A single day of step counts from your fitness tracker tells you far less than researchers once assumed. That’s the central finding of a preprint posted to medRxiv on January 30, 2026, which analyzed 11 million days of Fitbit data collected from 29,351 participants enrolled in the National Institutes of Health’s All of Us Research Program — one of the largest wearable-data analyses of its kind, and one with implications for how doctors, insurers, and consumers alike should interpret the numbers on their wrist.
What the Study Actually Measured
The research team derived four daily activity metrics from participants’ Fitbit data — step count, peak one-minute cadence, peak 30-minute cadence, and heart rate per step — and analyzed them across five different time windows ranging from a single day up to a full year. They then ran phenome-wide association analyses against more than 700 incident and over 1,300 prevalent disease outcomes drawn from participants’ linked electronic health records, among a study population with a mean age of 57.3 years and 69% women, nearly half of whom had contributed more than a year of Fitbit data.
Longer Windows, Stronger Signals
The headline result is straightforward but consequential: the longer the observation window, the stronger and more stable the associations with real health outcomes became. A full year of step-count data was associated with 373 prevalent and 37 incident health outcomes after statistical correction for multiple comparisons — compared to just 231 prevalent and 17 incident outcomes when researchers looked at single-day step counts alone. In other words, a snapshot from any given Tuesday is a poor substitute for a year’s worth of trend data when it comes to predicting actual disease risk.
Surprising New Associations
Among the associations that emerged from the longer time windows were links researchers hadn’t necessarily expected going in, including connections between wearable-derived activity metrics and chronic pain syndrome, SARS-CoV-2 infection history, and autoimmune disease. These aren’t associations a clinician would typically think to investigate using activity data, and the scale of the All of Us dataset — spanning years of continuous, real-world tracking rather than a short controlled study — is what allowed the pattern to surface at all.
Why This Matters Beyond Academic Interest
The finding lands at a moment when both consumer wearable companies and clinicians are increasingly leaning on AI systems to interpret exactly this kind of data in real time. Oura and Whoop have both rolled out AI coaching assistants this year designed to translate biometric trends into daily recommendations, and the NIH results are a useful check on how much weight any single day’s readout should really carry. A single bad night’s sleep or one low-step day flagged by an app’s AI assistant may say far less about a person’s underlying health trajectory than a year-long pattern would — a distinction that’s easy to lose when an app delivers same-day feedback with confident, specific language.
The Limits Researchers Are Careful to Note
The study’s authors and independent voices in the field caution against over-reading the associations. Phenome-wide association studies like this one are built to generate hypotheses at scale, not to prove that a given activity metric causes or prevents a given disease — some of the strongest correlations could reflect reverse causation, where an undiagnosed illness reduces a person’s activity levels rather than low activity causing the illness. The All of Us cohort, while large, also skews toward people willing to enroll in a long-term biomedical research program and wear a Fitbit consistently for months or years, a group that may not represent the broader population evenly.
A Companion Study on Data Quality
A related paper examining Fitbit physical activity and sleep data within the All of Us program has separately flagged data exploration and processing considerations that researchers need to account for before drawing conclusions — a reminder that wearable datasets, however large, carry their own quirks around missing data, device-switching, and inconsistent wear time that can distort findings if not carefully handled.
What’s Next
The All of Us program continues to grow its wearable dataset, and the research team’s approach — prioritizing long observation windows over single-day snapshots — is likely to influence how future studies using Fitbit, Oura, Whoop, and other consumer wearable data are designed. For everyday users, the takeaway is a useful corrective to the instant-feedback culture wearable companies have built their AI coaching products around: a single day’s numbers, however confidently an app interprets them, are a far weaker predictor of your actual health than the pattern you’ve built up over months.
