A Local Enrichment Computational Framework to Identify Disease-Enriched Physical Activity Patterns

Wearable accelerometers enable continuous monitoring of physical activity (PA), yet conventional summary measures and population-wide clustering approaches may overlook disease-associated behavioral heterogeneity. Here we introduce LocPA, a computational framework that identifies disease-enriched PA patterns through confounder-adjusted activity profiling, similarity network construction, local disease-enrichment testing and community detection. Applying LocPA to 24-hour accelerometry profiles from the UK Biobank, we identified two recurring PA phenotypes across six diseases spanning cardiometabolic, neurological, psychiatric, renal and cardiovascular conditions: a low-activity phenotype characterized by reduced daytime activity and a sleep-disrupted phenotype characterized by reduced daytime and elevated nocturnal activity. In hypertension and diabetes, the sleep-disrupted phenotype was associated with greater disease burden and less favorable clinical and metabolic profiles, differences not consistently captured by conventional PA summary measures. LocPA also identified disease-specific temporal patterns, particularly in Parkinsons disease. Analyses in an independent National Health and Nutrition Examination Survey cohort supported the recurrence of hypertension- and diabetes-enriched patterns across populations and accelerometry measurements. LocPA provides a flexible, disease-driven approach for uncovering clinically relevant temporal PA patterns, with potential applications in disease risk characterization and personalized behavioral interventions.