Our analysis was based on data from the UK Biobank study . Briefly, UK Biobank is a large population-based, prospective cohort study that recruited 502,505 participants aged 40–69 years in the UK from 2006 to 2010 (protocol available at https://www.UKbiobank.ac.UK/key-documents/). All participants provided questionnaire information, physical measurements, and biological samples at the baseline. UK Biobank was approved by the North West Multi-Center Research Ethical Committee (REF: 11/NW/03820). All participants gave written informed consent before enrolment in the study, which was conducted in accordance with the principles of the Declaration of Helsinki.
Among the 502,505 participants recruited at baseline, the present study excluded participants who have withdrawn from the UK Biobank (n = 46), pregnant women (n = 371), those with baseline CVD (n = 36,449), and those with missing information on physical activity (n = 91,624) or reported no physical activity (n = 7,449), leaving 366,566 participants for the final analytic sample (see Additional file 1). Baseline CVD was ascertained by self-reported information and hospital records . The distribution of baseline characteristics was comparable among participants before and after exclusion (see Additional file 2).
Physical activity assessment
Data on physical activity was obtained through questionnaires adapted from the IPAQ short version (https://biobank.ndph.ox.ac.uk/showcase/browse.cgi?id=1008&cd=browse), which was reported to have acceptable test–retest reliability and criterion validity in the UK population . Accordingly, we multiplied the energy expended for a specific category of activity (MET: 3.3 for walking, 4.0 for moderate activity and 8.0 for vigorous activity) [15, 16] by the corresponding frequency (times/week) and duration (minutes/time) and summed the corresponding amount to estimate MET-minutes/week of total MVPA. In the present study, we further defined MPA as physical activity of 3.0–5.9 METs therefore MET-minutes/week of walking and moderate activity were summed as the total amount of MPA, and VPA as physical activity of 6.0 METs or more such as running [15, 16]. Among participants with any MVPA, the proportion of VPA to MVPA was calculated by dividing the amount of VPA by the amount of MVPA. The proportion of VPA to MVPA was categorized into one of the three categories: 0% (no VPA), > 0% to ≤ 30%, and > 30%. The 30% cut-off was chosen according to a previous study, which showed a maximum reduction in risk of all-cause mortality appeared at 30% of VPA to MVPA .
The present study included demographic characteristics (age, sex [male or female]), sociodemographic characteristics (education [college/university or below college], income [< 18,000 or 18,000–52,000 or ≥ 52,000 ￡/year] , race [white or others], Townsend index), lifestyle factors (smoking status [never or current or former], alcohol consumption [0 or 0.1–30 or ≥ 30 g/day] [19, 20], diet quality score [see Additional file 3], sedentary behavior [hours/day] and MVPA [MET-minutes/week]), body mass index (BMI, kg/m2) and family history of CVD (yes or no) as covariates according to previous studies on similar topic [5, 7, 17]. Missing indicator approach was used in the regression model by coding missingness as a dummy variable.
Ascertainment of outcome
UK Biobank data are linked to Hospital Episode Statistics (HES; hospital diagnoses from the National Health Service) and the UK Biobank Cause of Death Registry . Incident CVD and mortality were determined based on the International Classification of Diseases edition 10 (International Classification of Diseases, ICD-10) . HES and mortality data were updated up until December 31, 2020. The primary outcome was incident CVD (ICD-10: I00-I99), its main subtypes including CHD (I20-25), heart failure (HF; I50, I500, I501, I509), and stroke (I60, I61, I63, I64). The secondary outcome was all-cause mortality, its main components including CVD mortality.
Standardized difference was calculated to compare the baseline characteristics across three groups of VPA to MVPA (0%, > 0% to ≤ 30% and > 30%) . A standardized difference less than 0.20 indicates a small difference of characteristic across different groups of VPA to MVPA (0%, > 0% to ≤ 30% and > 30%) . Cox proportional hazard models were used to calculate the hazard ratios (HRs) and 95% confidence intervals (CIs) for incident CVD and all-cause mortality according to three groups of VPA to MVPA. Models were adjusted for age, sex, education, income, race, Townsend index, smoking status, alcohol consumption, sedentary behavior (hours/day), MVPA (MET-minutes/week), BMI (kg/m2), diet quality score, and family history of CVD. To test for P trend, participants were assigned to the median value of each group of VPA to MVPA (0%, > 0% to ≤ 30% and > 30%), and then this continuous variable was entered into the Cox model . Comparison between adjusted HRs for the first half of follow-up years and for the subsequent years revealed no evidence of departure from the proportional hazards assumption for main analyses.
In order to explore the potential interaction, stratified analyses were performed by socio-demographic characteristics (age, sex, education, Townsend index), lifestyle risk factors (smoking status, alcohol consumption, total MVPA, sedentary behavior), BMI and chronic diseases (baseline hypertension and diabetes) , with adjustment for covariates the same as in the main model except for the stratified variable itself. The interaction effect of VPA to MVPA with stratified variable was assessed by introducing a multiplicative interaction term into the Cox models and the Wald test was used to calculate the P value for the interaction term .
We used the restricted cubic spline function to delineate the continuous exposure–response association between the proportion of VPA to MVPA and incident CVD and all-cause mortality with the %LGTPHCUTV9 macro, which fits restricted cubic splines to proportional hazards regression models to examine non-parametrically the relation between an exposure and the incidence rate ratio of the outcome of interest . The output includes the set of P-values from the likelihood ratio tests for non-linearity, a linear association, and any association. We set three knots at the 50th, 75th and 95th percentiles of the proportion of VPA to MVPA.
To find the optimal combination of MPA and VPA, we also conducted an analysis to examine the joint association of MPA (0 to < 150, 150 to < 300, ≥ 300 min/week) and VPA (0 to < 75, 75 to < 150, ≥ 150 min/week)  with different outcomes by creating a combined variable with 3×3 mutually exclusive groups, taking the lowest MPA and VPA as reference.
Finally, to test the robustness of our main findings, we conducted sensitivity analyses by extended adjustment for hypertension, diabetes status, medical center, lipid-lowering treatment, antihypertensive medications, and diabetes medication, by extended adjustment for employment information, by conducting competing risks analyses to compare end point-specific survival between the proportion of VPA and CVD events using the Fine and Gray competing risk model , by excluding participants who developed CVD or died within the first two years of follow-up, and by excluding participants with missing covariates.
All statistical analyses were performed using the SAS 9.4 (SAS Institute, Cary, North Carolina, USA) and R (R Foundation for Statistical Computing, Vienna, Austria). Statistical significance was defined as a two-sided P < 0.05 except for the subgroup analyses, in which statistical significance was set at P < 0.0045 (0.05 / 11 subgroups) to account for the possible type I error generating from multi-comparisons .