A prospective study of physical activity and cardiometabolic diseases in association with cancer risk

Abstract

Background

Most physical activity research focuses on general populations, and its cancer-preventive benefits in adults with cardiometabolic diseases remain poorly understood. This study investigated associations between non-occupational physical activity and cancer risk in adults with and without cardiometabolic diseases (cardiovascular disease and/or type 2 diabetes).

Methods

We conducted a meta-analysis of individual participant data of 598,890 men and women, aged 35–70 years at recruitment, across six European countries from the European Prospective Investigation into Cancer and Nutrition (EPIC) and UK Biobank. Participants with cancer at baseline were excluded. Physical activity was assessed at baseline using self-reported validated questionnaires. We used multivariable-adjusted Cox regression to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations between non-occupational physical activity and the risk of physical activity-related cancers (a composite of 15 cancers), with a multiplicative interaction between non-occupational physical activity and time-varying cardiometabolic disease status.

Results

We show that after a median follow-up of 11 years, 37,182 participants developed a first primary physical activity-related cancer (EPIC and UK Biobank combined). In the meta-analysis of both cohorts, a 1 standard deviation increment of non-occupational physical activity is associated with a lower risk of physical activity-related cancer, with HRs of 0.96 (95% CI: 0.93, 0.98) and 0.93 (95% CI: 0.91, 0.96) in adults without and in those with a cardiometabolic disease, respectively (all p-interaction ≥ 0.17).

Conclusions

The findings of this study suggest that higher non-occupational physical activity is equally beneficial for cancer prevention in adults with and without cardiometabolic diseases.

Plain language summary

People with cardiometabolic diseases (CMD), such as cardiovascular disease and type 2 diabetes, are encouraged to be physically active, but most evidence comes from general populations. It is therefore unclear whether physical activity influences cancer risk similarly in people with and without CMD. To address this, we analysed combined data from two European studies to examine how non-occupational physical activity (NOPA) relates to cancers known to be affected by physical activity. We found that higher NOPA was linked to a lower risk of these 15 cancer types combined, and all cancers, regardless of CMD status. These findings show that NOPA provides comparable cancer-preventive benefits for adults with CMD and support promoting regular physical activity for better long-term health.

Gebremariam et al. analyze large European cohorts to assess how non-occupational physical activity relates to cancer risk in adults with and without cardiometabolic diseases. They find that higher activity similarly reduces cancer risk in both groups.

Introduction

Higher levels of non-occupational physical activity (NOPA), which comprises all physical activity outside occupational settings, are inversely associated with the risk of total cancer and several types of cancers in the general population^1,2. The 2020 physical activity (PA) guidelines from the World Health Organization (WHO) include recommendations for adults with chronic conditions such as cancer, hypertension, or type 2 diabetes (T2D)³. A caveat is that these recommendations are largely based on extrapolation of evidence from the general population, with limited direct evidence on the role of PA in cancer prevention for individuals with, for example, T2D. Although the guidelines acknowledge that higher levels of PA are associated with improved outcomes in individuals with coronary heart disease, this condition, or other cardiovascular diseases (CVD), were not included among the chronic conditions subjected to formal evidence review³. The exercise guidelines of the European Society of Cardiology do recommend at least 150 min/week of moderate-intensity exercise to individuals with cardiac disease, but not specifically for cancer prevention⁴. Taken together, PA recommendations for cancer prevention tailored to individuals with CVD and/or T2D, here referred to as cardiometabolic diseases (CMD)⁵, remain limited, hampering personalized prevention.

The co-occurrence of CMD and cancer in individuals is becoming increasingly common⁶, which may be partly due to shared risk factors, such as low levels of PA⁷. Evidence also indicates an association between CMD and cancer risk⁸. T2D is a recognized risk factor for certain types of cancer, such as liver, pancreatic, and endometrial cancer⁹. Three prior prospective cohort studies have examined whether a history of T2D modifies the association between PA and the risk of colon cancer¹⁰, hepatocellular carcinoma¹¹, and endometrial cancer¹². The findings of these studies suggest that higher PA levels may lower the risk of these cancer types, regardless of T2D status. However, previous studies have not investigated the joint associations of PA and T2D with cancer risk. Additionally, the studies relied on self-reported diabetes status^10,12, had small sample sizes¹⁰, and focused on a limited range of cancer types^10–12. Although there are common risk factors shared between CVD and cancer, the extent to which CVD increases cancer risk is not yet well understood¹³. Only one study assessed the association between PA and cancer risk among patients with CVD, but did not compare risk with those without CVD or evaluate interaction effects¹⁴.

We examine associations between NOPA and the risk of a composite outcome of 15 PA-related cancers in adults with and without CMD in the European Prospective Investigation into Cancer and Nutrition (EPIC) and UK Biobank (UKB). We additionally evaluate all cancers combined, breast cancer, and colorectal cancer. In UKB, we assess occupational PA (OPA) and CMD in relation to cancer risk. We show that higher NOPA is associated with a lower cancer risk regardless of CMD status, and that the combination of low NOPA and CMD is associated with increased cancer risk.

Methods

Study settings and participants

The EPIC study has been described previously^15,16. Briefly, the EPIC cohort comprises half a million adults aged mostly 35 to 70 years at recruitment (1992-2000) from 23 research centres across 10 European countries. Participants completed health and lifestyle questionnaires, had anthropometric measurements, and were followed until the last date of centre- and event-specific ascertainment of CVD, T2D, cancer, death, loss to follow-up, or end of the study (December 2008), whichever came first¹⁵. We excluded participants from France, Greece, Norway, and Sweden due to insufficient information regarding their CVD and T2D status or administrative reasons. We further excluded individuals with cancer at baseline and those with missing data for PA. After these and other exclusions, a total of 242,239 EPIC participants remained for analysis as detailed in Supplementary Fig. 1.

UKB is a prospective cohort study of about half a million adults aged 40 to 69 years at recruitment (2006-2010) from 22 centres in England, Scotland, and Wales. Data on socio-demographic characteristics, lifestyle factors, diet, and anthropometric measurements were collected at recruitment¹⁶. Participants were followed until they were diagnosed with cancer, died, lost to follow-up, or last ascertainment of cancer or death (between February 2020 and January 2021, depending on the centre). We excluded participants with prevalent cancer at recruitment, and those with missing data for PA, resulting in a sample size of 356,651 UKB participants (Supplementary Fig. 2).

Ethics approval and consent to participate

All participants from UKB and EPIC have provided written informed consent to participate in the respective cohorts. UKB has ethical approval from the Northwest Multi-Centre Research Ethics Committee. The EPIC study was approved by the Cancer Ethical Review Committee of IARC and by local ethical committees at the participating centres (Supplementary text 1). The current study was approved by the IARC’s Ethics Committee (No. 23–40).

Physical activity assessment

In both cohorts, PA was calculated in metabolic equivalent task hours per week (MET-h/week) based on self-reported PA questionnaires. In EPIC, PA was assessed at recruitment using an interview-based questionnaire¹⁵. In UKB, PA was assessed at recruitment through a self-completed touch-screen questionnaire based on the short International Physical Activity Questionnaire (IPAQ)¹⁶. The validity of the PA assessment tools used in EPIC and UKB studies was evaluated to confirm their ability to rank individuals by PA levels, with a weighted kappa of 0.60 in EPIC¹⁷ and 0.67 in UKB¹⁸.

In both EPIC and UKB, NOPA was defined as the total MET-h/week from recreational activities (at least 10 min in UKB) such as walking, cycling, swimming, or jogging, household and do-it-yourself activities, and stair climbing. In UKB, activities not covered in the IPAQ were derived from additional questions that asked participants’ lifestyle activities of the past four weeks, such as light or heavy do-it-yourself activities. Hours spent in a typical week on each activity were multiplied by their respective MET values as previously recommended^19,20. More details are provided in Supplementary Table 1.

In UKB, we also estimated OPA for participants with paid employment by multiplying the length of a typical working week (in hours) by the relative proportions of time spent doing heavy physical work or walking and standing at work (categorical), as previously done¹⁸: 0% for ‘never/rarely’, 33% for ‘sometimes’, 66% for ‘usually’, and 100% for ‘always’, whereby heavy physical work (minutes per week) was multiplied by 4.5 METs and walking/standing at work (min/week) by 2.25 METs to obtain MET-h/week of OPA¹⁸. In EPIC, OPA was only recorded qualitatively using four categories (sedentary, standing, manual, and heavy manual) without information on proportion of time spent in these categories. OPA in EPIC was therefore only used as a covariate.

Cancer ascertainment

The primary outcome of interest was the incidence of PA-related cancers combined, defined as site-specific cancers for which evidence supports an inverse association with PA^1,21–24: pre- and postmenopausal breast cancer, adenocarcinoma of the oesophagus, endometrial, colon, kidney, gastric cardia, urinary bladder, gallbladder, liver, lung, small intestine, multiple myeloma, myeloid leukemia, rectal, and head and neck cancer. A list of these cancer types, along with corresponding risk estimates or evidence evaluations, is provided in Supplementary Table 2.

Overall cancer, defined as all primary cancers combined, excluding non-melanoma skin cancer, was our secondary outcome. Non-melanoma skin cancer cases were censored at the diagnosis date but not classified as cancer cases, since it has a distinct biological pathway, and is primarily caused by exposure to ultraviolet radiation²⁵. Furthermore, we provided risk estimates for colorectal (C18-C20 and C26.0) and breast cancer (C50), the two most common PA-related cancers. Cancers were coded according to ICD-10 and information on tumour morphology and histology using ICD-O-3 (Supplementary Table 3). Cancer cases were identified through health insurance records, cancer pathology registries, and active follow-up in EPIC, and for UKB, data on cancer diagnosis were obtained from the National Health Service (NHS) Digital and Public Health England for participants from England and Wales. For participants residing in Scotland, it was obtained from the NHS Central Register (NHSCR).

Cardiometabolic disease ascertainment

The ascertainment of CVD and T2D has been described previously⁵. Cardiometabolic disease included CVD and T2D, coded using ICD-10. Cardiovascular disease refers to a composite of ischemic heart diseases (I20-I25), and cerebrovascular disease (I60-I69), while T2D was defined as E11. In EPIC, baseline CVD and T2D were identified primarily through self-reported medical history. Incident CVD was identified and validated through the EPIC-Heart study using questionnaires, medical records, and death certificates. Incident T2D cases were identified and validated in the EPIC-Interact study via self-reports, care register linkages, medication use, and hospital admissions. In UKB, both prevalent and incident CVD and T2D cases were identified using hospital admission records⁵. Prevalent cases were defined as those recorded before or at baseline, and incident cases as those occurring after baseline follow-up started.

Assessment of covariates

Unless otherwise specified, covariates were assessed in both cohorts at baseline using self-reported questionnaires (Supplementary Tables 4). All covariates are reported overall and by sex-specific quartiles of NOPA (MET-h/week) in Supplementary Tables 5 and 6. Due to limited information about occupational status in EPIC, we used OPA as a proxy for the type of occupation (sedentary or standing vs. manual or heavy manual). Diet quality included the modified relative Mediterranean diet score in EPIC categorized as low, medium, and high²⁶, and a healthy diet score in UKB as a scale ranging from 0 (unhealthier) to 6 (healthier)²⁷. Other covariates included height (as a proxy for early life nutrition), and, among women, use of hormones for menopause, and menopausal status. Height and weight were self-reported in the EPIC Oxford centre and measured in all other EPIC centres¹⁵ and in UKB. In UKB, but not in EPIC, data on medication use, average total household income, overall self-rated health, the Townsend deprivation index, and sedentary time (derived by summing daily hours of TV watching, computer use, and driving) were available.

Statistics and reproducibility

After filtering for the exclusion criteria (Supplementary Figs. 1 and 2), the overall missing rate of the covariates was below 4% (Supplementary Table 4). Covariates with missing values were imputed using multiple imputation with chained equations (MICE)²⁸. Each of the covariates was imputed conditionally on the others using appropriate models (e.g., polytomous regression for unordered categorical variables) with 5 imputations and 10 iterations for each using the mice package in R.

To address right-skewness of MET-h/week in EPIC, we capped the outliers at 1.5 times the interquartile range above the third quartile (234 MET-h/week corresponding to the 99th percentile). In UKB, participants with incomplete responses or missing information regarding the number of days or duration were excluded, along with those who reported more than 960 minutes of total PA per day. Additionally, for each of the behaviours, the total time spent walking, total moderate intensity activity, and total vigorous intensity activity were capped at 180 min/day²⁰.

We used Cox proportional hazards regression to estimate cause-specific hazard ratios (HRs) and 95% confidence intervals (CIs) for associations between NOPA and cancer outcomes. Associations were modelled per one standard deviation (SD) increment in NOPA, standardized within each cohort before meta-analysis to account for differences in scale. The SD was 51.4 MET-h/week in EPIC and 34.0 MET-h/week in UKB.

The entry time for cancer follow-up was age at recruitment, and the exit time was age at first primary cancer diagnosis, end of follow-up, loss to follow-up, or death, whichever occurred first. Deaths from any cause were censored. Follow-up for CVD and T2D also started at age of recruitment, and the exit time was the same as for cancer. To address potential detection bias, i.e., the elevated likelihood of cancer diagnosis shortly after CMD diagnosis^29,30, we implemented a 12-month lag period after a CMD diagnosis in all analyses.

Three models were evaluated: Model 1 was adjusted for all covariates as listed in the footnotes of Fig. 1 and Table 1. Estimates for NOPA were adjusted for OPA and vice versa (note: OPA was only modelled in UKB). Model 2 was further adjusted for CVD and T2D status (modelled as binary time-varying variables), and time since diagnosis of CVD and T2D. Model 3, our main model, further included a multiplicative interaction between PA (continuous, time-invariable) and each of the two CMD (time-varying categorical). All models were stratified by age (5-year categories), sex (except for the breast cancer analysis), and recruitment centre, allowing the baseline hazard to vary across these strata and thus removing the violated proportional hazards assumption of these variables.

Fig. 1. Associations between non-occupational physical activity (per 1 SD increment in MET-h/week) and the risk of physical activity-related cancers by cardiometabolic disease status.

One SD of MET-h/week corresponded to approximately 51.4 MET-h/week in EPIC and 34.0 MET-h/week in UKB. The models were stratified by age (5-year categories), sex, and recruitment centre, and adjusted for education, BMI, height, smoking status, alcohol consumption, diet quality, occupational PA/employment, menopausal status (in women), menopausal hormone use (in women), CVD and T2D (modelled as binary time-varying variables) and time since diagnosis of CVD and T2D, and interaction terms between NOPA and each condition (T2D, CVD, CMD) in each model. Hazard ratios for each cohort were estimated using multivariable Cox proportional hazards models. The overall estimates (in bold) were obtained using a random‑effects meta‑analysis. All statistical tests were two‑sided. The measure of centre of the diamonds or the squares represents the point estimates of hazard ratios, and the error bars represent 95% confidence intervals. Abbreviations: CI: confidence interval; CMD: cardiometabolic diseases; CVD: cardiovascular disease; EPIC, European Prospective Investigation into Cancer and Nutrition; HR: hazard ratio; MET-h/week, metabolic equivalent task hours per week; NOPA, non-occupational physical activity; SD, standard deviation; T2D: type 2 diabetes; UKB: UK Biobank.

Table 1.

Associations between non-occupational physical activity (per 1 SD increase in MET-h/week^a) and the risk of physical activity-related cancers by cardiometabolic disease status

Models	EPIC							UKB
	N	Cases	HR	95% CI	p1	p2	p3	N	Cases	HR	95% CI	p1	p2	p3
Overall4	242239	18650	0.962	0.946, 0.979	<0.001			356651	18532	0.940	0.925, 0.955	<0.001
Overall, Adj T2D and CVD5	242239	18650	0.964	0.947, 0.980	<0.001			356651	18532	0.943	0.928, 0.958	<0.001
Without T2D6	224172	17242	0.965	0.948, 0.982	<0.001			335470	17476	0.940	0.925, 0.956	<0.001
With T2D6	18067	1408	0.948	0.898, 1.002	0.0575	0.546	0.545	21181	1056	0.981	0.925, 1.04	0.517	0.167	0.170
Without CVD6	227735	17489	0.966	0.949, 0.983	<0.001			306864	16302	0.946	0.93, 0.962	<0.001
With CVD6	14504	1161	0.932	0.877, 0.992	0.0256	0.274	0.272	49787	2230	0.923	0.887, 0.961	<0.001	0.255	0.253
Without CMD6	212326	16269	0.966	0.948, 0.984	<0.001			294705	15672	0.945	0.929, 0.962	<0.001
With CMD6	29913	2381	0.947	0.907, 0.988	0.0123	0.380	0.384	61946	2860	0.926	0.894, 0.96	<0.001	0.296	0.295

^a One SD = 51.4 MET-h/week and 34.0 MET-h/week in EPIC and in UK Biobank, respectively.

¹ p values for the association (from Wald tests in Cox proportional hazards models).

² p values for local interaction tests (from the full models including interaction terms).

³ p value for the global interaction test (from the likelihood ratio test comparing models with and without interaction).

⁴ Model 1: adjusted for education, BMI, height, smoking status, alcohol consumption, diet quality, occupational PA/employment, menopause status, menopausal hormone use, and stratified by age (5-year categories), sex, and recruitment centre.

⁵ Model 2: further adjusted for T2D and CVD, and the duration of these comorbidities.

⁶ Model 3: further including interaction terms between NOPA and T2D, CVD, and CMD (adjusted multiplicative interaction computed from the full model with interaction). All statistical tests were two-sided. p values smaller than 0.001 are shown as ‘<0.001’ for readability.

BMI body mass index, CI confidence interval, CMD cardiometabolic disease (T2D and/or CVD), CVD cardiovascular disease, EPIC European Prospective Investigation into Cancer and Nutrition, HR hazard ratio, MET-h/week metabolic equivalent task hours per week, NOPA non-occupational physical activity, SD standard deviation, T2D type 2 diabetes, UKB UK Biobank.

Likelihood ratio tests were used to compare models with the interaction term to nested models without interaction. We checked the proportional hazard assumption using Schoenfeld residuals, which was met. The linearity of the association between primary exposure, NOPA, and cancer risk was assessed using natural splines with three knots at the 25^th, 50^th, and 75^th percentiles of NOPA. Departure from linearity was evaluated using likelihood ratio tests comparing models including spline terms with models assuming a linear association (Supplementary Figs. 3, 4, 5, 6).

Next, we assessed separate and joint associations of NOPA and CMD with cancer outcomes, and quantified additive interaction through relative excess risk due to interaction (RERI). NOPA was categorized using a percentile-based cut-point of the MET-h/week distribution: low ( < 20^th percentile), moderate (20^th – 60^th percentile), and high ( ≥ 60^th percentile). We built six mutually exclusive categories: (1) high PA, without CMD (reference); (2) high PA, with CMD; (3) moderate PA, without CMD; (4) moderate NOPA, with CMD; (5) low NOPA, without CMD; and (6) low NOPA, with CMD (joint effect). We calculated additive interaction using the RERI for each CMD and each cancer outcome as RERI_HR  =  HR₁₁ – HR₁₀ – HR₀₁  +  1, with HR₁₁ the risk of being exposed to both factors (e.g., low NOPA and T2D), HR₁₀ exposed to one of the factors (low PA), and HR₀₁ to the other one (e.g., T2D). Moderate NOPA was omitted for RERI estimation. Estimates of 95% CIs were based on the delta method³¹.

Each of the two cohorts’ models was fitted separately, and the results were meta-analysed using a random-effects model. We used I² to describe the percentage of overall variability attributable to between-study heterogeneity.

Subgroup and sensitivity analyses were carried out to assess the robustness of results. First, in both cohorts, we evaluated the associations across population subgroups defined by smoking status (never-smokers versus ever smokers), sex (male versus female), and other subgroups, as shown in the Supplementary File 1. In UKB, we further examined the associations by medication use (individuals on prescribed blood pressure, cholesterol-lowering drugs, or insulin versus those not taking any of these medications), household income, self-rated health status, the Townsend deprivation index, and sedentary time. Lastly, we also conducted a complete case analysis after excluding those with missing data on any of the covariates.

All statistical tests were two-sided, and significance was considered at a p value of less than 0.05. Data were processed and analysed using R version 4.3.1³².

Results

A total of 598,890 participants were included in the analysis, with 242,239 from the EPIC study and 356,651 from UKB. Supplementary Tables 5 and 6 show the participants’ characteristics by levels of NOPA in EPIC and UKB, respectively. In EPIC, higher levels of NOPA were more common among women, never smokers, individuals with higher diet quality, those who were unemployed, and those with lower educational levels (Supplementary Table 5). In UKB, higher levels of NOPA were more common among individuals who were retired and postmenopausal women (Supplementary Table 6).

During a median follow-up of 15.3 years (IQR 12.5–16.7) in EPIC, 30,211 first primary cancers were identified over 3,403,149 person-years, including 18,659 PA-related cancers. In UKB, with a median follow-up of 11.8 years (IQR 11.0–12.4), 35,392 first primary cancers occurred over 4,035,112 person-years, of which 18,532 were PA-related cancers. Age-standardized transition rates from baseline to cancer and from CMD to cancer for both cohorts are shown in Supplementary Figs. 7 and 8. The proportion of PA-related cancers across the two cohorts is shown in Supplementary Table 7.

Non-occupational physical activity (NOPA) and cancer risk by cardiometabolic disease status

Multivariable-adjusted model 1, without adjusting for CMD status, showed an inverse association between higher levels of NOPA and the risk of PA-related cancers in both cohorts. These inverse associations were similar after further adjusting for T2D, CVD, and their durations (model 2) (Table 1). In our main model 3, we estimated associations by CMD status. In the meta-analysis of both cohorts, a 1 SD increment in MET-h/week of NOPA was associated with lower risk of PA-related cancers among those without CMD with a summary HR of 0.96 (95% CI: 0.93, 0.98), and among those with CMD with a summary HR of 0.93 (95% CI: 0.91, 0.96) (Fig. 1). Findings for T2D or CVD separately were similar in magnitude. There was little evidence for a multiplicative interaction between NOPA and CMD status (all p interaction ≥ 0.17) (Table 1). Findings for the risk of all cancers combined were slightly lower in magnitude but otherwise similar (all p interaction ≥ 0.10) (Supplementary Fig. 9 and Supplementary Table 8).

Among participants with T2D, the association of NOPA with colorectal cancer was notably stronger, corresponding to a summary HR of 0.88 (95% CI: 0.81, 0.96) per 1 SD increment in MET-h/week of NOPA, as compared to those with CVD (HR: 0.97; 95% CI: 0.91, 1.05) or no CMD (HR: 0.96; 95% CI: 0.93, 0.99) (Supplementary Table 9 and Supplementary Fig. 10). In contrast, associations with breast cancer were more pronounced among women with CVD, corresponding to a summary HR of 0.91 (95% CI: 0.84, 0.99) per 1 SD increment in MET-h/week of NOPA, as compared to those with T2D (HR: 1.04; 95% CI: 0.95, 1.13) or no CMD (HR: 0.97; 95% CI: 0.95, 0.99) (Supplementary Table 9 and Supplementary Fig. 11). However, these observed associations among adults with/without CMD were not formally different from those among adults without these CMD (all p interaction ≥ 0.089) (Supplementary Table 9).

In sensitivity analyses, in both cohorts, associations between NOPA and PA-related cancers remained consistent among adults with and without CMD across population subgroups (all p interaction ≥ 0.051) (Supplementary Table 10). An exception was a stronger inverse association among adults with CMD, who did not use medication (blood pressure, cholesterol-lowering drugs, or insulin), with a HR of 0.84 (95% CI: 0.76, 0.93) (p interaction = 0.043) (Supplementary Table 10). This pattern was also observed for all cancers combined (Supplementary Table 11). The main results were consistent in complete case analyses (Supplementary Table 12 and Supplementary Fig. 12).

In joint analyses of NOPA and CMD, compared to adults with high levels of NOPA (> 60^th percentile) and without CMD, there was a graded increase in risk of PA-related cancers across NOPA-CMD combinations, with the highest risk among adults with low NOPA ( < 20^th percentile) and CMD (HR = 1.40; 95% CI: 1.11, 1.77) (Table 2). Similar risk patterns were observed for combinations of NOPA and T2D or CVD. However, there was little evidence for additive interactions as quantified by RERI values around the null (Table 2). Corresponding results for all cancers combined were consistent but lower in magnitude (Supplementary Table 13).

Table 2.

Joint associations of non-occupational physical activity (in categories ^a) and cardiometabolic diseases with the risk of physical activity-related cancers

Joint exposures	Study	N	Cases	HR (95% CI)	p value
High NOPA, without T2D	UKB	139854	7155	Ref.
	EPIC	95443	6763	Ref.
High NOPA, with T2D	UKB	7754	413	1.272 (1.037, 1.561	0.021
	EPIC	6588	478	1.122 (0.922, 1.364)	0.250
	Overall (I2 = 0.0%)			1.191 (1.034, 1.372)
Moderate NOPA, without T2D	UKB	142444	6849	1.115 (1.076, 1.155)	<0.001
	EPIC	94495	6988	1.041 (1.004, 1.080)	0.029
	Overall (I2 = 90.0%)			1.078 (1.007, 1.153)
Moderate NOPA, with T2D	UKB	7820	353	1.204 (0.977, 1.483)	0.082
	EPIC	6931	556	1.219 (1.007, 1.475)	0.042
	Overall (I2 = 0.0%)			1.212 (1.053, 1.395)
Low NOPA, without T2D	UKB	69525	3472	1.210 (1.156, 1.266)	<0.001
	EPIC	45915	3491	1.100(1.050, 1.152)	<0.001
	Overall (I2 = 90.0%)			1.154 (1.051, 1.267)
Low NOPA, with T2D	UKB	5607	290	1.459 (1.176, 1.810)	<0.001
	EPIC	4548	374	1.272 (1.041, 1.555)	0.019
	Overall (I2 = 0.0%)			1.356 (1.170, 1.570)
High NOPA, without CVD	UKB	134238	6612	Ref.
	EPIC	96331	6857	Ref.
High NOPA, with CVD	UKB	22346	956	1.160 (1.009, 1.334)	0.037
	EPIC	5088	384	0.978 (0.792, 1.209)	0.840
	Overall (I2 = 40.0%)			1.086 (0.923, 1.278)
Moderate NOPA, without CVD	UKB	138383	6408	1.097 (1.058, 1.138)	<0.001
	EPIC	94661	7074	1.044 (1.007, 1.083)	0.020
	Overall (I2 = 70.0%)			1.070 (1.019, 1.123)
Moderate NOPA, with CVD	UKB	18074	794	1.340 (1.162, 1.546)	<0.001
	EPIC	5884	470	1.025 (0.833, 1.262))	0.813
	Overall (I2 = 80.0%)			1.185 (0.912, 1.539)
Low NOPA, without CVD	UKB	67654	3282	1.189 (1.134, 1.246)	<0.001
	EPIC	46169	3558	1.096 (1.046, 1.148)	<0.001
	Overall (I2 = 80.0%)			1.141 (1.054, 1.236)
Low NOPA, with CVD	UKB	9367	480	1.542 (1.321, 1.801)	<0.001
	EPIC	3532	307	1.159 (0.931, 1.443)	0.186
	Overall (I2 = 80.0%)			1.352 (1.023, 1.786)
High NOPA, without CMD	UKB	133075	6364	Ref.
	EPIC	93821	6424	Ref.
High NOPA, with CMD	UKB	26655	1204	1.184 (1.048, 1.338)	0.007
	EPIC	10861	817	1.08 (0.931, 1.253)	0.310
	Overall (I2 = 0.0%)			1.141 (1.038, 1.254)
Moderate NOPA, without CMD	UKB	137294	6197	1.103 (1.063, 1.145)	<0.001
	EPIC	92608	6603	1.044 (1.005, 1.083)	0.025
	Overall (I2 = 80.0%)			1.073 (1.017, 1.133)
Moderate NOPA, with CMD	UKB	22646	1005	1.326 (1.171, 1.502)	<0.001
	EPIC	11759	941	1.137 (0.982, 1.315)	0.086
	Overall (I2 = 60.0%)			1.234 (1.062, 1.434)
Low NOPA, without CMD	UKB	66840	3111	1.187 (1.132, 1.246)	<0.001
	EPIC	44861	3242	1.095 (1.043, 1.148)	<0.001
	Overall (I2 = 80.0%)			1.140 (1.053, 1.234)
Low NOPA, with CMD	UKB	12645	651	1.572 (1.375, 1.798)	<0.001
	EPIC	7293	623	1.241 (1.064, 1.448)	0.006
	Overall (I2 = 80.0%)			1.401 (1.111, 1.766)

^a Reference category: High NOPA (>=60^th percentile of the MET-h/week) without cardiometabolic disease.

^a Moderate NOPA: 20th–60^th percentile of the MET-h/week; Low NOPA: <20^th percentile of the MET-h/week.

In EPIC, the 20^th and 60^th percentiles were 42 MET-h/week and 94.54 MET-h/week, respectively. In UKB, the 20^th and 60^th percentiles were 4.8 MET-h/week and 22.3 MET-h/week, respectively.

N = Number of participants; Cases = Number of PA-related cancer events. Hazard ratios for each cohort were estimated using multivariable Cox proportional hazards models. The overall estimates were obtained using a random‑effects meta‑analysis. All statistical tests were two-sided. p‑values smaller than 0.001 are shown as ‘<0.001’ for readability.

Relative excess risk due to interactions (RERI) computed for those who had low NOPA with T2D, CVD, and CMD was 0.02 (95% CI: −0.42, 0.46), 0.14 (95% CI: −0.55, 0.84), and 0.14 (95% CI: −0.69, 0.96), respectively.

CI confidence interval, CMD cardiometabolic diseases, CVD cardiovascular disease, EPIC European Prospective Investigation into Cancer and Nutrition, HR hazard ratio, MET-h/week metabolic equivalent task hours per week, NOPA non-occupational physical activity, SD standard deviation, T2D type 2 diabetes, UKB UK Biobank.

For colorectal cancer, compared to adults who had high NOPA without T2D, those with low NOPA and T2D had a summary HR of 1.28 (95% CI: 0.94, 1.75); although the RERI was positive, confidence intervals included the null, providing no clear evidence of additive interaction (RERI = 0.23; 95% CI: −0.08, 0.55) (Supplementary Table 14). For breast cancer, compared to women with high NOPA without CVD, those with low NOPA and CVD had a summary HR of 1.27 (95% CI: 0.94, 1.73), with no clear evidence of additive interaction (RERI = 0.30; 95% CI: −0.36, 0.95) (Supplementary Table 15).

Occupational physical activity (OPA) and cancer risk by cardiometabolic disease status

OPA was not associated with the risk of PA-related cancers, with the notable exception of a positive association among adults with a history of CVD (HR per 1 SD increment in MET-h/week of OPA = 1.12; 95% CI: 1.04, 1.19, p interaction = 0.005) (Supplementary Table 16 and Supplementary Fig. 13). This positive association was slightly more pronounced among never smokers (HR per 1 SD increment in MET-h/week of OPA = 1.14; 95% CI: 1.02, 1.27) (Supplementary Table 17) and for breast cancer (HR per 1 SD increment in MET-h/week of OPA = 1.18; 95% CI: 0.99, 1.41) (Supplementary Table 18).

In the joint analysis of OPA and CMD status in relation to the risk of PA-related cancers, compared with adults who had high OPA (> 60^th percentile) and no CMD, adults with high OPA and CMD had a higher risk of PA-related cancers (HR = 1.59; 95% CI: 1.29, 1.95). This pattern was more pronounced among those with CVD as compared to those with T2D (Supplementary Table 19). Joint analyses for colorectal and breast cancer were mostly inconclusive due to low number of events in some of the exposure categories (Supplementary Table 20 and Supplementary Table 21).

Discussion

In this meta-analysis of two of the largest European prospective cohorts with together over 500,000 adult participants, we found that higher NOPA levels were associated with a lower risk of PA-related cancers and overall cancer among adults with CMD (T2D and/or CVD), with risk estimates comparable to those without CMD. Conversely, low NOPA (below the study-specific 20^th percentile) and a history of CMD were each, and jointly, associated with a higher risk of PA-related cancers and overall cancer compared with adults with high NOPA (above the 60^th percentile) and no CMD. In addition to our main findings, we also found that higher OPA levels were positively associated with the risk of PA-related cancers among adults with CVD, while no association was observed among those without CVD.

Numerous studies have shown that NOPA is consistently associated with a reduced risk of various types of cancers in adults of the general population^1,2,22,33,34. For example, a meta-analysis of nine prospective cohort studies reported that individuals who had NOPA levels of 7.5–15 MET-h/week had a 6% to 27% lower risk of cancers such as breast, kidney, and liver compared to individuals with no NOPA²². A prospective study in UKB reported a 13% (95% CI: 9%, 17%) lower risk of PA-related cancers per 1 SD increment (8.3 milligravity units) of accelerometer-based total PA³⁵. Our findings align with previous evidence and offer further insights into the role of NOPA in cancer prevention among adults with CMD.

Few studies have investigated PA-related cancer risk among adults with CMD^10–12,14, and have not consistently investigated multiplicative and additive interactions between NOPA and CMD status. In UKB, a similarly lower risk for PA-related cancers was observed among adults with and without diabetes in relation to higher total PA³⁵. A meta-analysis of both retrospective and prospective studies found an inverse association between higher PA and renal cancer, with little evidence for effect modification by T2D status³⁶. In our study, higher NOPA was associated with a lower risk of colorectal cancer and breast cancer, regardless of CMD status. The association with colorectal cancer was suggestively stronger among adults with T2D, whereas the association with breast cancer was slightly more pronounced among women with CVD, albeit with little evidence for being different from adults without these CMD. The joint analyses were suggestive of an additive interaction between low NOPA and CMD, in particular for the combination of low NOPA and T2D in relation to colorectal cancer risk. However, the confidence interval was wide due to a relatively low number of colorectal cancer events in the group exposed to both low NOPA and T2D (n = 170).

To our knowledge, only one previous study has stratified PA-associated cancer risk by CVD status³⁵. In that study, higher levels of total PA were similarly associated with a lower risk of PA-related cancers among adults with and without a history of CVD³⁵. However, separate and joint associations were not tested, and only baseline CVD status without accounting for CVD duration was investigated³⁵. Our study, therefore, adds evidence by suggesting that higher NOPA levels are associated with similar cancer risk reductions among adults with and without CVD. This is clinically relevant, as people living with CVD may be reluctant to engage in regular PA because of uncertain health implications^37,38.

Evidence on OPA and cancer risk is limited, and to the best of our knowledge, interaction by CMD status has not been investigated. In a Norwegian cohort of healthy adults, higher OPA, compared to sedentary occupations, was associated with a lower incidence of all cancers combined and of cancers of the colon, rectum, prostate, endometrium, and breast³⁹. Similarly, the EPIC Italy study⁴⁰ and a prospective U.S. cohort study⁴¹ reported an inverse association between OPA and risk of breast cancer, a pattern that we also observed in our analysis among those without CMD for OPA levels up to ~75 MET-h/week (Supplementary Fig. 13). In contrast, among adults with CMD, OPA levels of ~50 MET-h/week or higher were positively associated with the risk of PA-related cancers. These contrasting patterns align with the ‘physical activity paradox’⁴², whereby OPA does not confer the same health benefits as NOPA. OPA often involves prolonged, repetitive, or static workloads, which due to incomplete recovery, could lead to elevated systemic inflammation and eventually could contribute to cancer promotion⁴³. However, an alternative explanation for the positive association between OPA and cancer risk among adults with CMD is unobserved confounding and/or selection bias, whereby an unobserved confounder could increase the risk of both CMD and cancer independent of OPA. Given the scarcity of OPA research stratified by underlying chronic disease status, more targeted studies are needed to disentangle how different dimensions of OPA (intensity, duration, workload, and psychological stress) interact with cardiometabolic vulnerability to influence cancer risk.

Our study has notable strengths. The analysis is based on multinational data across six European countries from two large prospective cohort studies, which allowed us to model the temporal relationship between PA, objective T2D, and CVD status, and cancer risk. We also assessed multiplicative interaction and joint associations of PA and T2D/CVD on cancer risk. Our findings, however, should be interpreted considering the following limitations. First, PA exposure was assessed only at baseline, and potential changes over time were not evaluated. Also, as it relied on self-reported questionnaires, reporting bias may be inevitable. However, this bias is likely minimal, as we used standardized rankings instead of the actual responses, and previous research supports the questionnaire’s validity for ranking PA levels^17,18. EPIC assessed domain-specific activities (e.g., walking, gardening, housework) over the past year using task-specific MET values, while UKB captured walking, moderate, and vigorous PA over the past week using intensity-based METs and a 10-min bout criterion. EPIC’s approach is more detailed but may be more prone to recall bias and tends to smooth out seasonal and short-term variation. Notably, the median MET-h/week was higher in EPIC than in UKB, reflecting differences in assessment scope and time frame. Despite these methodological differences, hazard ratios were similar across cohorts, suggesting that both instruments effectively rank individuals by physical activity level. Additionally, we were unable to assess the effects of specific types of physical activity beyond aerobic exercise because such information was not collected consistently across the two cohorts. Second, the low participation rate of only 5% in the UKB cohort^44,45 suggests that the study participants may not accurately represent the overall UK population. Third, despite extensive adjustment for covariates, residual confounding remains possible. The lack of data on medication use, sedentary behaviour, and employment information in the EPIC cohort could have impacted our estimates. However, accounting for these covariates in UKB did not affect estimates. Nevertheless, other unmeasured factors may also contribute to residual confounding. Finally, the methods used to handle PA data may have influenced our estimates. Exclusion of extreme PA values could introduce bias if such values reflect true behaviour. For practical reasons in joint analyses, we categorized PA into three groups, which may oversimplify the exposure-outcome relationship.

Conclusions

In this study among European adults, higher levels of NOPA were associated with lower cancer risk, which was similar among adults with and without a history of T2D and/or CVD. Low NOPA and a history of CMD were each separately and jointly associated with a higher risk of cancer, compared with higher NOPA and no history of CMD. Taken together, our findings add to the growing body of evidence supporting NOPA as a key cancer prevention strategy encompassing population subgroups with cardiometabolic diseases.

Author contribution

H.F. conceived and designed the work. A.G., H.F., E.F., and P.F. were responsible for data analysis and interpretation, wrote the original draft, and carried out subsequent editing. A.L.R.H., J.L.M.A., K.S., R.F., M.B.S., C.S., S.P., M.Fra (Matteo Franco), N.C.O.M., E.M., J.B., M.Fer (Marta Farràs), M.J.Sa (Maria-José Sánchez), M.G., O.M.C., S.T.T., A.K.H., D.A., E.R., Q.G., L.P.N., R.C.R., H.N., P.B., H.B., M.F.L. (Michael F. Leitzmann), M.J.S. (Michael J. Stein), C.M.F., B.F., and O.P. contributed to data acquisition or interpretation of results. All authors contributed to writing, editing, and final approval of the manuscript. HF is accountable for all aspects of the work.

Peer review

Peer review information

Communications Medicine thanks Tongyu Ma, Yuichiro Nishida and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

Funding for IIG_FULL_2021_027 was obtained from World Cancer Research Fund (WCRF UK), as part of the World Cancer Research Fund International grant programme, and the French National Cancer Institute (l’Institut National du Cancer, INCA_16824). The funders had no role in study design, data collection and analysis, decision to publish, or manuscript preparation.

Data availability

The data for this study are available from UK Biobank, which is an open-access resource. However, UK Biobank has an ethical responsibility to ensure that participants’ data are accessed only by eligible scientific researchers conducting health-related research in the public interest. In this sense, bona fide researchers can apply to use the UK Biobank dataset by registering and applying at https://www.ukbiobank.ac.uk/use-our-data/apply-for-access/. Registrations are reviewed within 10 working days of submission. The EPIC data for this study are available for investigators who seek to answer important questions on health and disease in the context of research projects that are consistent with the legal and ethical standard practices of IARC/WHO and the EPIC Centres. For information on how to apply to gain access to EPIC data and/or biospecimens, please follow the instructions http://epic.iarc.fr/access/. Data access proposals are usually reviewed monthly, and after signing a data use agreement, data can be accessed via a central analysis platform. Source data for Fig. 1 can be accessed from ‘Supplementary Data 1’ of this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s43856-026-01665-9.