Risk of cancer in regular and low meat-eaters, fish-eaters, and vegetarians: a prospective analysis of UK Biobank participants

Potential participants were first identified for the UK Biobank study using National Health Service (NHS) records, and 9.2 million eligible individuals, aged 40–70 and living within 25 miles of one of the assessment centres in the UK, were invited to participate in the study. Over 500,000 participants (5.5% response rate) consented to participate between 2006 and 2010 [27] and visited one of 22 assessments centres across England, Wales, and Scotland. A full description of the study protocol can be found on the UK Biobank website [27].

The UK Biobank was approved by the NHS North West Multicentre Research Ethics Committee (21/NW/0157). All participants provided informed consent at recruitment, allowing for follow-up using data-linkage to health records. The study was performed in accordance to the Declaration of Helsinki.

Participants were excluded from this analysis if they withdrew consent over the study period (n = 871), had a prevalent cancer diagnosis at recruitment (excluding non-melanoma skin cancer International Statistical Classification of Disease (ICD-10) code: C44; n = 29,504), their genetic sex was different from their reported sex (n = 321), or they did not contribute any follow-up time (n = 2; Additional File 1 Fig. S1). Participants who responded as ‘do not know’ or ‘prefer not to say’ for all dietary questions regarding meat intake were also excluded from the analyses (n = 282). This left a total of 472,337 participants, of whom 217,937 were males and 254,400 were females. For prostate cancer analyses, women were excluded, and for postmenopausal breast cancer analyses, women who were premenopausal at recruitment and did not reach the age of 55 over the follow-up time (n = 16,222), and men, were excluded.

The baseline touchscreen questionnaire also asked participants about sociodemographic, reproductive, and lifestyle factors. In addition, all participants had their blood drawn and anthropometric measurements, including height and weight, taken by a trained professional. Further information on covariate data collection and classification can be found in the Additional File 1 Supplementary Methods.

Non-fasting blood samples were provided by 99.7% of participants at recruitment and were shipped to the central processing laboratory at 4 °C prior to serum preparation, aliquoting, and cryopreservation in the central working archive. Biochemistry markers were measured including insulin-like growth factor-I (IGF-I) and testosterone, as well as sex hormone-binding globulin which we used to calculate an estimate of free testosterone [28]. Further description of the UK Biobank biomarker measurements can be found online [29].

Data on cancer diagnosis were ascertained using a combination of records from the NHS Digital (cancer registry) and Public Health England for participants from England and Wales, NHS Central Register for participants from Scotland [30] as well as the Hospital Episodes Statistics (HES) data for English participants and Scottish Morbidity Records (SMR) for Scottish participants (please see details in the Additional File 1 Supplementary Methods). Using the World Health Organization’s ICD-10 codes, participants were classified as having an event if they had an incident diagnosis of cancer recorded as: all cancer (C00-97 excluding non-melanoma skin cancer: C44), colorectal cancer (C18-C20), breast cancer (C50), or prostate cancer (C61), or if no prior incident diagnosis was reported their primary underlying cause of death was the respective cancer. Participants contributed follow-up time from the date of recruitment until the date of the first cancer registration or cancer first recorded on death certificate, date of death, or last day of follow-up available from HES and SMR data (28 February 2021 for England and Scotland). Cancer registry data were available until 31 July 2019 for England and Wales, and 31 October 2015 for Scotland; after this time, only HES and SMR data were used for the follow-up of participants. For Welsh participants, hospital episode data did not extend past the cancer registry censoring date and therefore were not used. For breast cancer, analyses were restricted to postmenopausal breast cancer and women contributed follow-up time beginning when they turned 55 years of age or their date at recruitment if they were categorised as being postmenopausal from questions asked at baseline (see Additional File 1 Supplementary Methods for further details) [31].

Baseline characteristics of UK Biobank participants were summarised across diet groups for all participants, and separately for men and women.

Cox proportional hazards regressions were used, with age as the underlying time variable, to estimate hazard ratios (HR) and 95% confidence intervals (CI). Minimally adjusted models were stratified by sex (for all cancer and colorectal cancer analyses only) and age at recruitment (< 45, 45–49, 50–54, 55–59, 60–64, ≥ 65 years) and adjusted for region at recruitment (North-West England, North-Eastern England, Yorkshire & the Humber, West Midlands, East Midlands, South-East England, South-West England, London, Wales, and Scotland).

Multivariable-adjusted Cox regression models for all analyses were further adjusted for height (eight sex-specific categories increasing by 5 cm, and unknown/missing (0.51%)), physical activity (low: 0–9.99, medium: 10–49.99, high: ≥ 50 metabolic equivalent of task-hours /week, and unknown/missing (4.04%)), Townsend deprivation index (quintiles from most deprived to least deprived, and unknown/missing (0.13%)), education (completion of national exam at age 16, completion of national exam at age 17–18, college or university degree, or other/unknown/missing (18.7%)), employment status (employed, retired, not in paid employment, or unknown (1.15%)), smoking status (never, former, light smoker: ≤ 15 cigarettes/day, medium smoker: 16–29 cigarettes/day, heavy smoker: ≥ 30 cigarettes/day, or missing/unknown (0.65%)), alcohol consumption (none drinkers, < 1, 1–9.99, 10–19.99, ≥ 20 g/day, or unknown/missing (0.73%)), ethnicity (White, Mixed race or other, Asian or British Asian, and Black or Black British, or missing/unknown (0.56%)), and diabetes status (no, yes, or unknown (0.53%)).

For colorectal cancer and for all cancer sites, multivariable models were further adjusted for female specific covariates: menopausal hormone therapy (MHT) use (no, former, current, or unknown (0.58%)) and menopausal status at recruitment (premenopausal, postmenopausal, or unknown (9.0%)). Moreover, for colorectal cancer, multivariable models were adjusted for non-steroidal anti-inflammatory drug use (NSAID; no reported use, irregular use, regular use of aspirin/ibuprofen). For prostate cancer, models were additionally adjusted for marital status (not living with a partner, living with a partner) [32]. For postmenopausal breast cancer, models were additionally adjusted for MHT use (same as above), age at menarche (≤ 12 years, 13 years old, ≥ 14 years, or unknown (22.5%)), parity and age at first birth (nulliparous, 1–2 children < 25 years old, 3+ children < 25 years old, 1–2 children 25–29.9 years old, 3+ children 25–29.9 years old, 1–2 children 30+ years old, 3+ children 30+ years old, or missing (0.3%)). Further information on covariate classification can be found in Additional File 1 Supplementary Methods. In all models, the proportional hazards assumption was evaluated using Schoenfeld residuals, and no violations were observed.

We considered BMI as a potential confounder as well as a mediator. When BMI was considered as a potential confounder, BMI measured at recruitment was added to multivariable models (multivariable adjusted + BMI; < 20, 20–22.49, 22.5–24.9, 25.0–27.49, 27.5–29.9, 30–32.49, 32.5–34.9, ≥ 35 kg/m², or unknown/missing (0.57%)). Models assessing BMI as a mediator are explained below in the mediation analyses section.

To determine if there was heterogeneity in the associations of diet groups with cancer risk, and to assess the influence of confounder adjustments [33, 34], χ² statistics and p-values for including the diet group in the model were estimated using likelihood ratio tests (LRT) comparing a model without the diet groups variable to the model with the diet groups variable.

No evidence of heterogeneity was observed across BMI subgroups in the associations between diet groups and risk of all cancer, colorectal, postmenopausal breast, and prostate cancer (Additional File 1 Tables S3-S6). For colorectal cancer, there was evidence of heterogeneity by sex (P_het = 0.007), with male low meat-eaters, fish-eaters, and vegetarians having a lower risk of colorectal cancer (0.89, 0.83–0.95; 0.69, 0.47–1.01; 0.57, 0.36–0.91, respectively) in comparison to regular meat-eaters, whereas no significant association was observed across diet groups for females (Additional File 1 Table S4). For smoking status, some evidence of heterogeneity was observed in the association between diet groups and all cancer risk (p_het = 0.056); amongst ever smokers, the low meat-eaters, fish-eaters, and vegetarians had lower risks of all cancer sites than regular meat-eaters (0.97, 0.94–0.99; 0.86, 0.78–0.95; 0.79, 0.70–0.90, respectively), whereas these associations were non-significant for non-smokers (Additional File 1 Table S3). However, when censoring participants who developed lung cancer during follow-up, the test for heterogeneity by smoking status became non-significant between diet groups (p_het = 0.223; Additional File 1 Table S3) although the association with diet group for all cancer sites was still only significant amongst ever smokers.

Associations remained largely the same when analyses were restricted to participants of white European ancestry and when participants with missing data were excluded (Additional File 1 Fig. S2). When the participants who had an event or were censored in the first 2 years of follow-up were excluded, results remained mostly the same except that being a fish-eater was more strongly associated with a lower risk of prostate cancer (HR 0.69, 0.55–0.88) in comparison to regular meat-eaters (Additional File 1 Fig. S3). In analyses, additionally adjusted for intake of fruit and vegetables in the multivariable models, no changes in associations were observed (Additional File 1 Fig. S3). For prostate cancer risk, when PSA testing was added to multivariable models, the associations were not materially changed (Additional File 1 Table S2).

Adjusted and relative means of BMI, IGF-I, and free testosterone across diet groups are shown in Additional File 1 Table S7. Explorations of potential mediators for significant diet-cancer associations are shown in Table 2. When we considered the potential of mediation via BMI in the associations of diet groups and risk of all cancer, this was not found to substantially mediate the observed associations (Table 2). For colorectal cancer risk, BMI was not found to mediate the observed lower risk in low meat-eaters compared with regular meat-eaters (Table 2); hormonal biomarkers were not explored because no differences in concentrations were observed between regular and low meat-eaters (Additional File 1 Table S7). For postmenopausal breast cancer risk, BMI was found to be a potential mediator for the observed difference in the risk between vegetarians and regular meat-eaters, with a decomposed HR^NIE of 0.83 (95% CI: 0.63–1.08) implying that BMI may explain nearly 93% of the lower risk observed in vegetarian women although this was not statistically significant (Table 2). When IGF-I was explored independently and after adjusting for BMI, a HR^NIE of 0.91 (95% CI: 0.73–1.15) was observed. For prostate cancer risk, IGF-I and free testosterone concentrations did not seem to mediate the observed difference in risk between vegetarians and regular meat-eaters, and free testosterone was not found to mediate the difference in risk between fish-eaters and regular meat-eaters (Table 2).

In this study, vegetarians, fish-eaters, and low meat-eaters all had a lower risk of developing all cancer in comparison to regular meat-eaters. It is important to consider that although some cancers may have similar aetiologies, some cancer sites may not be associated with dietary or nutritional factors and that using all cancer incidence as an outcome may crudely capture other lifestyle factors, outside of diet, that may be associated with cancer risk and may confound the associations observed; therefore, these results should be interpreted carefully. In the two largest previous prospective studies following vegetarians, EPIC-Oxford and AHS-2 found that being a vegetarian was associated with a 10% and 8% lower risk of all cancer than being a meat-eater, respectively, after adjusting for lifestyle risk factors and BMI [12, 13]. Fish-eaters in EPIC-Oxford had a lower risk of developing all cancer [12], but no association with risk for all cancer was observed for fish-eaters in comparison to meat-eaters in AHS-2 [13]. In the current analysis, we observed some evidence of heterogeneity by smoking status, and when we removed lung cancer from all cancer cases, significant associations were only observed across diet groups within the ever smoker subgroup. Therefore, the differences observed between diet groups for all cancer outcomes combined may not be due to diet and might be due to residual confounding by differences in other lifestyle factors, such as smoking.

The risk of colorectal cancer was lower in low meat-eaters in comparison to regular meat-eaters whereas there was no significant difference for fish-eaters and vegetarians, potentially due to lack of power as the point estimates suggested lower risks in both these non-meat-eating diet groups. In both EPIC-Oxford and AHS-2, being a fish-eater was associated with a lower risk of colorectal cancer in comparison with meat-eaters, whereas no association was observed for being vegetarian and risk of colorectal cancer compared to regular meat-eaters [12, 16]. We also observed heterogeneity by sex, in that significant inverse associations were observed with risk across diet groups in men, when compared to regular meat-eaters, but not for women. This may in part be due to dietary differences between sexes; however, the number of colorectal cancer cases in some diet groups was too small to draw a clear conclusion. The intake of processed meat has been evaluated by the World Health Organization and World Cancer Research Fund to be a definite cause of colorectal cancer [40] and red meat as a probable cause of colorectal cancer [40, 41]. This is likely to at least in part explain the lower risk of colorectal cancer in low meat-eaters, and mechanisms suggested include chemicals in meat such as nitrosamines [40, 42]. Overweight and obesity also increase the risk for colorectal cancer [43, 44], but in mediation analyses, BMI did not appear to mediate the difference observed between low meat-eaters and regular meat-eaters.

A borderline significantly lower risk for postmenopausal breast cancer was observed for vegetarian women, which appeared to be largely due to their lower BMI as evidenced in mediation analyses and the attenuation of estimates when analyses were adjusted for BMI. We also observed a small potential effect for mediation for lower risk of postmenopausal breast cancer for vegetarians through lower IGF-I concentrations, perhaps influenced by the inclusion of vegans in this group [23]. To date, studies have reported a non-significantly lower risk of breast cancer for women following a vegetarian or pescatarian diet with or without adjustment for BMI [12, 14, 17, 45], which may be due to lack of power to detect modest associations in individual studies. Breast cancer is a heterogeneous disease, with differing risk factors by menopausal status and hormone receptor status [46]. BMI is robustly associated with higher postmenopausal breast cancer risk, probably due to higher circulating oestrogen derived from aromatisation of androgens in the adipose tissue [46]. As such, being vegetarian would be expected to confer a lower risk of postmenopausal breast cancer in comparison to meat-eaters because vegetarians generally have a lower BMI, but whether BMI is a confounder or a mediator for this association is not clear; if vegetarians have a lower BMI because of their diet then BMI may be considered a mediator, but if vegetarians have a lower BMI that is not due to their dietary intake but rather due to other non-dietary lifestyle factors (e.g. physical activity), then BMI would be considered to be a confounder.

Previous research has also suggested that vegetarian women are less likely to use MHT or to attend breast cancer screening [47]. In this analysis, we adjusted for MHT use at baseline, but residual confounding due to differences in use of MHT during follow-up is still possible. Data on breast cancer screening during follow-up were not available in this cohort; therefore, full adjustment for screening attendance was not possible and differences between diet groups in screening may have influenced our findings.

The risk of prostate cancer was lower in men who were vegetarians or fish-eaters in comparison to regular meat-eaters, but no difference in risk was observed for low meat-eaters. Previous analyses in the EPIC-Oxford cohort found a non-significantly lower risk of prostate cancer for British vegetarians and fish-eaters in comparison to meat-eaters [12]. In the AHS-2 study, no difference was found for vegetarians or fish-eaters, whereas being vegan was associated with a 35% lower risk of prostate cancer (based on 1079 cases in the cohort of which only 59 were in vegans) [15]. To date, no established dietary risk factor has been found in relation to prostate cancer risk, although there is some evidence which suggests that higher intake of dairy products, and possibly milk specifically, may increase the risk of prostate cancer [48]. This association has been proposed to be possibly mediated through IGF-I [22, 49], a hormone shown to be positively associated with both milk intake and prostate cancer risk [25, 50]. In this cohort, slightly lower IGF-I concentrations have been observed in vegetarians compared to regular meat-eaters [21], and IGF-I has also been associated with prostate cancer risk [25]; however, the difference in IGF-I concentrations between these diet groups is small and may not confer a substantial difference in prostate cancer risk. As might be expected, in mediation analyses, the estimates were imprecise and there was no evidence that the difference in IGF-I concentrations between diet groups mediates the observed associations with cancer risk.

In this cohort, vegetarian men were less likely than meat-eaters to have had a PSA screening test at recruitment; therefore, vegetarians may have a lower risk of having prostate cancer diagnosed following a PSA test. Similarly, two other cohorts have also reported that vegetarian men were less likely to have had a PSA test [47, 51]. When data at recruitment and available general practice records during follow-up were assessed for PSA testing in the UK Biobank, there was only a small difference with 40% of regular meat-eaters and 37% of vegetarians reporting having had a PSA test (although general practice records were only available for half of the participants) after adjusting for age differences. Adding PSA screening in the multivariable-adjusted model did not attenuate the estimates, suggesting the differences in PSA screening in vegetarians or fish-eaters in comparison to regular meat-eaters does not explain the observed associations, but other differences in attendance for medical examinations could possibly also contribute. Due to unavailable data in UK Biobank, we were also unable to assess associations by tumour subtypes which may be aetiologically different [35]. Considering the substantial difference in risk we observed for vegetarian men, differences in detection and residual confounding, as well as chance, may contribute to this observed difference.

The role of residual and unmeasured confounding must be considered when interpreting the findings from this study. Vegetarians and fish-eaters differ from meat-eaters in many non-dietary lifestyle factors such as lower smoking and alcohol consumption, and higher physical activity [52]. Although relevant potential confounders were added to the multivariable models to adjust for these differences, imperfect measurements and/or changes in these confounders over time may result in incomplete adjustment for these variables. For example, the evidence of heterogeneity by smoking status when looking at all cancer as an outcome suggested that residual confounding by smoking may be present.

Differences in BMI between diet groups have also been suggested to explain the lower cancer incidence observed amongst vegetarians [18]; however, when BMI was considered as a potential confounder and mediator, the difference between BMI by diet groups only slightly attenuated the estimates, with the exception of postmenopausal breast cancer. Whether differences in BMI by diet group is due solely to their diet or other lifestyle factors remains unclear, making it difficult to tease out whether BMI mediates or confounds the associations between diet group and cancer risk.

Strengths of this study include the prospective nature and moderately long follow-up time of participants. Data-linkage to health records was used to determine cancer diagnoses which minimises misclassification and loss to follow-up of participants. The UK Biobank study also gathered data on an array of potential confounders and biochemical biomarkers amongst participants; thus, we were able to adjust the models for potential confounding as well as conduct mediation analysis exploring potential mediators between diet groups and cancer risk. When analyses excluded the first 2 years of follow-up, the results remained largely the same, reducing the chance that these associations are due to reverse causality.

There are some limitations to consider in these analyses. Although there were many cancer cases accrued during the follow-up period, these analyses may still be underpowered to detect moderate associations due to the relatively low numbers of cancer cases amongst vegetarians and fish-eaters in this cohort. We also used hospital admissions data to follow-up participants after 2015 in Scotland and 2019 in England because cancer registry data were not available in UK Biobank after this date, which may result in some missing cancer cases and relatively later dates of diagnosis. We also were unable to adjust for total energy intake as this could not be calculated due to the limited number of dietary questions asked at recruitment. As detailed above, the results may be influenced by unmeasured and residual confounding, as well as chance with numerous comparisons, and causality cannot be confirmed. Misclassification of diet may also be possible, as participants may have underreported their intake or changed their diet over the follow-up period, possibly resulting in attenuation of the risk estimates. Vegetarian diets are characterised by not consuming meat, however, this does not necessarily mean that all vegetarians follow a healthy diet, which may influence their risk of cancer and these results. The mediation analyses only explored three potential mediators, and other possible mediating factors such as other biomarkers relevant for cancer, such as oestradiol, were not available. Moreover, baseline BMI and biomarkers were used to assess mediation, and therefore, these measures may not represent BMI during the follow-up and long-term biomarker concentrations, although correlations with repeat measures of BMI and biomarkers showed high agreement [53]. As well, we used the IORW method to explore mediation with bootstrapping CIs, which may make estimates less statistically efficient in comparison to parametric methods, but the IORW has the advantage that it can be applied in survival analysis and provides estimates of the proportion mediated for the mediators of interest. The UK Biobank has been shown to have a healthier risk profile than the UK population [54] and only included British participants most of whom are of white European ancestry (94%); this may limit generalizability to other populations. However, the risks estimated may still be valid to estimate relative differences for risk-factor disease associations [55].