, on its own, a fairly blunt instrument. Polygenic scores for depression explain only a small share of variation in who develops the condition, which is why they have not entered routine clinical use despite years of work.
Behavior is the opposite. It changes week to week and can potentially be acted on, but it is difficult to measure objectively at scale. Self-reported activity and sleep are unreliable, and asking someone with depressive symptoms to recall their sleep introduces the exact bias a researcher is trying to avoid.
Wearables measure passively and continuously, which is what makes this pairing possible in the first place. The same group has published related work on physical activity trajectories preceding diagnosis in the same cohort and on large-scale digital phenotyping in a general population sample. The interest is not in either measure alone but in whether behavioral drift adds information on top of a fixed genetic baseline.
Why This Is Not a Smartwatch That Detects Depression
Several limitations sit between this result and anything a consumer could use, and they are worth stating clearly.
The outcome was an electronic health record diagnosis of major depressive disorder, not a structured clinical assessment. That captures people who sought care, received a diagnosis and had it coded, and misses people who were depressed and never diagnosed. Access to care is unevenly distributed, so the outcome variable partly measures who reaches a clinician.
The sample was restricted to 3,030 adults of genetically inferred European ancestry, with 284 events. Polygenic scores developed largely in European-ancestry populations lose predictive accuracy when applied to other groups, which is why the restriction exists, and it also means the results should not be assumed to hold for the broader US population.
Reverse causation is the most serious interpretive problem. Reduced activity, fragmented sleep and irregular schedules are symptoms of depression, not merely precursors. Depressive episodes have a prodromal phase, and a decline in step count during the months before a recorded diagnosis may indicate that the illness is already underway rather than a signal preceding it. The 180-day baseline helps but does not resolve this.
Measurement accuracy is also a live question. Consumer wearable sleep metrics, including sleep efficiency, agree less closely with laboratory polysomnography than clinical devices do, so a measure labeled sleep efficiency in this dataset is an algorithmic estimate rather than a laboratory finding.
Fitbit users within All of Us are also self-selected. People who buy a fitness tracker, wear it consistently for years, and consent to share their data differ from the general population in terms of health behavior and socioeconomic status. Similar constraints apply across other wearable research in the same program.
Where This Fits in the Broader Evidence
The direction of these findings is consistent with earlier work in the same cohort. A study published in Nature Medicine used Fitbit step monitoring across All of Us participants over a median of about four years and found an inverse relationship between daily steps and incident major depressive disorder, alongside similar patterns for obesity, sleep apnea, and reflux disease.
Consistency across analyses strengthens the case that these associations are real. It does not answer whether they are causal, and observational data cannot settle whether adding steps to a sedentary week reduces depression risk or whether people already heading toward an episode simply move less.
The realistic application of this line of research is clinical rather than consumer. If passive behavioral data reliably flags elevated risk, that information belongs with a clinician who is already following a patient, as one input among many. It is not a feature that should generate an alert on a wrist.
Nobody should read a step count or a sleep score as a verdict on their mental health, and a bad month of sleep data is not a diagnosis. Anyone noticing persistent changes in mood, sleep, energy or interest in things they normally enjoy should raise it with a doctor or mental health professional, who can assess what is actually happening.
Key Questions Answered
What did the analysis find?
Higher polygenic risk, lower daily step counts, lower light and vigorous physical activity, lower sleep efficiency, and greater sleep duration variability were each associated with a higher risk of a later recorded diagnosis of major depressive disorder.
Has this been peer-reviewed?
No. It is currently posted as a preprint, which means the methods and conclusions have not been vetted by independent reviewers.
How many people were studied?
3,030 adults of genetically inferred European ancestry in the All of Us Research Program, of whom 284 developed a recorded diagnosis after a 180-day baseline period.
Does this mean a wearable can predict depression?
No. The associations are group-level, and the design cannot separate early symptoms from precursors. Reduced activity and disrupted sleep are themselves features of depression, so a decline before diagnosis may be the illness already underway.
Why was the sample limited to one ancestry group?
Polygenic scores were developed largely in European-ancestry populations and lose accuracy when applied to other populations. That restriction also means the results should not be assumed to generalize to the wider US population.
How reliable are consumer sleep metrics?
Less reliable than clinical sleep studies. Sleep efficiency from a consumer wearable is an algorithmic estimate that agrees imperfectly with laboratory polysomnography.
What should someone do if their step count or sleep score drops?
Not treat it as a diagnosis. Persistent changes in mood, sleep, energy or interest are worth raising with a doctor or mental health professional who can actually assess what is happening.
