This study utilized data from the UK Biobank, a prospective cohort study comprising approximately 500,000 participants aged 40–69 years recruited between 2006 and 2010, from across the United Kingdom. Details of this study have been described elsewhere [16]. Extensive information on socio-demographic, lifestyle, and health-related factors was collected through a self-completed touch-screen questionnaire and computer-assisted interviews. In addition, physical and functional measurements were conducted, and data on cancer, death, and primary care were obtained through linkage to national cancer and death registries and electronic health records. Ethical approval for the UK Biobank was obtained from the North West Multi-centre Research Ethics Committee (MREC) as a Research Tissue Bank (RTB), the approval was renewed in 2021 (21/NW/0157), and all participants provided electronic signed informed consent. For this analysis, only participants with no previous cancer diagnosis (except non-melanoma skin cancer) and without missing body mass index (BMI) values at recruitment were included.

Weight measurements were taken using the Tanita BC-418 MA body composition analyzer, and standing height was measured using a Seca 202 height measure during the initial assessment visit [17]. BMI was determined by dividing individuals’ weight in kilograms by the square of their height in meters. The World Health Organization (WHO) categories were used to classify BMI: <18.5 kg/m2 (underweight), ≥18.5-<25 kg/m2 (normal weight), ≥25-<30 kg/m2 (overweight), ≥30-<35 kg/m2 (obesity class I), ≥35-<40 kg/m2 (obesity class II), and ≥40 kg/m2 (obesity class III) [18].

Information on cancer incidence was obtained from national cancer registries through linkage with the UK Biobank data. The International Statistical Classification of Diseases (ICD-10) was used to determine incident cancer cases. Thirteen cancer types with established causal association with excess weight according to IARC [1], which are listed in Supplemental Table 1, were included in the analyses. In this manuscript, these cancers are referred to as obesity-related cancers. Of these, GI cancers comprised cancers of the esophagus, stomach (cardia), colorectum, liver, gall bladder, and pancreas, while all other obesity-related cancers were categorized as non-GI cancers. The number of incident cancer cases by cancer type is also shown in Supplemental Table 1. This study includes complete cancer follow-up data until 29th of February 2020 for England and Wales and 31st of January 2021 for Scotland.

All statistical analyses were conducted using SAS software version 9.4. Baseline characteristics of the cohort were summarized using descriptive statistics. The association between BMI and cancer risk was evaluated using multivariable Cox proportional hazards models. Follow-up time was defined as the time from initial assessment visit to the first cancer diagnosis, date of death, date of loss to follow-up, or the end of the follow-up period, whichever came first. Two models were fitted: the first model was adjusted for age at baseline (years, continuous) and sex (male, female), and the second (fully adjusted model) was adjusted for additional covariates including height (cm, continuous), self-reported ethnic background (classified as white, or other), socioeconomic status (Townsend deprivation index, continuous), educational qualifications (higher academic/professional, lower academic/vocational, or none), smoking status (never, former, current), pack-years of smoking (years, continuous), alcohol consumption (never, special occasions only, 1–3 times a month, once or twice a week, 3–4 times a week, daily or almost daily), level of physical activity determined by the International Physical Activity Questionnaire (IPAQ) [19] (low, moderate, high), fruit intake (pieces/day, continuous), vegetable intake (tablespoons/day, continuous), and red and processed meat intake (never, less than once a week, once a week, ≥ 2 a week), hormone replacement therapy (no, yes/women only), menopausal status (pre-menopausal, post-menopausal/women only), history of bowel cancer screening (no, yes), history of mammography (no, yes/women only), and family history of breast and colorectal cancer (no, yes). Schoenfeld residuals plots were examined to assess deviations from the proportionality assumption and none was detected. To address missing covariate values, the SAS multiple imputation procedure PROC MI was employed. The analysis involved combining five imputed datasets using PROC MIANALYZE. Specifically, physical activity data had a 20% rate of missing values, while all other covariates had less than 2% missing values and age and sex had no missing values.

Hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) were calculated to quantify the cancer risk for each BMI category compared to the reference group of participants with normal BMI (18.5 kg/m2≤BMI<25 kg/m2). HRs (95% CIs) were computed for all obesity-related cancers, obesity-related GI cancers, and obesity-related non-GI cancers. Due to small numbers in the underweight category and obesity sub-categories, further analyses were conducted excluding the underweight participants (BMI<18.5 kg/m2) and using combined obesity sub-categories (≥30 kg/m2). PAFs (95% CIs) of cancer cases associated with excess weight (BMI≥25 kg/m2) for cancer incidence were then estimated based on the HRs calculated for the association between overweight and obesity with the various cancers and the prevalences of overweight and obesity in the population, using Mietinnen’s formula [20] modified for risk factors with multiple categories of exposure (Supplemental Text). Population prevalence of different BMI categories by sex- and 5-year age groups for years 2006–2010 (years of the UK Biobank recruitment) was extracted from a nationally representative survey, the “Health Survey for England – 2010” [21]. The data source is summarized in Supplemental Table 2.

PAFs were calculated as age- and sex-weighted averages accounting for the substantial variability in cancer incidence observed across different age and sex groups. Initially, age- (at baseline; 5-year increments) and sex-specific PAFs were calculated. The overall PAF was subsequently determined as a weighted mean of these specific PAFs, employing weights corresponding to the age- and sex-specific number of cancer cases. All PAFs were calculated as proportion of obesity-related cancer cases attributable to excess weight. Additionally, we calculated PAFs as proportion of total cancer cases (ICD-10 C00-C97 excluding C44) attributable to excess weight in our dataset (number of total cancer cases not shown).

To address the potential underestimation of calculated HRs and PAFs, the following analyses were performed (this approach has been detailed previously [15]): firstly, a standard cohort analysis was performed using the complete available follow-up period at the time of analysis (14 years). Subsequently, we performed separate analyses by including either only the initial four years of follow-up (0-4 years), or only the later years of follow-up (>4–14 years). The rationale for this analysis is as follows: During the early years of follow-up, it is likely that a significant proportion of newly diagnosed cancer cases originated from participants who already had preclinical cancer at the time of their recruitment, which may have led to some weight loss. For cancers diagnosed more than four years after diagnosis, relevant weight loss before recruitment would appear unlikely. We additionally did a sensitivity analysis by including 0–3/>3–14 and 0–5/>5–14 years of follow-up to assess the robustness of the findings from the main analysis.

All analyses were conducted for all obesity-related cancers combined, as well as separately for obesity-related GI and non-GI cancers.

Subgroup analyses were performed based on age group, sex, smoking status, and diabetes considering different follow-up time windows (0–4, >4–14, 0–14). All p-values reported in this study are two-sided, and statistical significance was defined as p-values less than 0.05.