Accelerometer-derived physical activity and mortality in individuals with type 2 diabetes

Type 2 diabetes (T2D) affects hundreds of millions globally and is associated with about 60% higher mortality than in non-diabetic adults. Physical activity (PA) is a key modifiable factor linked to lower mortality risk, yet people with T2D often face barriers to being active. Prior evidence in T2D largely relies on self-reported PA, which is prone to recall and measurement biases and poorly captures light-intensity PA (LPA). Objective accelerometry may provide more accurate estimates of PA–mortality associations. The study aims to quantify dose–response relationships between accelerometer-measured PA duration at different intensities (LPA, moderate-intensity PA [MPA], vigorous-intensity PA [VPA], and moderate-to-vigorous PA [MVPA]) and risks of all-cause, cancer, and cardiovascular disease (CVD) mortality in adults with T2D from the UK Biobank. The study also examines whether associations vary by diabetes severity indicators (glycemic control, diabetes duration, and medication use).

Meta-analyses and cohorts based on self-report indicate substantial reductions in all-cause mortality among adults with diabetes at higher PA levels (e.g., ~40% lower mortality comparing highest versus lowest PA; 14–37% lower mortality with ≥150 min/week MVPA or ≥75 min/week VPA). However, heterogeneity in PA categorization and reliance on self-report limit precision and the ability to specify exact PA volumes. Accelerometer-based studies in the general population show stronger associations (e.g., up to ~60% lower mortality with MVPA) than self-report studies. In T2D, accelerometer-based evidence is limited: small studies from NHANES, Look AHEAD, and a T2D-risk trial reported lower mortality or CVD outcomes with higher objectively measured PA but had smaller sample sizes (<2000), hip/waist placement (potentially missing upper-body activity), and lacked dose–response analyses. This study addresses these gaps using wrist-worn accelerometry and comprehensive dose–response modeling in a large T2D cohort.

Design and population: Prospective cohort analysis within the UK Biobank (application 79095). Participants provided informed consent. Among 502,401 participants, those with T2D at accelerometry baseline (2013–2015) were identified via multiple sources: self-report, insulin use or HbA1c ≥48 mmol/mol per ADA criteria, and ICD-10 E11 codes from linked health records. Of 32,709 with T2D, 4604 had accelerometry data; after exclusions for insufficient wear time (<72 h), daylight saving time shift during wear, and missing covariates, 4003 participants were included. Median follow-up was 6.9 years (through Nov 12, 2021). Exposure assessment: Participants wore an Axivity AX3 triaxial accelerometer on the dominant wrist for 7 days (100 Hz, ±8 g). Minutes/week were computed for intensity bands: LPA = 30–125 mg, MPA = >125–400 mg, VPA = >400 mg; MVPA = MPA + VPA. Outcomes: All-cause mortality; cancer mortality (ICD-10 C00–C97); CVD mortality (ICD-10 I00–I99), ascertained via NHS death registries. Covariates: Based on a DAG, included demographics (age, sex, ethnicity, education), accelerometer variables (season, wear duration), lifestyle (smoking, alcohol, diet score 0–7, sleep score 0–5), adiposity (BMI, waist circumference), health status (self-rated health; long-standing illness/disability; serious illness/injury/bereavement/stress in past 2 years), history of cancer, CVD, hypertension, and diabetes duration. Statistical analysis: Baseline characteristics summarized across MVPA categories. Dose–response associations modeled using restricted cubic splines within Cox proportional hazards models (rms package in R). Reference values set at the 1st percentile; knots at the 5th, 35th, 65th, and 95th percentiles. Nonlinearity assessed via Wald tests. PA durations categorized: LPA <1750, 1750–2099, 2100–2449, ≥2450 min/week; MPA <150, 150–299, 300–449, ≥450 min/week; VPA 0, 1–14, 15–29, ≥30 min/week; MVPA <275, 275–449, 450–624, ≥625 min/week. Cox models estimated hazard ratios (HRs) with 95% CIs for all-cause and cause-specific mortality. Three models: Model 1 adjusted for age, sex, ethnicity, education, season, wear duration; Model 2 added smoking, alcohol, diet, sleep; Model 3 (main) additionally adjusted for BMI, waist circumference, self-rated health, long-standing illness/disability, recent serious life events, history of cancer/CVD, history of hypertension, and diabetes duration. Age, season, and history of cancer/CVD were treated as time-varying covariates via interactions. Proportional hazards assumptions evaluated by Schoenfeld residuals (no violation). Joint associations of PA intensities assessed with a risk matrix (least active combinations as reference). Population-attributable fractions (PAFs) estimated for all-cause mortality comparing less active to most active groups. Stratified analyses by age, sex, BMI, waist circumference, smoking, alcohol intake, diet, sleep, and hypertension history; interactions tested by Wald tests. Sensitivity analyses: excluded baseline cancer/CVD for cause-specific analyses; excluded poor self-rated health; excluded deaths in first 2 and 4 years; stratified and adjusted for diabetes severity factors (HbA1c ≥7.0%, diabetes duration ≥10 years, insulin use); additionally adjusted for T2D-related complications; mutually adjusted LPA, MPA, VPA; stratified cancer deaths into T2D/obesity-related vs independent; multiple imputation by chained equations; competing risks (Fine and Gray) for cause-specific mortality; alternative PA metric using total MET-min/week; analyses by meeting vs not meeting guideline PA recommendations. Software: STATA 16 and R 4.1.3; two-sided P<0.05.

- Cohort: 4003 adults with T2D; mean age 64.9 years (SD 6.9); 63.0% male. Median follow-up 6.9 years; 339 all-cause deaths, including 158 cancer and 83 CVD deaths. - Dose–response: Longer PA duration at all intensities associated with lower all-cause mortality, showing L-shaped associations for LPA, MPA, VPA, and MVPA. Inflection-point risk reductions (vs least active 1st percentile): LPA ~1800 min/week, −60%; MPA ~300 min/week, −70%; VPA ~30 min/week, −45%; MVPA ~300 min/week, −70%. Beyond inflection, additional reductions were limited for VPA (≈1% up to +100 min/week) but notable for LPA (≈15% up to 3000 min/week), MPA (≈20% up to +1000 min/week), and MVPA (≈20% up to +1000 min/week). - Categorical analyses (Model 3): Compared with <150 min/week MPA, HRs (95% CI) for all-cause mortality were 0.61 (0.47–0.79) for 150–299, 0.41 (0.29–0.56) for 300–449, and 0.24 (0.15–0.36) for ≥450 min/week (P-trend<0.001). For MVPA vs <275 min/week, HRs were 0.54 (0.41–0.71) for 275–449, 0.36 (0.24–0.55) for 450–624, and 0.30 (0.16–0.54) for ≥625 min/week (P-trend<0.001). For LPA vs <1750 min/week, HRs were 0.90 (0.70–1.16) for 1750–2099, 0.60 (0.43–0.84) for 2100–2449, and 0.60 (0.39–0.93) for ≥2450 (P-trend<0.001). For VPA vs 0 min/week, HRs were 0.58 (0.44–0.75) for 1–14, 0.52 (0.37–0.73) for 15–29, and 0.43 (0.26–0.70) for ≥30 (P-trend<0.001). - Cause-specific mortality: L-shaped associations for cancer mortality and negative linear associations for CVD mortality across intensities. Compared with <150 min/week MPA, ≥450 min/week MPA associated with 63% lower cancer mortality (95% CI: 33%–80%) and 68% lower CVD mortality (95% CI: 21%–87%). VPA ≥30 min/week vs none: HR 0.61 (95% CI: 0.32–1.17) for cancer mortality and 0.35 (95% CI: 0.10–1.15) for CVD mortality. - Population-attributable fractions (all-cause mortality): For MVPA, PAFs were 47.1% for <275 min/week, 8.8% for 275–449, and 1.0% for 450–624 compared to ≥625 min/week. Lowest activity groups’ PAFs: LPA 18.8%, MPA 28.0%, VPA 31.1%. - Joint intensity combinations: Numerous combinations of LPA, MPA, and VPA yielded similar risk reductions in all-cause mortality (e.g., 300–449 min/week MPA and 2100–2449 min/week LPA: HR 0.35; 95% CI 0.19–0.63; 1–14 min/week VPA plus 2100–2449 min/week LPA: HR 0.33; 95% CI 0.17–0.63; 1–14 min/week VPA plus 300–449 min/week MPA: HR 0.35; 95% CI 0.22–0.54). - Robustness: Associations were consistent across strata (no significant interactions by age, sex, BMI, waist circumference, smoking, alcohol, diet, sleep, hypertension) and across multiple sensitivity analyses, including exclusion of early deaths and additional adjustments for diabetes severity and complications.

This study demonstrates that greater duration of accelerometer-measured PA at any intensity is associated with substantially lower risks of all-cause, cancer, and CVD mortality among adults with T2D. The L-shaped dose–response patterns, without clear lower or upper thresholds, support the message that any increase in PA is beneficial, including light-intensity activity. The results address prior uncertainties arising from self-reported PA by using objective, wrist-worn accelerometry, showing larger effect sizes and clarifying intensity-specific volumes linked to risk reduction. Importantly, clinically meaningful reductions in risk occurred at around 300 min/week of MPA and as little as 30 min/week of VPA, with diminishing returns beyond these levels for VPA. CVD mortality exhibited negative linear associations across intensities, suggesting continuous benefit with greater activity. The findings were robust to extensive adjustment, stratification by diabetes severity indicators, and multiple sensitivity analyses, indicating that confounding and reverse causation are unlikely to fully explain the associations. The joint intensity analyses offer practical flexibility, indicating that different combinations of LPA, MPA, and VPA can confer similar survival benefits, which is relevant for tailoring PA prescriptions in T2D to individual capabilities and preferences. These results have implications for refining PA guidelines for T2D, including potentially emphasizing benefits of LPA, recognizing strong gains up to ~300 min/week MPA, and considering lower VPA thresholds for suitable patients.

In a large cohort of adults with T2D using wrist-worn accelerometry, higher durations of PA across intensities were associated with markedly lower all-cause, cancer, and CVD mortality, with L-shaped dose–response relationships for all-cause and cancer mortality and linear inverse associations for CVD mortality. The study supports updating PA recommendations for T2D to emphasize benefits across the activity spectrum, including LPA, strong benefits around 300 min/week of MPA, and potential gains from as little as 30 min/week of VPA for eligible individuals. Future research should validate these associations in more diverse national cohorts with longer follow-up, refine accelerometer intensity thresholds using data-driven methods, evaluate additional PA features (fragmentation, diurnal and distributional patterns), and further address causality and potential selection biases.

- PA measurement period was 7 days; although correlated with longer-term PA in prior validation work, may not perfectly represent habitual activity, potentially biasing estimates toward the null if random error is present. - Wrist-worn accelerometers may under-capture activities with minimal arm movement (e.g., cycling) compared with ankle-worn devices, despite better intensity discrimination than waist-worn devices. - Only duration and intensity were analyzed; other PA characteristics (fragmentation, diurnal patterns, distribution) were not assessed. - Intensity cut-points (mg thresholds) may vary by device and demographics, introducing subjectivity; more objective, machine learning–based approaches are needed. - Observational design precludes causal inference; residual/unmeasured confounding and reverse causation cannot be fully excluded despite extensive adjustments and sensitivity analyses. - Potential selection bias: UK Biobank’s low response rate and availability of accelerometry in a subset; included T2D participants were generally healthier and more active than the broader T2D population, possibly limiting generalizability. - Some attenuation observed after excluding baseline cancer/CVD or early deaths, though overall patterns remained consistent.