Physical activity (PA) guidelines recommend engaging in 150-300 minutes per week of moderate-intensity PA (MPA) or 75-150 minutes per week of vigorous-intensity PA (VPA), while minimizing sedentary behavior. However, since these recommendations are based on a single PA metric, they may not fully capture the complexity of real-world PA patterns, potentially hindering effective implementation. Digital phenotyping based on “PA patterns” may provide more actionable and individualized recommendations. In this study, participants are randomly divided into derivation and validation cohorts. In the derivation cohort, we apply Uniform Manifold Approximation and Projection (UMAP) to reduce high-dimensional accelerometer-derived PA data to two dimensions, followed by k-means clustering to identify distinct PA patterns. The clustering approach is then replicated in the validation cohort. Demographics and PA metrics are compared across clusters to characterize their unique features. The risks of cardiovascular events, kidney failure, and all-cause mortality are subsequently evaluated for each cluster. This analysis will identify favorable PA patterns-beyond the duration of a single PA metric-that are associated with better cardiovascular and kidney outcomes.
