Wearable accelerometer-derived physical activity and incident disease

Physical activity may have important health benefits, though the effects of physical activity across the range of human disease are poorly quantified. Examining relations between activity and disease risk may provide a comprehensive understanding of the benefits of physical activity, a modifiable lifestyle behavior. Prior studies often used self-reported activity, which is subject to recall bias and correlates modestly with measured energy expenditure, and many prior wearable sensor studies report metrics (e.g., step counts or raw acceleration) that are harder to map to consensus moderate-to-vigorous physical activity (MVPA) recommendations. Past work typically evaluated limited outcomes or focused on prevalent disease. To address these limitations, the authors leverage the UK Biobank cohort, where more than 90,000 individuals wore wrist-worn triaxial accelerometers for one week, to systematically identify associations between accelerometer-measured MVPA (as continuous minutes and as a binary threshold ≥150 min/week) and incident risk of approximately 700 diseases spanning the phenome. The study aims to inform future mechanistic research and guide prevention strategies by quantifying activity-disease associations and by benchmarking device-measured versus self-reported MVPA within the same population.

The paper outlines that numerous studies link higher physical activity to lower risk of cardiometabolic conditions, but most relied on self-report, which can be biased and imprecise. Device-based studies have shown associations between accelerometer-measured activity and reduced risks of cardiovascular and neurological disease and mortality; however, they often assessed limited outcomes or non-guideline metrics. The authors highlight the evolution and advantages of accelerometer methods and note that current activity guidelines are largely based on self-reported data, underscoring the need to contextualize device-derived MVPA against guideline thresholds and to examine a broad spectrum of disease outcomes.

Study population: UK Biobank prospective cohort (N=502,629) recruited 2006–2010 with linkage to national health records. Between 2013–2015, 236,519 were invited to a 1-week wrist accelerometer substudy; 103,695 submitted data. After quality control (excluding insufficient wear time <72 h, calibration failures, and missing covariates), 96,244 participants formed the primary accelerometer analysis sample. Separately, 456,374 participants provided self-reported activity via the short-form IPAQ at enrollment. Exposures: The primary exposure was accelerometer-derived MVPA minutes/week using an Axivity AX3 wrist triaxial accelerometer (100 Hz, ±8 g). Raw signals were calibrated to gravity, summarized in 5-second epochs (vector magnitude), and non-wear (≥60 minutes stationary; SD <13 mg on all axes) was imputed using time-of-day matching. MVPA was defined as epochs with mean acceleration ≥100 mg and aggregated in 5-minute bouts where ≥80% of epochs met the threshold to reduce misclassification of random wrist movement. Weekly MVPA was extrapolated for the small subset with >72 h but <7 days of wear. A binary guideline adherence exposure was defined as ≥150 min/week MVPA. Secondary exposures included overall mean acceleration (global activity surrogate) and minutes of vigorous activity (>430 mg). Self-reported MVPA was computed from IPAQ responses to mirror accelerometer-based MVPA and a self-reported guideline adherence variable (≥150 min/week) was also constructed. Covariates: Age, sex, BMI, smoking status, alcohol use (grams/week), systolic and diastolic blood pressure, anti-hypertensive medication use, Townsend Deprivation Index (socioeconomic), educational attainment (years), and diet quality (poor/intermediate/good). Outcomes: Incident diseases were defined using Phecode Map v1.2 (1867 phecodes grouped into clinically meaningful categories) mapped from ICD-9/10 codes via linked hospital and general practitioner electronic health records. For accelerometer analyses, person-time began at the end of accelerometer wear; for self-report analyses, at enrollment; follow-up ended at event, death, or last available date (England/Scotland: March 31, 2021; Wales: February 28, 2018). Pregnancy conditions and congenital anomalies were excluded; only diseases with ≥120 events were tested, yielding 697 conditions in the primary analysis. Statistical analysis: Associations between accelerometer-derived MVPA (per 150 min/week, approximately one SD) and incident disease were tested using Cox proportional hazards models adjusted for the covariates above. Multiplicity was addressed using a 1% false discovery rate (FDR) via tail-area based thresholds (R package fdrtool). E-values were reported for point estimates and the CI bound closest to the null to assess robustness to unmeasured confounding. Additional models assessed binary guideline adherence (≥150 min/week) and self-reported MVPA exposures, as well as overall mean acceleration. Risk stratification plots showed 5-year cumulative incidence for exemplar diseases (heart failure, type 2 diabetes, cholelithiasis, chronic bronchitis) by guideline adherence; sex-specific adjusted risk curves were also generated. Dose-response was assessed via hazard ratios across MVPA quintiles relative to the lowest quintile. Secondary analyses included: age subgroup analyses (<55, 55–64, ≥65 years); alternative MVPA thresholds (≥75 and ≥300 min/week); vigorous activity; landmark analysis starting person-time 2 years post-accelerometer wear; hospital-only outcome definitions; and models excluding BMI and blood pressure covariates to explore potential mediation. Analyses used R v4.0.3 with survival, data.table, and fdrtool packages.

• Sample: 96,244 participants with valid accelerometer data (mean age 62 ± 8 years; 56% female); median MVPA 135 min/week (Q1: 60, Q3: 250); 46% met ≥150 min/week guideline threshold. Median follow-up 6.2 years (Q1: 5.7, Q3: 6.7). - Primary associations: MVPA was associated with risk of 373/697 (54%) incident diseases at FDR 1%. Of significant associations, 367 (98%) indicated lower risk with greater MVPA (HR per +150 min/week MVPA: 0.70–0.98). Strong associations included: atherosclerosis HR 0.57 (95% CI 0.44–0.74), type 2 diabetes HR 0.74 (0.70–0.79), chronic bronchitis HR 0.44 (0.37–0.53), depression HR 0.84 (0.79–0.88). - Disease category enrichment among lower-risk associations: cardiac (16%), digestive (14%), endocrine/metabolic (10%), respiratory (8%), with significant associations observed across all 16 categories (chi-square p < 0.01). Largest median effect sizes were seen in endocrine/metabolic, respiratory, and infectious diseases. - Higher-risk associations: 6 associations indicated higher risk with greater MVPA (HR 1.08–1.24), all within musculoskeletal, injuries/poisonings, or dermatologic categories (e.g., disorders of muscle/ligament/fascia HR 1.09 [1.03–1.15]; fracture of radius/ulna HR 1.09 [1.02–1.15]). - Guideline adherence (≥150 min/week): 306 significant associations with lower risk (HR range 0.11–0.91). Notable reductions: heart failure HR 0.65 (0.55–0.77), type 2 diabetes HR 0.64 (0.58–0.71), cholelithiasis HR 0.61 (0.54–0.70), chronic bronchitis HR 0.42 (0.33–0.54). Cumulative 5-year risk curves showed clear stratification by accelerometer-derived guideline adherence. - Dose-response: For most conditions, higher MVPA quintiles showed progressively lower risk; in a minority (notably musculoskeletal/injury categories), lowest risk was at intermediate MVPA levels. Alternative thresholds (≥75 and ≥300 min/week) showed similar patterns, with fewer significant associations at ≥300 min/week. - Self-reported activity: Within 456,374 participants, self-reported MVPA showed qualitatively similar patterns but substantially smaller effect sizes (e.g., heart failure HR 0.84 [0.80–0.88] for guideline adherence) and more associations indicating greater disease risk, primarily in musculoskeletal and injuries/poisonings. - Secondary analyses: Results were consistent using overall mean acceleration and for vigorous activity (fewer associations). Associations were robust across age subgroups, hospital-only outcomes (n=343 significant), when excluding potential mediators (BMI/BP) increased significant associations (n=500), and in a 2-year landmark analysis (n=259 significant).

The study demonstrates that greater device-measured MVPA is broadly associated with lower incidence of hundreds of diseases across all major categories, directly addressing the research question of how physical activity relates to incident disease risk across the phenome. Associations were strongest and most numerous for cardiac, digestive, endocrine/metabolic, and respiratory conditions, yet present across all categories, underscoring the pervasive health relevance of physical activity. Both continuous MVPA and guideline adherence showed substantial risk reductions, and dose-response analyses indicated progressively lower risk with higher MVPA, suggesting benefits even below the guideline threshold and continued benefits above it. Comparisons with self-reported activity within the same cohort revealed that accelerometer-derived measures provided stronger effect sizes and more robust risk stratification, highlighting the superior information content of device-based activity measurement and potential biases in self-report. The findings extend prior literature by using a wearable-based, guideline-contextualized MVPA metric and phenome-wide outcome assessment, suggesting that optimizing objectively measured physical activity could serve as a powerful lever for disease prevention and risk stratification. Differences between accelerometer and self-reported associations, including more frequent higher-risk associations in self-report for musculoskeletal/injury categories, suggest distinct information content and measurement error characteristics that warrant further investigation.

Using wrist-worn accelerometry in over 90,000 individuals, the study quantified phenome-wide associations between MVPA and incident disease, revealing that higher device-measured activity and adherence to ≥150 min/week are associated with substantially lower risks for more than 350 conditions spanning all major disease categories. Device-measured activity outperformed self-reported activity in both the number and magnitude of significant associations. These results prioritize diseases for mechanistic follow-up, support the integration of wearable-based activity monitoring in public health and preventive cardiometabolic strategies, and suggest that increasing objectively measured MVPA—even below traditional thresholds—may reduce future disease incidence. Future research should: (1) identify causal pathways linking activity to specific diseases; (2) refine accelerometer-based thresholds aligned with health outcomes; (3) evaluate interventions that use device-based activity targets to reduce incident disease; and (4) assess generalizability across diverse populations and settings.

• Activity measured over one week may incompletely capture habitual behavior; longer monitoring might improve classification. - Comparisons between accelerometer and self-report may be biased by differing exposure timing and follow-up windows. - Observational design: despite adjustment, E-values, and a 2-year landmark analysis, residual confounding and reverse causality may persist; causal inference is not established. - Outcome definitions rely on diagnosis codes; individuals with lower activity might have different healthcare contact patterns, potentially biasing incident diagnoses. - Accelerometer processing choices (e.g., cut-points, bouting, non-wear imputation) can influence exposure classification and comparability across studies. - Guideline thresholds (≥150 min/week) are based largely on self-reported data; optimal accelerometer-derived thresholds may differ and remain to be defined. - Some subgroup analyses (e.g., younger age strata) may be underpowered due to fewer events. - Generalizability may be limited: UK Biobank participants are comparatively healthy, predominantly White, and may alter behavior due to being monitored (Hawthorne effect).