SCIENCE / SUMMARY / WATCHES AND WEARABLES

Is sleep recorded by a wristband over years linked to later diagnoses?

SINGLE STUDY · COHORT ■■□□ low evidence: large but observational; selected participants, sleep data from a closed device algorithm

Who

6,785 adults who met the criteria, out of 14,892 people sharing Fitbit data in the All of Us research programme in the United States. 71% were women, 84% white and 71% college graduates; median age 50.2. In total there are about 6.5 million nights of data. People with less than six months of data, or with under 4 hours of recorded sleep on more than 30% of days, were excluded.

What they did

They used the daily summaries Fitbit shows its users: sleep length, restless sleep, light, deep and REM percentages, irregularity (the standard deviation of daily sleep length) and the share of weekdays with sleep onset between 8 p.m. and 2 a.m. These were matched with new diagnoses in electronic health records. To reduce reverse effects, diagnoses made in the first 180 days of tracking were not counted. A scan across 1,636 diagnosis groups came first (with a Bonferroni threshold for multiple comparisons), followed by time varying Cox models for chronic diseases chosen in advance.

What they found

Median sleep length was 6.7 hours and median sleep onset 11:10 p.m. The scan found 48 significant associations, 24 of them involving irregularity. Each 1 hour increase in irregularity went together with higher odds of high blood pressure (odds ratio 1.56), obesity (1.49), major depression (1.75) and high blood lipids (1.39). Compared with the median of 6.8 hours, people averaging 5 hours had 1.29 times the risk of high blood pressure, 1.64 times for major depression and 1.46 times for generalised anxiety disorder; these risks were also higher at 10 hours (a J shaped pattern). A higher deep sleep percentage went with less atrial fibrillation (hazard ratio 0.59 in the time varying model, confidence interval 0.35 to 0.99); after accounting for sleep apnea diagnoses, the same link in the scan (odds ratio 0.86) did not reach the Bonferroni threshold.

Limits

As the authors note: participants were relatively young and mostly women, white and college educated; device owners are healthier than the general population. Sleep stages are estimated by Fitbit's closed algorithm from heart rate and movement; atrial fibrillation can affect heart rate and distort these estimates. Reverse effects cannot be fully excluded; diagnosis codes may be misclassified; unmeasured confounders such as occupation remain. Funding: the US National Institutes of Health and, in part, an unrestricted gift from Google. Four authors are Google employees holding Alphabet stock and one sits on the ResMed board; according to the authors, Fitbit and Google staff took part in design, interpretation and writing, but not in data collection or analysis.

Link to Nefesly

One of the reasons behind Nefesly showing sleep as a trend across several nights rather than a single night. Nefesly draws no disease result from watch data; this is not a health claim. How we built it (Turkish)

Reference

Zheng NS, Annis J, Master H, Han L, Gleichauf K, Ching JH, et al. Sleep patterns and risk of chronic disease as measured by long-term monitoring with commercial wearable devices in the All of Us Research Program. Nature Medicine. 2024;30(9):2648-2656. doi:10.1038/s41591-024-03155-8 · Open access (CC BY 4.0).

Why chosen

One of 15 sources chosen from 1,118 candidates on the watch and wearables axis; included as the strongest study showing that long term consumer watch sleep data can be linked to health outcomes.

Summarised by

Nefesly, from the full text, 11 October 2026. Awaiting clinician check.

Copyright

Summarised from the full text, not translated from the abstract.