Diet and colorectal cancer in UK Biobank: a prospective study

Background

In October 2015, the International Agency for Research on Cancer (IARC), part of the World Health Organization (WHO), concluded that there was sufficient evidence to classify processed meat as carcinogenic to humans (Group 1) and red meat as probably carcinogenic to humans (Group 2A).¹ A recent systematic review of the published evidence on foods and beverages and colorectal cancer by the World Cancer Research Fund–American Institute for Cancer Research (WCRF-AICR) Continuous Update Project concluded there was convincing evidence that processed meat and alcoholic drinks increase the risk of colorectal cancer. The evidence that red meat increases the risk and that dairy products, wholegrains and foods containing dietary fibre reduce risk was judged as probable. The evidence for other foods and beverages was weak.² Few studies included in the review corrected the results for measurement error and error in dietary assessment tends to attenuate the associations between diet and disease.^3,4 Some studies included in the review did not publish results on all food types, which may indicate the presence of publication bias. In addition, many of the large previous studies were based on dietary intakes in the 1990s^5–7 and food-supply data indicate that meat consumption has changed from the 1990s, with a much higher proportion of total meat supply in Europe and the USA now coming from poultry as opposed to beef and pork.⁸ Thus, it is not certain whether previous risk estimates are relevant to current eating patterns.

In addition, tumour characteristics⁹ as well as the underlying aetiology of bowel cancer¹⁰ may differ by anatomical sub-site, but there are few data on the associations between dietary factors and risk of sub-sites of colorectal cancer.

UK Biobank is a prospective cohort study of half a million men and women recruited between 2006 and 2010, and provides the opportunity to investigate prominent hypotheses related to diet and colorectal cancer in a contemporary population-based cohort in the UK. We have previously shown that the dietary data collected from the UK Biobank short food-frequency touchscreen questionnaire, which generally asks about frequency of consumption of main foods and food groups, is highly reproducible.¹¹ These data are available on all UK Biobank participants. In addition, dietary intakes were re-measured in a large sub-sample of participants (n = 175 402) who completed at least one online 24-hour dietary assessment and these data can be used to correct for regression dilution and other forms of measurement error.¹² The objective of the current study is to systematically examine the associations of colorectal-cancer risk with the intake of foods and food groups included in the short food-frequency touchscreen questionnaire: meat, fish, fruit, vegetables, milk, cheese, alcohol, tea and coffee, as well as fibre intake, and to use re-measured dietary intakes to quantify the risk at actual levels of intake in UK Biobank. We also examined the associations between intakes of food and food groups with anatomical sub-sites of colorectal cancer.

Methods

Results

Among 475 581 participants without prevalent cancer (excluding non-melanoma skin cancer) at recruitment, followed for an average of 5.7 years (maximum 8.6 years), 2609 cases of colorectal cancer occurred (1504 in men and 1105 in women). Table 1 (men) and Table 2 (women) compare participant characteristics between those in the lowest and highest categories of reported consumption of red meat, processed meat, poultry, fish, dairy milk and cheese. Compared with those in the lowest category, participants in the highest category of reported total red-meat intake were slightly older, more likely to be smokers, had a higher BMI and body-fat percentage, had a higher alcohol intake and had lower intakes of fruit, vegetables and fibre. The same was true for processed-meat intake, with the exception of age, which was similar between the two categories. Participants in the highest category of processed-meat intake (men and women combined) reported higher intakes of bacon, ham and sausages in the online 24-hour dietary assessments (mean daily intakes: 11.3 g of bacon, 7.5 g of ham and 6.2 g of sausages, total = 25.0 g per day) than those who reported never consuming processed meat at recruitment (mean daily intakes: 1.4 g of bacon, 1.2 g of ham and 0.8 g of sausages, total = 3.4 g per day) (data not shown). Supplementary Tables 1 (men) and 2 (women), available as Supplementary data at IJE online, compare participant characteristics between those in the lowest and highest categories of reported consumption of fruit, vegetables, total fibre, alcohol, tea and coffee.

The multivariable associations between foods, food groups and fibre and colorectal-cancer risk are shown in Figure 1 (red and processed meat, red meat, processed meat, poultry, fish, dairy milk and cheese) and Figure 2 (fruit, vegetables, fibre, alcohol, tea and coffee). The values from the multivariable model, as well as the results from the minimally adjusted model and the multivariable model in those that did not report a major change in their diet in the past 5 years because of illness are shown in Supplementary Tables 3 and 4, available as Supplementary data at IJE online. In the multivariable adjusted model, the HR for incident colorectal cancer was 1.20 [95% confidence interval (CI): 1.04–1.37] for those who reported consuming an average of 76 g/day of red and processed meat compared with those who reported consuming an average of 21 g/day [and per 50-g/day increment was 1.17 (95% CI: 1.04–1.32)]. For red meat, the HR was 1.15 (0.98–1.36) for those who reported consuming an average of 54 g/day compared with those who reported consuming an average of 8 g/day [and per 50-g/day increment was 1.18 (1.00–1.39)] and, for processed meat, the HR was 1.19 (1.01–1.41) for those who reported consuming an average of 29 g/day compared with those who reported consuming an average of 5 g/day [and per 25-g/day increment was 1.19 (1.03–1.38)]. We further adjusted the red and processed-meat analyses for milk, cheese and fibre from bread and breakfast cereals and the results were unchanged [the HR for each 50-g/day increment in red and processed meat was 1.18 (1.05–1.33)] for each 50-g/day increment in red meat was 1.19 (1.00–1.41) and for each 25-g/day increment in processed meat was 1.20 (1.03–1.40).

The HR for each 25-g/day increment in fish was 0.96 (0.86–1.07). The HR for each 100-mL/day increment in milk intake was 0.93 (0.87–1.01); we further adjusted the milk analyses for red meat, processed meat and fibre from bread and breakfast cereals and the HR for each 100-mL/day increment was 0.94 (0.87–1.02). The HR for each 5-g/day increment in total fibre was 0.94 (0.86–1.03) and for each 10-g/day increment in alcohol was 1.08 (1.04–1.12). We further adjusted the alcohol analyses for red meat, processed meat, milk, cheese and fibre from bread and breakfast cereals and the HR for each 10-g/day increment was 1.06 (1.03–1.10). The intakes of poultry, cheese, fruit, vegetables, tea and coffee were not associated with colorectal cancer. When we excluded participants who reported a major change in their diet in the past 5 years due to illness, the HRs were of similar magnitude.

When we adjusted for BMI instead of waist circumference, the results were very similar (data not shown). When we examined the effect of adjusting for each covariate sequentially (Table 3), using the total of red and processed meat as the main exposure, waist circumference attenuated the HR, as did alcohol and, to a lesser extent, smoking. Overall, the HR was attenuated by 7% after adjustment for 14 covariates. The HR (95% CI) for each 50-g/day increment in red-meat intake was 1.18 (1.00–1.39) and after adjustment for processed meat was 1.12 (0.94–1.34). The HR for each 50-g/day increment in processed-meat intake was 1.40 (1.04–1.88) and after adjustment for red meat was 1.33 (0.98–1.83).

Supplementary Table 4, available as Supplementary data at IJE online, shows the main results in men and women separately. There was heterogeneity by sex for the associations between colorectal cancer and red and processed meat (P_{heterogeneity} = 0.008), with a positive association seen in men [HR for each 50-g/day increment in red-meat intake = 1.39 (1.17–1.64)] and no association was seen in women [HR for each 50-g/day increment in red meat = 0.99 (0.83–1.19)]. There was also heterogeneity by sex for red meat (P_{heterogeneity} = 0.008) and for processed meat (P_{heterogeneity} = 0.022). There was also heterogeneity by sex for alcohol, with a positive association seen in men [HR for each 10-g/day increment in alcohol intake = 1.12 (1.08–1.17)] and no association seen in women [HR for each 10-g/day increment = 0.99 (0.93–1.06), P_{heterogeneity} = 0.002]. There was no heterogeneity by sex for the associations between fish, dairy milk, cheese, fruit, vegetables, fibre, tea or coffee and colorectal cancer.

When we examined oily fish and non-oily fish separately, there were no associations between oily fish [HR for each 25-g/day increment = 0.87 (0.74–1.02)] or non-oily fish [HR for each 25-g/day increment = 0.94 (0.76–1.17)] and colorectal cancer (Supplementary Table 5, available as Supplementary data at IJE online).

When we examined the source of fibre, intake of fibre from bread and breakfast cereals was associated with a reduced risk of colorectal cancer [HR for the highest vs lowest fifth of intake = 0.86 (0.76–0.98), P_trend = 0.005]. We further adjusted the fibre from bread and breakfast cereals for red meat, processed meat, milk and cheese and the results were unchanged. Intakes of fruit and vegetable fibre were not associated with the risk of colorectal cancer (P_trend = 0.728 and P_trend = 0.633, respectively) (Supplementary Table 6, available as Supplementary data at IJE online).

When we examined the type of alcoholic drink, beer was associated with an increased risk of colorectal cancer [HR for each 10-g/day increment of alcohol from beer = 1.11 (1.06–1.18)]. The HR for each 10-g/day increment of alcohol from wine was 1.05 (1.00–1.10) and for each 10-g/day increment of alcohol from spirits was 1.08 (0.90–1.31) (Supplementary Table 7, available as Supplementary data at IJE online).

Supplementary Table 8, available as Supplementary data at IJE online, shows the results for colon and rectal cancer separately and Supplementary Table 9, available as Supplementary data at IJE online, shows the results for proximal and distal colon cancer separately. There was no heterogeneity between colon and rectal cancer for any of the foods or nutrients examined. There was heterogeneity between proximal and distal colon for red and processed meat, processed meat and alcohol; in all cases, a positive association was observed for distal colon cancer and no association for proximal colon cancer. When we examined the associations between type of alcoholic drink and colon and rectal cancer separately, the HR for each 10-g/day increment of alcohol from beer was 1.09 (1.02–1.16) for colon cancer and 1.16 (1.07–1.27) for rectal cancer, but there was no heterogeneity (P_{heterogeneity} = 0.223) (Supplementary Table 10, available as Supplementary data at IJE online).

Discussion

In this large contemporary prospective study of half a million men and women from the UK that included 2609 cases of incident colorectal cancer, we found that a 25-g/day increment in processed-meat intake (equivalent to about one rasher of bacon or one slice of ham²¹) was associated with an 19% (95% CI: 3–38, P_trend = 0.020) greater risk of incident colorectal cancer. We also found that each 50-g/day increment in red-meat intake (equivalent to about one thick slice of roast beef or the edible portion of one lamb cutlet²¹) was associated with a 18% (95% CI: 0–39, P_trend = 0.049) greater risk of incident colorectal cancer. Each 10-g/day increment in alcohol intake [equivalent to about half a pint of 4.5% alcohol (by volume) beer] was associated with an 8% (95% CI: 4–12, P_trend < 0.001) greater risk of incident colorectal cancer. Intake of fibre from bread and breakfast cereals (but not fibre from vegetables or fruit) was also associated with a reduced risk.

The directions of our results for red and processed meat are in agreement with the latest systematic review from the WCRF Continuous Update Project, which summarized the evidence from available prospective studies for red and processed meat.²² However, the magnitudes of our estimates are nearly twice as large. This may be in part because we used re-measured intakes (using the data from the large sub-sample, n = 175 402) to fit the trends in risk per increment in mean intake. This approach reduces the impact of measurement error in the assessment of food intakes and measurement error tends to bias associations towards the null.¹² Adjusting for waist circumference, alcohol and, to a lesser extent, smoking attenuated the results for red and processed meat. However, the fully adjusted models were adjusted for 18 variables that are known or suspected confounders and, for processed meat in particular, the fully adjusted results were still strong. We further adjusted red and processed meat for milk, cheese and fibre from bread and breakfast cereals and the results were unchanged. When we excluded participants who reported having made major changes to their diet in the past 5 years due to illness, the point estimates remained similar, indicating that reverse causality is not likely to explain the associations.

We found an increased risk for greater alcohol intake, in line with the WCRF review.²² When we looked at type of alcoholic drink, we found an association between beer and colorectal cancer. Our results for fibre are in agreement with those reported in a recent meta-analysis on fibre and colorectal cancer, which found that, in analyses by fibre type, only fibre from cereals, but not from fruit or vegetables, was inversely associated with risk.²³ Different types of fibre may have different effects on stool transit time and weight, which may explain the different associations with colorectal cancer, but more research is needed to clarify the physiological effects of different types of fibre. Alternatively, the intake of fibre from cereals will be correlated with wholegrain intake, which has also been shown to be associated with a reduced risk of colorectal cancer.²³ It is possible that phytochemicals or other non-fibre components of wholegrains are responsible for the observed associations. The WCRF review found an inverse association between fish and risk of colorectal cancer; however, this finding was driven by one study and was not robust to removal of that study from the meta-analysis.²² We did not find any associations between fruit and vegetables and colorectal cancer. The WCRF systematic review found a very weak inverse association between vegetables and colorectal cancer, which might be explained by residual confounding, and no association for fruit.²² In our study, the HR for 100-mL/day higher milk intake was 0.93 (95% CI: 0.87–1.01). The WCRF systematic review found an inverse association between milk intake and colorectal cancer [HR for 200 g/day = 0.94 (95% CI: 0.92–0.96)].²⁰ The UK Biobank short food-frequency touchscreen questionnaire did not contain all sources of milk and therefore there may be more measurement error in our categorization of milk intakes, which would tend to bias the association towards the null.

Interestingly, we found heterogeneity by sex for red and processed meat, red meat, processed meat and alcohol, with the association stronger in men and null in women. The explanation for these differences by sex is unclear and could be a chance finding. However, there is evidence that the association between other factors, such as BMI²⁴ and height²⁵ and colorectal-cancer risk, differs between men and women. We also found heterogeneity by sub-site of the colon for the same food items—red and processed meat, processed meat and alcohol—with the associations stronger for distal cancer and null for proximal cancer, and previous work has shown that women are more likely to develop cancers in the proximal colon, whereas men are more likely to develop cancers in the distal colon.²⁶ Future epidemiological studies examining dietary, lifestyle and anthropometric factors in relation to colorectal-cancer risk should examine associations by sex and by site and/or characteristics of the tumour.

Strengths of this study are the large number of participants and the ability to adjust for known and potential confounders. We have previously reported good reproducibility of the short food-frequency touchscreen questionnaire.¹¹ This indicates that the participants’ reporting of food groups, including meat consumption, in UK Biobank is stable. We were also able to reduce the impact of regression dilution bias and other measurement errors, and quantify actual intakes, using the re-measured dietary intakes on the large sub-sample of participants who completed at least one 24-hour dietary assessment. Recruitment for UK Biobank is recent, from 2006 to 2010, so the results we present here indicate that red and processed meat, as is typically consumed in the present day, is associated with a higher risk of colorectal cancer. We were not able to adjust for total energy intake because the short food-frequency touchscreen questionnaire did not ask about every food group and therefore estimated total energy intake was not available. Importantly, we adjusted for measures of body size (waist circumference in the main analysis and BMI in a sensitivity analysis) and physical activity; adjusting for body size and physical activity has been shown to better approximate objectively measured total energy expenditure (and therefore true energy intake) than estimated energy intake from a food-frequency questionnaire.²⁷ A large sub-sample completed a 24-hour dietary assessment during follow-up. Previous work has shown that UK Biobank participants who completed at least one 24-hour dietary assessment were more likely to be women, older, of white ethnicity, less deprived and more educated than those that did not complete a 24-hour dietary assessment.¹⁴ Nevertheless, the approach we used to calculate the trends based on re-measured intakes has been shown to be robust to various factors influencing which participants provide a second measure.²⁸ We did not have information about participation in bowel-screening programmes during follow-up and participants who underwent bowel screening would be more likely to be diagnosed with bowel cancer and it is possible that participating in screening could be associated with dietary intakes. Although this study is one of the largest single studies to report on dietary factors and cancers at sub-sites of the colorectum, more data are needed for reliable estimation of the associations between dietary factors and sub-sites of the colorectum.

Conclusions

In conclusion, in this systematic analysis of a contemporary cohort of half a million men and women from the UK population, we found that consumption of red and processed meat and alcohol was associated with an increased risk of colorectal cancer. We also found that fibre from bread and breakfast cereals was associated with a reduced risk. The current recommendation by the UK Government Department of Health is that people should not eat more than 90 g of red and processed meat a day.²⁹ Participants in the highest category of red and processed-meat consumption were consuming an average of 76 g of red and processed meat per day and thus this group was on average meeting the current recommendation but still had a 20% (95% CI: 4–37) increase in risk of colorectal cancer compared with those who ate an average of 21 g of red and processed meat per day. Therefore, our results suggest that reductions in meat intake below the current recommendation may further reduce the risk of colorectal cancer.

Funding

This work was supported by Girdlers’ New Zealand Health Research Council Fellowship (K.E.B.), Cancer Research UK (T.J.K.; C8221/A19170 and 570/A16491), the UK Medical Research Council (T.J.K.; MR/M012190/1) and the Wellcome Trust [T.J.K.; Our Planet Our Health, project: Livestock, Environment and People (LEAP) (205212/Z/16/Z)]. The funders did not influence the design of the study, analysis or interpretation of the data, the writing of this report or the decision to publish.

Supplementary Material