WHAT IS ALREADY KNOWN ON THIS TOPIC

WHAT THIS STUDY ADDS

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE or POLICY

From the wrist accelerometry substudy activity data, we included only cancer survivors with valid accelerometer data (online supplemental methods). We further excluded participants with baseline CVD (coronary artery disease (CAD) [ICD-10, I21-25], heart failure (HF) [ICD-10, I50] or stroke [ICD-10, I60-I61, I63-I64]) or missing covariates (figure 1, online supplemental figure S1). Cancer data linkage was obtained through national cancer registries. We examined all cancers (C00–C97, except for non-melanoma skin cancer C44, in situ and non-well-defined cancers).16 Methods for cancer assessment are provided in the online supplemental methods section. Detailed coding and numbers of site-specific cancers are presented in online supplemental table S1. The sample size justification related to the different endpoints is presented in the online supplemental methods.

Cancer types were further categorised as obesity-related and non-obesity-related based on the IARC criteria12 because previous studies on PA have mainly focused on obesity-related cancers. Obesity-related cancers include endometrial, oesophageal, stomach, liver, kidney, multiple myeloma, brain, pancreatic, colorectal, gallbladder, breast, ovarian and thyroid cancers.17

MVPA behaviours were classified using a previously published two-stage machine-learning Random Forest activity classifier developed and validated for use with the UK Biobank18 19 (see online supplemental methods). The corresponding MVPA behaviours are detailed in online supplemental table 2). This machine-learning approach was based on the CAPTURE-24 study containing the largest free-living dataset and demonstrated superior performance in practice. Compared with traditional ‘vector magnitude’ methods, the machine-learning approach leverages multiple data features to classify behaviours with greater accuracy.19 The MVPA volume was further categorised into four groups based on guideline-recommended weekly MVPA volume (>0 to <75 min/week, 75 to <150 min/week, 150 to <300 min/week and ≥300 min/week).

The primary outcome was the occurrence of CVD, including CAD (ICD-10, I21-25), HF (ICD-10, I50) or stroke (ICD-10, I60-I61, I63-I64). The incident CVD information was gathered from self-report, primary care and hospital admission sources. Participants were followed up from the completion of accelerometer wear to the incidence of CVD, loss to follow-up, death or the end of follow-up on 31 December 2022, whichever occurred first.6 . Online supplemental tables S3 and S4 contain complete definitions of outcomes and variables, as well as selection rationales.

Complete-case analysis was conducted in this study. Descriptive characteristics were reported as mean (SD) for normally distributed variables or median (IQR) for skewed continuous variables and number and percentage for categorical variables.

Analysis in Model 1 showed an unadjusted HR, while Model 2 was adjusted for age, sex, ethnicity, Townsend index of deprivation, educational levels, smoking status, alcohol intake, diet score, sleep duration, medications (lipid-lowering, antihypertensive, antidiabetic), self-reported parental history of CVD or cancer and time from cancer diagnosis to accelerometer-worn completion. The selection of variables was guided by a directed acyclic graph (DAG), developed through a comprehensive literature review. This process is outlined in detail in online supplemental table S4, with the resulting DAG shown in online supplemental figure S2. Model 2 served as the main analysis.

The dose-response analysis of MVPA duration with outcomes was conducted using restricted cubic splines, with knots positioned at the 10th, 50th and 90th percentiles according to the previous literature,20–22 and the reference group set to zero minutes per week. For HR analyses, the departure from linearity was assessed by a Wald test examining the null hypothesis that the coefficient of the second spline was equal to zero.22 Poisson regression models with log-transformed follow-up time as the offset were used to estimate the crude and adjusted incidence rates across MVPA groups, reported as events per 1000 person-years. Model assumptions were evaluated by calculating the ratio of residual deviance to df for overdispersion and using residual distribution plots to assess the linearity of quantitative predictors. Adjusted incidence rate differences and rate ratios were estimated using g-computation with standard parametric regression models.23 Adjusted HR with 95% CI was estimated using Cox proportional hazards regression models. Cox proportionality assumptions were assessed using Schoenfeld residuals, with no observed violations. The cumulative risks of CVD across MVPA groups were compared as adjusted survival curves.24 25 We also calculated the 5 year absolute risk across MVPA groups using Fine and Grey competing risks regression models. Exploratory analyses were further conducted in survivors of obesity-related and non-obesity-related cancers, respectively.12 17

In sensitivity analysis, we further excluded people with diseases of the circulatory system (I0, I11, I13, I20-I51, I60-I69) at baseline. A landmark analysis was performed by delaying the start of follow-up by 2 years following accelerometer-worn completion to account for potential reverse causality. We employed the Fine and Grey and cause-specific hazard model to account for competing risks (non-CVD deaths as competing risks). Body mass index (BMI) may mediate the relationship between PA and CVD, so it was excluded from the main analysis. However, as BMI could also act as a confounder, an additional analysis was conducted with BMI adjustment. Furthermore, we classified MVPA based on WHO standards: below WHO guidelines (0–150 min), meeting WHO guidelines but below the extended recommendation (150–300 min) and exceeding the extended WHO recommendation (≥300 min) to further validate the robustness of our results. An additional analysis was performed using covariates from the interview closest to the accelerometer assessment. Missing covariate values were imputed using multiple imputation by chained equations (see online supplemental methods and online supplemental tables S5 andS6). Finally, multiplicative and additive interaction (using the ‘interactionR’ package in R) analyses and subgroup analyses were performed. All analyses were performed using R version 4.3.1.

The author group is gender balanced, consisting of junior, mid-career and senior researchers from several disciplines, including cardiology, oncology, computing technology and sport. Our study cohort included male and female cancer survivors of different race and ethnicity, but the majority were white.

The study sample comprised 6109 cancer survivors, with a mean (SD) age of 65.4 (6.9) years (table 1). Of these, 3722 (60.9%) were women, 2387 (39.1%) were men, and 5991 (98.1%) were White individuals. The most common specific cancer types were breast (n=2063, 33.8%), prostate (n=1151, 18.9%), small intestine (n=593, 9.7%), melanoma skin (n=506, 8.3%) and lymphatic and haematopoietic tissue (n=457, 7.5%). Details of the number of cancer survivors by specific types are presented in online supplemental table S1. Cancers were further divided into obesity-related (n=3401, 55.7%) and non-obesity-related (n=2708, 44.3%) categories.

The median time between cancer diagnosis and completion of accelerometer wear was 6.9 (IQR, 3.1–12.0) years (table 1). The median duration of MVPA was 201.0 (IQR, 90.0–358.8) min per week, with 1768 (29.0%) cancer survivors meeting the standard guideline recommendation of 150–300 min/week, and 1980 (32.4%) meeting the extended recommendations of ≥300 min/week. Individuals with longer MVPA duration were more likely to be younger, male, nonsmokers, have a higher educational level, lower BMI and blood pressure levels, report better diet quality and use fewer medications.

During a median follow-up of 8.01 (IQR 7.95, 8.04) years, there were 539 incident CVD events, including 361 incident CAD events, 155 incident HF events and 109 incident stroke events. And 451 participants died from other causes. An inverse near-linear dose-response relationship between MVPA volume and the incidence of CVD was observed (P for nonlinear=0.49), with no maximal threshold for the benefits (online supplemental figure S3). The adjusted rates of CVD incidence (95% CIs) across MVPA groups (0–75 min/week, 75–150 min/week, 150–300 min/week, ≥300 min/week) were 15.30 (12.90, 18.10), 13.5 (11.00, 16.40), 12.00 (10.20, 14.10) and 9.86 (8.35, 11.60) per 1000 person-years, respectively (table 2). Adhering to MVPA within the recommended duration (150–300 min per week) and surpassing it (≥300 min per week) were associated with a 23% (HR, 0.77, 95% CI, 0.61 to 0.97) and 37% (HR, 0.63, 95% CI, 0.49, 0.80) lower hazard of CVDs, respectively, compared with those with the lowest PA duration (0–75 min per week) (figure 2). For specific types of CVD, patients with longer MVPA duration were associated with a reduced risk of CAD (≥300 MVPA min/week vs 0–75 MVPA min/week: HR, 0.68 [0.51, 0.91], p=0.01, S-value=6.64). There was weaker evidence for associations between MVPA and HF (≥300 MVPA min/week vs 0–75 MVPA min/week: HR, 0.66 [0.42,1.06], p=0.08, S-value=3.6) or stroke (≥300 MVPA min/week vs 0–75 MVPA min/week: HR, 0.72 [0.42, 1.23], p=0.23, S-value=2.12). However, the data are still compatible with the potential protective effects of MVPA on these outcomes, as suggested by the direction of the hazard ratios. Adjusted survival curves for CVD and subtypes of CVD across MVPA groups were depicted in online supplemental figure S4. Furthermore, we employed the Fine and Gray competing risks regression model to calculate the 3 and 5 year cumulative CVD risks presented in online supplemental figure S5. Likewise, a higher MVPA volume reduces the CVD incidence across different subgroups.

The CVD benefits remained consistent regardless of cancer type. There was an inverse near-linear relationship between MVPA duration and CVD risk (non-obesity-related cancers: P for nonlinearity=0.51, P for trend=0.035; obesity-related cancers: P for nonlinearity=0.82, P for trend<0.001) (online supplemental figure S6, table 3). For non-obesity-related cancers, the HR (95% CI) for 75–150 min of MVPA per week was 0.65 (0.44–0.97), for 150–300 min of MVPA per week was 0.70 (0.51–0.97) and for exceeding 300 min MVPA/week was 0.64 (0.46–0.88), compared with the lowest MVPA group (0–75 min/week). For obesity-related cancers, the reduced HRs (95% CI) are 0.82 (0.59–1.16) and 0.54 [0.37–0.81] for 150–300 min of MVPA per week and exceeding 300 MVPA minutes per week, respectively.

The inverse association between MVPA duration and CVD risk remained consistent across various subgroups, including age of cancer diagnosis (<60, ≥60), sex (male, female), educational background (university degree or not), smoking status (never, current or previous), alcohol intake (<median, ≥median), sleep duration (<7 hours/day, ≥7 hours/day), diet score (0–1, 2–3), hypertension (no or yes), diabetes (no or yes), BMI (<25 kg/m2, ≥25 kg/m2), low-density lipoprotein-cholesterol (<3.4 mmol/L, ≥3.4 mmol/L) and triglycerides (<1.7 mmol/L, ≥1.7 mmol/L) (online supplemental table S10). We did not observe statistically significant additive or multiplicative interactions for any other risk factors (online supplemental table S10).

In sensitivity analyses, adjusting for BMI (online supplemental figure S7), excluding people with diseases of the circulatory system (online supplemental figure S8), excluding the first 2 years of follow-up (online supplemental figure S9), accounting for competing risks from other causes of death (online supplemental table S10), analysing the total case sample, comprising 6334 individuals in the cohort (online supplemental figure S10), using covariates from the interview closest to the accelerometer assessment (online supplemental figure S11) and classifying MVPA based on WHO guidelines (<150 min/week, 150–300 min/week and ≥300 min/week) (online supplemental figure S12) yielded similar results with the main analysis.

Our research aligns with the third American Cancer Society guideline recommendations, suggesting that cancer survivors should engage in regular PA and aim to meet or exceed 150–300 min per week of MVPA when feasible.6 Our study further expands guideline knowledge on the effects of PA on CVD. Given the diversity in cancer diagnoses and treatments, survivors should personalise activity levels to their abilities while avoiding inactivity.

Not applicable.

This study involves human participants and was approved by the North West Multicenter Research Ethics Committee for the UK Biobank (reference number: 11/NW/0382), and all participants provided written informed consent. This study was conducted based on the UK Biobank cohort study under application number 91305. Participants gave informed consent to participate in the study before taking part.