Wearable-derived daily steps and first recorded cardiovascular diagnoses: outpatient and acute care in All of Us

Wearable-derived daily step counts are associated with cardiovascular health, but associations with the setting of cardiovascular disease (CVD) recording are less clear. We conducted a retrospective landmark cohort study using linked Fitbit, survey, and electronic health record data from the All of Us Research Program. Steps were averaged over the year before baseline. Eligibility required post-baseline sleep coverage and survey completeness, and classifiable pre-baseline smoking and drinking information; former smokers were excluded. After excluding participants with primary or secondary CVD recorded before baseline and 1,470 with either status during the first post-baseline year, outcomes were classified by the first qualifying primary CVD record during years two and three; a single qualifying outpatient record was sufficient. Among 8,292 participants, 5,780 had no recorded CVD, 2,127 had outpatient recording, and 385 had acute-care recording. Multinomial models adjusted for age, sex at birth, smoking, and drinking categories. Each one-standard-deviation increase in step count (3,299 steps/day) was associated with lower odds of outpatient recording versus no recorded CVD (odds ratio 0.786, 95% confidence interval 0.744–0.830), acute-care recording versus no recorded CVD (0.659, 0.586–0.742), and acute-care versus outpatient recording (0.839, 0.742–0.948). Both associations relative to no recorded CVD remained inverse across the tested hypothetical allocations of 455 missing-setting cases, whereas the acute-care-versus-outpatient point estimate approached or crossed the null under some scenarios. Secondary exploratory mediation analyses using a pooled score combining sleep, mental-health, and clinical-burden measures identified indirect associations for both settings versus no recorded CVD, but the between-setting indirect contrast was inconclusive. In this selected cohort, the findings highlight the importance of endpoint definition and recording completeness when interpreting wearable–EHR associations.