Data source and study population
The UK Biobank study recruited about half of a million adults aged 37–73 years in 22 study centers across England, Scotland, and Wales between 2006 and 2010. At baseline, comprehensive information covering sociodemographic variables, lifestyle factors, anthropometric indices, and health status were collected through touchscreen electronic questionnaires, face-to-face verbal interviews, as well as physical and biological measurements. Further information on the study design and procedures has been published previously [21]. The UK Biobank study was approved by the North West Multi-Center Research Ethics Committee (R21/NW/0157). All participants provided written informed consent.
This study was restricted to a subsample of participations in the accelerometry sub-study (n = 103,661). Prevalent hypertension at baseline was identified based on self-reported history of hypertension, current use of antihypertensive medications, SBP/DBP ≥ 140/90 mmHg, or diagnostic records from primary care and hospital admission [22]. Participants with insufficient accelerometer wear time (data for fewer than 3 days or missing data for any 1-h period of the 24-h cycle, n = 6992), poorly calibrated accelerometer data (n = 4), implausibly high accelerometer values (average vector magnitude scores > 100 g, n = 13), and missing data on daily MVPA (n = 1194) were excluded. Additionally, we further excluded participants without hypertension at baseline (n = 47,256), who were diagnosed with CVD prior to the accelerometry measurements (n = 5328), and who had missing data on covariates (n = 841). Moreover, participants with less than a full week of acceleration data (n = 1750) were excluded to ensure accurate classification of PA patterns based on the actual distribution of daily MVPA. Ultimately, 40,283 adults with hypertension were included in the final data analyses. Figure 1 shows the detailed flowchart of participants’ selection process.
Selection of study participants from the UK Biobank
PA patterns
Between June 2013 and December 2015, participants with a valid email address were invited to wear an Axivity AX3 accelerometer (Newcastle upon Tyne, UK) on their dominant wrist for 7 consecutive days to objectively measure PA [23]. Accelerometers were configured to capture triaxial accelerometer data with a sampling frequency of 100 Hz and a dynamic range between ± 8 gravity units. After the devices were returned, all acceleration signals were calibrated to local gravity. Non-wear time was identified using standard procedures, defined as consecutive episodes of at least 60 min during which standard error of each axis was less than 13.0 g. Details on data collection and procedure can be found at the related website (https://biobank.ndph.ox.ac.uk/showcase/refer.cgi?id=131600). Proportions of time spent in various movement behaviors (i.e., light intensity PA, MVPA, sedentary behavior, and sleep) were derived using a machine learning algorithm trained on an external independent cohort of 152 individuals in free-living conditions [24]. This algorithm demonstrated a good performance in a previous study, with a mean accuracy of 0.88 and a Cohen’s kappa of 0.80 [24].
Participants were divided into three groups as follows [16, 25]: the inactive group (MVPA < 150 min/week), the regularly active group (MVPA ≥ 150 min/week with ≥ 50% distributed over > 2 days/week), and the WW group (MVPA ≥ 150 min/week with ≥ 50% concentrated on 1–2 days/week). For example, if a participant accumulated 200 min of MVPA during the week, and 120 min (≥ 50%) occurred on Friday and Sunday combined, the participant was classified into the WW group. In contrast, if the total MVPA volume was more evenly distributed across the week (e.g., 30–40 min per day) and the combination of MVPA over any 2 days was less than 100 min, the participant was classified as having the regularly active pattern. As the optimal MVPA duration was unclear in individuals with hypertension, we also performed three sensitivity analyses to assess the robustness of our findings using additional MVPA thresholds to define PA patterns [16], including ≥ 25th percentile (100.8 min/week), ≥ 50th percentile (216.0 min/week), and ≥ 75th percentile (388.8 min/week) of the total MVPA volume in our study population.
Outcomes
The primary outcome in this study was overall CVD, encompassing fatal and non-fatal myocardial infarction (MI), atrial fibrillation (AF), heart failure (HF), and stroke, which was ascertained through self-reported medical history and linkage to hospital inpatient records and death registrations. Hospital inpatient records were provided by the Hospital Episode Statistics for England, the Patient Episode Databases for Wales, or the Scottish Morbidity Record for Scotland. Death registration data were obtained from the UK National Health Service (NHS) Digital of England and Wales or the NHS Central Register and National Records of Scotland. Incident cases were identified based on the 10th vision of International Classification of Diseases (ICD-10). Detailed disease definitions and their corresponding ICD-10 codes are available in Additional file 1: Table S1. The follow-up period spanned from the date of accelerometry data collection to the first occurrence of a CVD event, death, or the end of follow-up (30 November 2022), whichever came first.
Covariates
Selected covariates in this study were based on prior knowledge and previous analyses of WW pattern using the UK Biobank data [16, 17, 26]. Covariate definitions are provided in Additional file 1: Table S2. Briefly, the following covariates were included: age at the end of accelerometer wear (years, calculated from the date of birth and the completeness of accelerometer wear), sex (women and men), ethnicity (white and non-white), Townsend deprivation index (TDI, a comprehensive surrogate of socioeconomic status) [27], education (college or university degree, A levels/AS levels or equivalent, O levels/GCSE or equivalent, CSE or equivalent, NVQ or HND or HNC or equivalent, and others), employment status (employed and unemployed), smoking status (never, ever, and current), frequency of alcohol drinking (never, ≤ 1–2 times/week, and ≥ 3 times/week), sleep duration (< 7, 7–8, and > 8 h/day), healthy diet score (0–7 points, comprising fruits, vegetables, fish, processed meats, unprocessed red meats, whole grains, and refined grains) [26], body mass index (BMI, kg/m2), SBP/DBP (mmHg), glycated hemoglobin (HbA1c, mmol/mol), triglycerides (TG, mmol/L), low-density lipoprotein cholesterol (LDL-C, mmol/L), high-density lipoprotein cholesterol (HDL-C, mmol/L), and the use of antihypertensive, antihyperglycemic, and lipid-lowering medications (yes and no). Given the potential mediating roles of cardiometabolic factors (including BMI, SBP, DBP, HbA1c, TG, LDL-C, HDL-C, and use of antihypertensive, antihyperglycemic, and lipid-lowering medications) in the association between PA patterns and CVD outcome, these covariates were not adjusted for in the primary analyses but were included in the sensitivity analyses.
Statistical analyses
Baseline population characteristics were presented according to different PA patterns. Continuous variables were expressed as mean (standard deviation, SD) and categorical variables were showed as number (%). Group differences were tested by analysis of variance (ANOVA) or chi-square tests, as appropriate.
Kaplan–Meier survival curves were applied to intuitively visualize the cumulative risks of incident CVD stratified by PA patterns, with log-rank tests performed to examine differences between three groups. Multivariable-adjusted Cox proportional hazards regression models were performed to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations between PA patterns and the risk of incident CVD and its four subtypes. To account for different effects of confounding factors, three models were applied. Model 1 was adjusted for age, sex, and ethnicity. Model 2 was additionally adjusted for socioeconomic status (including TDI, education, and employment status). Model 3 was further adjusted for four key lifestyle factors (including smoking status, alcohol drinking, healthy diet score, and sleep duration). The proportional hazard assumption for Cox regression models was assessed using Schoenfeld residuals, with no evidence of violation observed (all P values > 0.05).
Stratified analyses were performed by age (< 65 and ≥ 65 years), sex (women and men), ethnicity (White and non-white), BMI (< 25 kg/m2 and ≥ 25 kg/m2), smoking status (never and ever/current), alcoholic drinking (< 1 time/week and ≥ 1 time/week), diet quality (healthy diet score < 4 and ≥ 4 points), and daily sleep duration (7–8 h/day and < 7/> 8 h/day) to examine whether associations varied across different subgroups of potential covariates. Effect modification by subgroup of included covariates was examined via the Z-test [28].
To assess the robustness of our primary results, we further conducted a series of sensitivity analyses. First, three additional thresholds of total MVPA, including the 25th, 50th, and 75th percentile of total MVPA volume in the study population, were used to re-define PA patterns. Second, the WW pattern was re-classified based on alternative definitions, including ≥ 75% of total MVPA concentrated on 1–2 days, ≥ 50% of total MVPA concentrated on 1–2 consecutive days, and ≥ 50% of total MVPA concentrated on 1–2 weekend days. Third, to minimize potential reverse causation, landmarks analyses were conducted after excluding participants who developed new CVD events within the first 2 years of follow-up period. Fourth, we additionally adjusted for several potential mediators including BMI, SBP, DBP, HAb1c, TG, LDL-C, HDL-C, and the use of antihypertensive, antihyperglycemic, and lipid-lowering medications. Fifth, to reduce the potential influence of baseline comorbidities, we performed a sensitivity analysis with additional adjustment for the history of diabetes, hyperlipidemia, chronic lung disease, gastrointestinal disorders, chronic kidney disease, and liver disease. Sixth, covariates with missing values were imputed using the multiple imputation method to examine the potential influence of missing data of covariates. Seventh, we additionally adjusted for wear time, the season of accelerometer wear, and total MVPA volume to reduce the potential impacts of measured variance and PA volume. Finally, participants without full-week accelerometer data were classified as physically inactive.
All statistical analyses were performed with R (software version 4.4.0, R Development Core Team, Vienna, Austria). A two-sided P value < 0.05 was considered as statistically significant.