A much longer view of daily activity
A new analysis of nearly 11 million days of Fitbit recordings suggests that months of activity data reveal more consistent relationships with health than measurements collected over just one day.
Published in Nature Communications on 14 October 2026, the study linked wearable activity measurements from 29,303 participants in the United States’ All of Us Research Program with electronic health records.
The researchers examined hundreds of medical outcomes and compared how well different activity summaries captured associations with disease. Their findings point to the importance of measurement duration rather than relying only on a single daily step total.
Crucially, the work does not show that a fitness tracker diagnoses disease. It demonstrates how long-term activity patterns can be studied alongside medical histories to identify associations worth investigating.
Why one day may not tell the whole story
Daily activity fluctuates for ordinary reasons. Weather, work schedules, illness, holidays and family responsibilities can all change the number of steps someone takes.
A single day may therefore be an unreliable description of a person’s typical activity. Even a week can capture an unusual period rather than a stable routine.
The researchers asked whether using longer windows would change the statistical relationships between wearable-derived activity measures and medical outcomes.
They compared summaries based on one day, one week, one month, six months and one year of measurements.
That comparison allowed them to test whether more sustained observations offered additional information or simply repeated what shorter samples already showed.
The distinction is important because many research studies use a short monitoring period, even when the health conditions of interest develop over years.
How the study was conducted
Evelynne S. Fulda and colleagues analysed data contributed through the All of Us Research Program, a large US initiative that combines participant information with health records and other research data.
The study included 29,303 participants with Fitbit information and linked clinical records, representing approximately 11 million recorded person-days.
Researchers calculated four daily activity measures: step count, peak one-minute cadence, peak 30-minute cadence and heart rate per step.
Cadence describes walking or movement pace in steps per minute. Peak cadence over different intervals provides a way to capture short bursts and more sustained periods of faster movement.
Heart rate per step was included as another way to characterise the physiological response associated with movement.
Each of the four measures was summarised over five time windows, creating 20 combinations for analysis.
The team then used a phenome-wide association approach. Instead of testing a single disease, this method screens relationships across many outcomes recorded in electronic health data.
The published analysis considered more than 700 incident outcomes and more than 1,400 prevalent outcomes. Incident outcomes concern new diagnoses occurring during the observation framework, whereas prevalent outcomes concern conditions already recorded.
The researchers applied statistical correction for the large number of comparisons, reducing the likelihood of treating chance findings as meaningful.
The headline numbers
One-year step counts were associated with 389 prevalent medical outcomes after Bonferroni correction, compared with 271 outcomes for a one-day step measure.
For incident outcomes, the corresponding figures were 61 for one-year step counts and 33 for one-day step counts.
That is an increase of 118 statistically associated prevalent outcomes and 28 incident outcomes when the longer measurement window was used.
These are counts of statistical associations, not counts of illnesses prevented or cases correctly diagnosed.
An association can arise because activity influences health, because illness changes activity, or because another factor affects both.
The central result is that longer measurement windows captured more stable relationships across a wider range of outcomes.
The findings also indicate that activity measurements can contain useful information beyond the familiar question of how many steps someone took yesterday.
Different measurements capture different aspects of movement
Step count is easy to understand and widely available, but it does not describe every aspect of physical activity.
Two people may take the same number of steps while moving at very different speeds or with different levels of exertion.
Peak one-minute cadence captures brief periods of relatively rapid stepping. Peak 30-minute cadence reflects activity sustained across a longer interval.
Heart rate per step offers a different perspective, although it can be influenced by fitness, medication, illness and device measurement quality.
The researchers found that the 20 derived activity summaries were highly correlated overall.
That similarity means the metrics often reflect related aspects of a person’s movement patterns, but it does not make them interchangeable for every health outcome.
Comparing metrics helps identify which features may be most informative for particular research questions.
It also reduces the temptation to treat a single wearable number as a complete measure of physical health.
Links extending beyond familiar conditions
Previous research has repeatedly linked physical activity with cardiometabolic health. The new study also found associations involving conditions not always central to wearable research.
The published report highlighted relationships with chronic pain syndrome and SARS-CoV-2-related outcomes among its findings.
These associations should not be interpreted as evidence that walking prevents every condition appearing in the analysis.
For example, a person experiencing chronic pain may move less because of symptoms. Lower activity could therefore be a consequence of the condition rather than its cause.
Similarly, a history of infection or recovery from illness can affect movement patterns, making the timing of measurement important.
The distinction between prevalent and incident outcomes helps researchers explore these possibilities, but it cannot resolve every causal question.
Follow-up studies would need stronger designs to determine which associations are useful for prediction and which reflect underlying illness or other circumstances.
What longer monitoring can add
Longer monitoring can smooth out unusually active and unusually inactive days, producing a better estimate of someone’s habitual behaviour.
It can also reveal whether patterns remain consistent over time, rather than reflecting a short-lived change.
In this analysis, the researchers found that longer windows generally produced stronger and more stable disease associations.
However, more data do not automatically guarantee better clinical decisions. The benefit depends on the question being asked, the quality of the measurements and how the resulting information is interpreted.
Devices may be removed, run out of battery or record activity differently across users and situations.
People who regularly wear a tracker for a full year may also differ systematically from those who contribute only a few days.
These practical considerations matter when deciding whether a monitoring approach could work outside a research programme.
A healthcare system would also need to establish whether the additional information improves decisions enough to justify the burden and cost of collecting it.
Why medical records were important
Wearable data can describe movement, but they do not independently establish whether someone has a diagnosed disease.
Linking activity information to electronic health records allowed researchers to investigate associations with conditions documented in clinical care.
That linkage greatly expands the number of questions that can be asked of a large activity dataset.
It also introduces limitations. Medical records reflect contact with healthcare services, coding practices and the completeness of available information.
An unrecorded diagnosis does not necessarily mean a participant has never experienced the condition.
Researchers must therefore distinguish between what is documented in a health record and the true underlying burden of disease.
The study’s findings are best understood as relationships between measured activity and recorded health outcomes, rather than a comprehensive census of all illnesses among participants.
What the findings could mean for future care
If replicated and validated, longer-term wearable measures might contribute to tools that help researchers identify changes in health status or estimate risk.
For example, an unusual sustained decline in activity could prompt investigation, although many harmless explanations would remain possible.
The current study did not establish a diagnostic threshold or show that clinicians should use a particular step count to screen for specific diseases.
Nor did it test whether giving people fitness trackers reduces hospital admissions or improves survival.
Any clinical application would require prospective testing to assess accuracy, fairness, patient benefit and the consequences of false alarms.
Privacy is another consideration. Combining continuous behaviour data with medical histories creates sensitive information that requires appropriate safeguards.
The researchers’ work therefore offers a foundation for future investigation, not a ready-made replacement for clinical assessment.
Limitations that matter
The study was observational, so it cannot demonstrate that the recorded activity patterns caused the associated medical outcomes.
Participants came from a US research programme and had linked Fitbit and health-record information. They may not represent the general population or communities with less access to wearable technology.
Longer observation windows may preferentially include people who consistently use devices, introducing selection effects.
Step counts and cadence estimates depend on how the wearable detects movement, and heart-rate measurements can vary with device placement and other conditions.
Electronic health records are also imperfect proxies for underlying disease status.
Even with multiple-testing correction, a large association screen can identify relationships influenced by confounding or reverse causation.
Finally, the results do not establish an ideal daily step target or prove that one year of monitoring is necessary for every person.
Independent validation across other populations, devices and health systems will be important before the findings are translated into routine practice.
The bigger lesson
The analysis demonstrates that the duration of activity measurement can substantially change what researchers learn from wearable data.
Across 29,303 participants and roughly 11 million days of Fitbit recordings, one-year step summaries were associated with more recorded health outcomes than one-day summaries.
That result strengthens the argument for studying sustained behaviour rather than isolated daily readings.
At the same time, it reinforces the need to separate statistical association from medical diagnosis and causation.
For consumers, a smartwatch can help reveal personal activity trends. For scientists, large and carefully analysed collections of those trends may offer a deeper view of population health.
The challenge now is to determine which insights are reliable, clinically useful and equitable across different groups of people.
Source Information
Study: 11 million days of longitudinal Fitbit data reveal novel future health insights
Authors: Evelynne S. Fulda, Bennett J. Waxse, Slavina B. Goleva and colleagues.
Journal: Nature Communications
Published: 14 October 2026
Methods: Observational phenome-wide analysis of approximately 11 million Fitbit person-days from 29,303 All of Us participants, comparing four activity metrics across five measurement windows with electronic health records.
