Cancer is a major and growing public health challenge. Lifestyle exposures including smoking, physical activity, diet, alcohol consumption and body mass index (BMI) are established modifiable risk factors for many cancers. Because etiologic pathways differ among cancer sites, determining which cancers are most strongly associated with a composite healthy lifestyle and with its individual components can inform targeted prevention strategies. The present study reports a pan-cancer analysis in UK Biobank examining associations of an established healthy lifestyle score (HLS) with overall and site-specific cancer incidence and estimating population attributable fractions (PAFs) for incomplete adherence to healthy behaviours.
The analysis used the UK Biobank prospective cohort. From the baseline recruitment pool the authors excluded participants who withdrew, were pregnant at baseline, had prevalent cancer (other than non-melanoma skin cancer), lacked necessary lifestyle data, or had follow-up under 2 years. The final analytic sample comprised 287,745 participants aged 37–73 years at recruitment. Follow-up for cancer outcomes was based on linkage to cancer and death registries, with end-of-follow-up dates varying by UK nation (England to 31 December 2020, Scotland to 30 November 2021, Wales to 31 December 2016).
The study used a previously benchmarked HLS that assigns one point for each of five healthy behaviours: non-smoking (never or former smoker with <30 pack-years), moderate alcohol intake (≤12 g/day for women; ≤24 g/day for men), a healthy diet (participants in the highest 40% of a diet-score distribution), adequate physical activity (≥500 MET-hours/week combining walking and moderate/vigorous activity), and a healthy BMI (18.5–<25 kg/m2). Points were summed and HLS categorized as low (0–1), medium (2–3) and high (4–5).
The primary outcome was first incident cancer diagnosis identified through registry linkage. Analyses included cancer sites with at least 100 cases in the study population; breast cancer was analyzed separately for premenopausal and postmenopausal cases. Follow-up time was calculated from recruitment to cancer diagnosis, loss to follow-up, death, or the relevant regional censoring date.
Missing data were handled by multiple imputation (Markov-Chain Monte Carlo, five imputed datasets). Cox proportional hazards models estimated hazard ratios (HRs) and 95% confidence intervals (CIs) comparing HLS categories, with the low-HLS group as reference. Model 1 adjusted for age and sex; Model 2 further adjusted for ethnicity, region, Townsend Deprivation Index, household income, education, cancer screening history, family history of specific cancers, hypertension, diabetes and height. Additional cancer-specific covariates were included where appropriate. Individual lifestyle components were also analyzed with mutual adjustment for other lifestyle factors. The Benjamini–Hochberg procedure corrected for multiple comparisons. Variance inflation factors were assessed and reported below thresholds suggesting problematic multicollinearity.
PAFs and 95% CIs were estimated for incomplete adherence (0–3 vs 4–5 HLS points) and for individual unhealthy behaviours using the graphPAF R package with 1,000 bootstrap replications; cancers were ranked by estimated PAFs and the number of cancer sites attributable to each unhealthy behaviour was quantified when PAF CIs excluded zero.
Mean age in the analytic cohort was 56.0 (SD 8.1) years. Participants with higher HLS were more frequently women, had higher socioeconomic status and education, were taller, had lower prevalence of hypertension and diabetes, reported less family history of cancer, and were more likely to have undergone cancer screening.
Over a median 11.5 years of follow-up, 12,947 participants developed cancer. Prostate cancer (men) had the highest incidence in this cohort, followed by breast cancer (women); colorectal cancer was the most frequent among cancers affecting both sexes. Comparing high versus low HLS, there was a decreased risk of overall cancer after adjustment (Model 2 HR approximately 0.71, 95% CI 0.68–0.74 as reported). In site-specific analyses, adherence to 4–5 versus 0–1 healthy behaviours was associated with decreased risks for 13 of 27 cancer types examined, including cancers of the oral cavity, pharynx, larynx, oesophagus, stomach, colorectum, liver, pancreas, lung, soft tissue sarcoma, kidney, bladder and postmenopausal breast cancer. Reported site-specific HRs in the abstract ranged from 0.12 to 0.79 for these inverse associations.
Significant PAFs for incomplete adherence to the HLS were observed for 14 cancer types. Among the individual components, smoking and overweight (i.e., not meeting the healthy BMI criterion) were associated with the largest number of cancer types for which PAFs were statistically significant. The authors therefore highlighted smoking cessation and weight control as key priorities for cancer prevention given their broader attributable impact across cancer sites.
Subgroup analyses stratified by age (<60 vs ≥60 years) and by sex (men vs women) were conducted for cancer sites with significant inverse associations in the primary analyses; interaction tests were performed by adding cross-product terms to the models. The manuscript reports that these subgroup analyses were part of the analytic plan, with further details available in supplementary materials.
The study extends evidence that a composite healthy lifestyle is linked to lower risks for both common and less-common cancers in this large prospective cohort. Because smoking and overweight were associated with more cancer types than other lifestyle components, the authors suggest prioritizing smoking cessation and weight-control interventions within cancer prevention strategies.
Key strengths include the large sample size, prospective design, use of an established HLS, comprehensive covariate adjustment and estimation of PAFs with bootstrap confidence intervals. The study followed STROBE reporting guidance and used multiple imputation for missing data and correction for multiple testing. Full numeric details for site-specific HRs, PAF estimates, supplementary tables, figures and additional methods are provided in the original article and supplementary materials referenced by the authors.