Wearable device metrics link to hundreds of human diseases
In 70,473 people tracked for up to 10 years, wearable metrics mapped 8,149 disease associations and predicted conditions including incident Parkinson's disease and dementia.
In a study of 70,473 human participants followed for up to 10 years, researchers used wearable accelerometers to analyze 11.8 million person-hours of behavioral data. The authors extracted 210 movement-derived phenotypes across 406 diseases and 956 health-related traits, identifying 8,149 disease associations. Step intensity and circadian rhythms emerged as the behavioral domains most strongly linked to disease. Using these data, the team built a framework named WATSNet, which achieved discrimination scores above an AUC of 0.7 for 112 incident diseases, including incident Parkinson's disease at an AUC of 0.873 and dementia at 0.841. Lower overall disease risks were associated with at least 0.8 hours daily of moderate-to-vigorous activity, 10,000 steps per day, and 7.8 hours of sleep.
Why it matters
Disrupted circadian rhythms and declining physical activity are hallmark features of aging. The findings demonstrate that digital tracking of these behaviors can help identify risks for age-related conditions like dementia long before clinical diagnosis.
Caveats
Because the design is observational, the reported behavioral benchmarks and phenotype associations do not demonstrate that changing activity patterns directly prevents disease. Predictive performance was externally evaluated in a smaller group of 4,406 individuals.
The paper
A wearable accelerometry landscape of human health and disease
Liang Y, Zhao Y, Deng Y et al.
Science Bulletin · 22 Sep 2026
