Predicting incident type 2 diabetes using wearable activity and polygenic risk: A survival-modeling study in All of Us
Authors' conclusion
Step-count thresholds for reducing T2D risk varied across genetic risk groups, indicating that step counts do not confer uniform protection. Step counts and PRS provided independent, complementary predictive information, and their combination improved prediction of incident T2D.
Study context
- Population
- Adults in the NIH All of Us Research Program with valid Fitbit step data and whole-genome-derived Polygenic Risk Score (PRS), excluding T2D before or within a 180-day lead-in
- Participants
- 4,589 adults
- Exposure
- Daily step count and PRS as predictors of incident T2D
- Outcome
- Incident T2D (HbA1c ≥ 6.5%, plasma glucose ≥126 mg/dL, or All of Us T2D condition record)
- Follow-up
- Median 2.92 years
Methodological detailssurvival modelling, Cox and machine-learning survival models
- Survival modelling
- Cox and machine-learning survival models
Main finding
As reported in the abstract
Incident T2D · median 2.92 years (15 340 person-years)
- Participants developing T2D
- 265
- Cumulative incidence
- 5.77%
- Incidence rate
- 17.27 per 1000 person-years
- Per additional 1000 steps/day, aHR
- 0.83 (95% CI 0.79-0.88; P < .005)
- Per 1-SD higher PRS, aHR (2.32-2.96; P < .005)
- 2.62
Observational data: associations, not effects.
Further findings
Risk-reducing step thresholds
- Overall threshold, steps/day (P < .001)
- ∼7000
- High PRS threshold, steps/day
- ∼7800
- Low PRS threshold, steps/day (difference by PRS group P < .001)
- ∼5800
Discrimination (C-index)
- Clinical model
- 0.748
- Clinical model + steps
- 0.774
- Clinical model + steps + PRS
- 0.867
- Penalized Cox (highest)
- 0.859
- Survival support vector machine
- 0.85
- Classical Cox
- 0.846
Calibration
- Calibration was best with Random Survival Forests
Original publication
Predicting incident type 2 diabetes using wearable activity and polygenic risk: A survival-modeling study in All of Us. J Clin Endocrinol Metab. Published 15 December 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →