For anyone who has worn a fitness tracker and wondered whether that data could actually improve health outcomes, a newly released resource moves that possibility substantially closer to reality. A dataset of this scale and demographic breadth—linked to genomic, clinical, and survey records—creates an infrastructure that could redefine how wearable-derived signals are validated against hard medical endpoints, rather than remaining isolated lifestyle metrics.
The All of Us Research Program has released what amounts to one of the largest and most demographically diverse wearable health datasets assembled to date. Drawing on Fitbit devices distributed to participants across the United States, the repository captures continuous physical activity and sleep data from more than 59,000 individuals over a 14-year span, encompassing upward of 39 million step observations and 31 million sleep recordings. Crucially, nearly half of these participants also contributed electronic health records, physical measurements, genomic data, and survey responses—enabling multi-modal analysis that far exceeds what single-modality wearable studies can accomplish.
This release matters beyond its sheer size. Most prior wearable research has suffered from two structural weaknesses: homogeneous participant demographics and the absence of linked clinical or biological data. The All of Us cohort was explicitly designed to oversample historically underrepresented populations, addressing a persistent gap in digital health validation. The multi-modal linkage—connecting step counts and sleep architecture directly to genomic variants and clinical diagnoses—opens the door to causal inference approaches that observational wearable data alone cannot support. Researchers could, for instance, explore whether polygenic risk scores for cardiovascular disease correlate with objectively measured sedentary behavior, or whether sleep fragmentation patterns precede clinical diagnoses by months or years. Limitations remain: device wear compliance varies across participants, Fitbit's proprietary algorithms introduce measurement standardization questions, and the dataset is observational rather than interventional. Still, as a foundational infrastructure for population-scale digital health science, this resource is genuinely paradigm-shifting rather than merely incremental.