Dietary antioxidants and the aetiology of pancreatic cancer: a cohort study using data from food diaries and biomarkers
- Paul J R Banim1,
- Robert Luben2,
- Alison McTaggart2,
- Ailsa Welch1,
- Nicholas Wareham3,
- Kay-Tee Khaw2,
- Andrew R Hart1,4
- 1Department of Medicine, University of East Anglia, Norwich, UK
- 2Institute of Public Health, University of Cambridge, Cambridge, UK
- 3MRC Epidemiology Unit, University of Cambridge, Cambridge, UK
- 4Department of Gastroenterology, Norfolk and Norwich University Hospital NHS Trust, Norwich, UK
- Correspondence to Dr Andrew R Hart, Norwich Medical School, University of East Anglia, Norwich NR4 7TJ, UK;
Contributors All authors were involved in the study concept, design, analysis and interpretation of data, drafting and critical revision of the manuscript and administration of the project. PJRB collected the data on patients with pancreatic cancer. The EPIC Study is coordinated and has been developed, together with the acquisition of baseline data by K-TK, NW, RL, AMT and AW. ARH supervised this project.
- Accepted 5 June 2012
- Published Online First 23 July 2012
Objective To investigate whether the dietary antioxidants vitamins C and E, selenium and zinc decrease the risk of developing pancreatic cancer, for the first time using 7-day food diaries, the most accurate dietary methodology in prospective work.
Design 23 658 participants, aged 40–74 years, recruited into the EPIC-Norfolk Study completed 7-day food diaries which recorded foods, brands and portion sizes. Nutrient intakes were calculated in those later diagnosed with pancreatic cancer and in 3970 controls, using a computer program with information on 11 000 foods. Vitamin C was measured in serum samples. The HRs of developing pancreatic cancer were estimated across quartiles of intake and thresholds of the lowest quartile (Q1) against a summation of the three highest (Q2–4).
Results Within 10 years, 49 participants (55% men), developed pancreatic cancer. Those eating a combination of the highest three quartiles of all of vitamins C and E and selenium had a decreased risk (HR=0.33, 95% CI 0.13 to 0.84, p<0.05). There were threshold effects (Q2–4 vs Q1) for selenium (HR=0.49, 95% CI 0.26 to 0.93, p<0.05) and vitamin E (HR=0.57, 95% CI 0.29 to 1.09, p<0.10). The HRs of quartiles for antioxidants, apart from zinc, were <1, but not statistically significant. For vitamin C, there was an inverse association with serum measurements (HR trend=0.67, 95% CI 0.49 to 0.91, p=0.01), but the threshold effect from diaries was not significant (HR=0.68, 95% CI 0.37 to 1.26).
Conclusion The results support measuring antioxidants in studies investigating the aetiology of pancreatic cancer. If the association is causal, 1 in 12 cancers might be prevented by avoiding the lowest intakes.
- Pancreatic cancer
- dietary antioxidants
- cancer epidemiology
- diverticular disease
Significance of this study
What is already known on this subject?
Known risk factors for pancreatic cancer are cigarette smoking and diabetes.
There is minimal and conflicting data on whether dietary antioxidants, which have anticarcinogenic effects, prevent pancreatic cancer.
No studies have used 7-day food diaries, the most accurate method of measuring diet in large-scale epidemiological studies, to investigate antioxidants.
What are the new findings?
Higher intakes of all the antioxidants selenium and vitamins C and E reduced the risk of pancreatic cancer by two-thirds.
For higher intakes of selenium and vitamin E, the risk of pancreatic cancer was associated with a reduction in risk of at least 40%.
Inverse associations were detected for vitamin C from both the food diary and serum data, but were only statistically significant in the latter.
How might it impact on clinical practice in the foreseeable future?
These results support the accurate measurement of dietary antioxidants in studies investigating the aetiology of pancreatic cancer, the malignancy with the worst prognosis of any cancer.
Trials of antioxidants in the general population and those at high risk should include pancreatic cancer as an end point.
Worldwide, adenocarcinoma of the pancreas causes more than a quarter of a million deaths annually and is the eighth commonest cause of death from cancer.1 Each year in the UK, this malignancy is diagnosed in 7500 people and it is the sixth commonest cause of cancer death, with a lifetime risk of 1 in 86.2 The survival rates are the worst for any tumour with a mortality to incidence ratio of 98%. Only one in six patients survives beyond 1 year and just 3% for more than 5 years,3 ,4 with most treatments involving palliative measures.
A detailed understanding of the aetiology of pancreatic cancer is required, so that preventive public health measures can be recommended. The aetiology of the cancer is poorly understood, although recognised risk factors include family history,5 smoking6 and type 2 diabetes.7 The risk for people with a positive family history increases as the number of relatives affected increases, rising to 32 times in those with three affected relatives.5 Cigarette smoking approximately doubles the risk of pancreatic cancer, and the magnitude of the effect is related to its duration and intensity.6 The mechanism for this association probably involves carcinogens reaching the pancreas via first the blood stream and second through refluxed bile. Type 2 diabetes7 nearly doubles the risk, although the biological mechanisms for this association are unknown. Possibilities include common aetiological agents, the metabolic consequences of diabetes including increases in body mass index (BMI), or residual confounders. Positive associations with pancreatic cancer have also been reported with certain occupations, chronic pancreatitis, Helicobacter pylori infection, and inverse associations with aspirin and statin use.8
Diet may be involved in the aetiology of pancreatic cancer and dietary variations between countries may explain the differences in incidence. For antioxidants, including vitamins C and E, selenium and zinc, there are several plausible biological mechanisms by which they might prevent pancreatic cancer, including inactivating free radicals and reducing oxidative DNA damage, stimulating immune function9 and through genetic effects.10 Antioxidants may interact with established risk factors, such as smoking, by inactivating pro-oxidant carcinogens in tobacco. To confirm if these micronutrients prevent cancer, epidemiological work is needed to examine whether patients with pancreatic cancer eat fewer antioxidants than controls before the development of symptoms. The strongest such nutritional aetiological data would be obtained from prospective cohort studies, which have several methodological advantages over case–control studies—namely, minimising both recall and selection biases.
Few cohort studies have investigated antioxidants and pancreatic cancer, although the Finnish male smokers cohort study of 27 111 participants reported no associations for the dietary intakes of vitamins C and E and selenium,11 but found that higher serum levels of vitamin E nearly halved the risk.12 In 13 979 residents of a retirement community, a higher intake of vitamin C was associated with a non-statistically significant decreased risk.13 Similarly, prospective studies investigating food groups rich in antioxidants—namely, fruits and vegetables, mainly reported null associations.14 However, methodological problems in accurately measuring diet may make any true differences difficult to detect. Existing work used food frequency questionnaires (FFQs), where subjects recorded their frequency of consumption of standard portion sizes of selected food items. FFQs, although quicker to complete are less accurate in measuring diet than detailed food diaries.15 Diaries have shown closer associations with biomarkers and weighed records than other questionnaire-based dietary assessments.15 For example, the correlation coefficients from food diaries for vitamin C and total energy intake are 0.70 and 0.59, compared with those from 24 h recalls of 0.54 and 0.42, using weighed records as the standard.15 Previously, no such cohort studies have used diaries to investigate nutrients in the aetiology of pancreatic cancer.
The aim of this investigation was to conduct a prospective cohort study of dietary antioxidants in the aetiology of pancreatic cancer using nutritional data derived for the first time from 7-day food diaries (7-DFDs). Consistent results from such work showing benefits of antioxidants would support public health recommendations for preventing pancreatic cancer.
Materials and methods
The cohort comprised 23 658 men and women, aged 40–74 years, who completed food diaries, recruited into the European Prospective Investigation of Cancer-Norfolk Study (EPIC-Norfolk) between the years 1993 and 1997. Participants were resident in the county of Norfolk, UK registered in 35 general practices in rural, suburban and inner city areas. The Norwich District Health Authority ethics committee approved the study and all participants gave signed consent for their medical notes to be reviewed in the future. At recruitment, participants completed detailed questionnaires on their demography, previous medical history, medication and smoking. Participants attended a baseline health check, supervised by a nurse, who took non-fasting blood samples and recorded anthropometric measurements.16 The non-fasting blood samples were immediately transported to the laboratory for analyses, including serum vitamin C. At the baseline health check, the nurse explained the 7-DFD, the first day of which was recorded with the nurse, as a 24 h recall of the participant's previous day's dietary intake. The remaining 6 days were completed by participants themselves at home, who recorded their entire dietary intake, including food types, portion sizes, brands, cooking methods and recipes in eight separate meal times each day. The name of commercially prepared foods consumed was included in the diary to allow more accurate nutritional assessments. Portion sizes were estimated by comparison with supplied photographs of varying household measures or recognisable standard units.
The 7-DFDs were returned to the study headquarters where they were coded by trained nutritionists, and the data input into a specially designed computer program called DINER (Data Into Nutrients for Epidemiological Research).17 Each entry in the diary was matched to one of 11 000 food items and 55 000 portion sizes within DINER, by selecting the food item which best described it. DINER facilitated the translation from participant-reported free text of food to structured data, which were electronically converted into nutrient values. The underlying nutrient database is based on the UK food composition database.18 Each 7-DFD took approximately 2.5 h to code and interpret, with an average of 220 individual food and drink items reported by participants in their diaries. An example of the detail of this method was that 337 specific types and brands of breakfast cereals were included in DINER. The computer program checked for potential errors in the coded diaries, such as unexpectedly large portion sizes or duplication, and any anomalies were checked by the nutritionists.
The cohort was subsequently monitored after recruitment to identify those participants who developed incident pancreatic cancer up to June 2010. Pancreatic cancer cases were identified by matching the EPIC-Norfolk database with the Norfolk Health Authority records of hospital admissions, the Eastern Cancer Registry and Information Centre. The notes of all potential cases were reviewed by a medical gastroenterologist to verify the cancer diagnoses and to obtain information on confirmatory investigations. Cases were excluded if there was diagnostic uncertainty, participants had pancreatic cancer before enrolment or if the diagnosis was made within 12 months of entering the study. This last reason for exclusion ensured that the dietary data were truly prospective before the development of symptoms.
In the assessment of food diary data a case–cohort analysis was performed between cases and a random sample of 3970 food diaries from controls, for which data were available. The baseline characteristics were compared between participants with and without incident pancreatic cancer using a t test for normally distributed continuous variables, a Mann–Whitney U test for non-parametric continuous variables and a χ2 test for categorical ones. Each antioxidant was divided into quartiles of intake across the distribution of the whole cohort. Cox proportional hazards regression models estimated the HRs, with 95% CIs of developing pancreatic cancer for quartiles of intake of each antioxidant, using the lowest quarter as the baseline value. To evaluate possible threshold effects, the quartile of intake for participants eating the amounts of antioxidants in the range of the lowest fourth of intake (Q1) was compared with a summation of the range of the three higher quartiles (Q2+Q3+Q4).
All analyses were adjusted for the covariates of age at recruitment, gender, smoking, diabetes, BMI and total energy intake. In a separate analysis, the HRs were also adjusted for the use of supplements containing that specific antioxidant—for example, dietary vitamin C adjusted for supplements containing vitamin C. For antioxidants for which inverse associations were documented, the HR was calculated for eating the lowest quarter of all of these antioxidants, compared with a summation of the three higher ones. The associated attributable fraction was computed—that is, the percentage of all cases in the lowest quartile which might be prevented if a causal association exists, by consuming the three higher quarters of antioxidants. As smoking induces oxidative stress, the effect of consuming the highest three quartiles of all of the antioxidants compared with the lowest, was calculated in both smokers and non-smokers. A similar analysis for all antioxidant intake was done for those with and without histology.
In the analyses, the primary outcomes were HRs after 10 years of follow-up as this period was estimated as the time during which a single baseline measure of dietary intake or serum vitamin C might be representative of future intake. The secondary outcome was the risk of pancreatic cancer after the full follow-up time of 17 years after recruitment. For the analysis of serum vitamin C concentrations, a full cohort analysis was performed using serum data on 22 474 (95%) participants who gave blood samples. These serum values were divided into quartiles and the HRs adjusted for age at recruitment, gender, smoking and diabetes. Vitamin C levels in blood correlate significantly with those estimated from diet diaries, with a 5.6 µmol/l of vitamin C per one quintile increase in dietary vitamin C for the food diary (p trend<0.001).19
In the cohort of 23 658 participants (55% women) who attended the baseline health check, 49 participants developed pancreatic cancer (55% men) between 1 and 10 years of follow-up, which increased to 86 (44% men) after 17 years (table 1). The mean time between recruitment and diagnosis in those diagnosed within the first 10 years was 5.8 years (SD 2.4) and for 17 years it was 8.6 years (SD 3.8). For 10 years, most patients had either local or metastatic spread (72%) and there was histological confirmation in 35%. For the patients without histology, the diagnosis was achieved with at least two radiological imaging modalities. The clinical characteristics of the patients with and without histology were similar—namely, local or metastatic spread (78% vs 73%, p=0.65) and median survival (3 vs 4 months, p=0.73). In the 10-year group, the treatments included surgery (13%), chemotherapy (38%) and palliative measures (50%). The median survival of these patients after diagnosis was 5.0 months (range 0.5–25.0) and for those within 17 years it was 6.0 months (range 0.3–25.0). Compared with controls, the patients diagnosed after 10 years were more likely to be older at recruitment and former smokers, but there were no differences in median BMI or diabetes (table 1). There was a borderline statistically significant lower median vitamin E intake in cases at both 10 years (p=0.09) and 17 years (p=0.08), but no other differences in the median intakes of other antioxidants. The characteristics of the subcohort of controls used in this analysis were almost identical to the other controls, whose data were not used (age at recruitment 59.3 vs 59.2 years, men 43.8% vs 45.5% and BMI at recruitment 26.3 kg/m2 vs 26.4 kg/m2, respectively).
For the primary outcome—that is, analysing data at 10 years of follow-up, there were inverse associations with each quartile of antioxidant intake and the risk of pancreatic cancer, apart from zinc, although these did not reach statistical significance (table 2), with no trends across quartiles. As the magnitudes of the effect sizes in most quartiles were similar, the threshold effects were calculated. Here, there were inverse associations between the lowest quartile and a summation of the three higher ones for selenium (HR=0.49, 95% CI 0.26 to 0.93, p=0.03) and vitamin E (HR=0.57, 95% CI 0.29 to 1.09, p=0.09), but not vitamin C (HR=0.68, 95% CI 0.37 to 1.26, p=0.22) or zinc (HR=0.91, 95% CI 0.44 to 1.91, p=0.81) (table 3). The threshold effects for all antioxidants were similar when adjusted for supplement use (table 3). Those consuming the three higher quarters of all of vitamins C and E and selenium had a decreased risk of pancreatic cancer (HR=0.33, 95% CI 0.13 to 0.84, p<0.05) at 10 years of follow-up (table 4). Based on these figures and if the association is causal, then 8.2% of all pancreatic cancers could be prevented by avoiding the combined lowest intakes of vitamins C and E and selenium. The HR for consuming the three higher quartiles of all of vitamins C, E and selenium was lower in smokers (HR=0.09, 95% CI 0.01 to 0.69, p=0.02) than in non-smokers (HR=0.41, 95% CI 0.14 to 1.21, p=0.11), although the difference was not statistically significant (p=0.19). When all analyses were performed using dietary data at 17 years the associations were inversely related, but attenuated (tables 2–4). The associations for those with and without histology were similar at 17 years for those in the lowest quartile of all of vitamins C and E and selenium compared with those in a summation of the three higher ones (HR=0.30, 95% CI 0.03 to 2.86, p=0.30 and HR=0.35, 95% CI 0.12 to 0.98, p=0.046 respectively).
In this cohort, 95% of participants donated serum samples for measuring vitamin C. For the serum concentrations at 10 years, the three higher quartiles were all inversely associated with risk (table 5), although only the highest one reached statistical significance (HR=0.19, 95% CI 0.06 to 0.68) with a trend across quartiles (HR trend=0.67, 95% CI 0.49 to 0.91, p=0.01). For the 17-year follow-up data, there were smaller inverse effect sizes (highest vs lowest quartile HR=0.42, 95% CI 0.20 to 0.91, p=0.08, HR trend =0.83, 95% CI 0.67 to 1.03, p<0.10).
The main finding of this study was a threshold inverse association for the development of pancreatic cancer within 10 years of recruitment between those in the lowest quartile and those in a summation of the three higher dietary intakes of all of selenium, vitamins E and C. Participants eating higher intakes of all of these micronutrients were 67% less likely to develop pancreatic cancer than those eating lower amounts. For selenium, the effect size was large with an approximately halving of the risk with greater intakes, and for vitamin E, there was a slightly smaller effect of borderline statistical significance. For vitamin C, inverse associations were seen in the data from both the food diaries and serum, although only the latter was statistically significant. As the effects were mainly threshold effects and the median intake was similar between cases and controls, the potential benefit of these micronutrients is only applicable to those with the lowest intakes. These results support recording the intake of dietary antioxidants in aetiological studies of pancreatic cancer. At 17 years of follow-up, the associations were also inverse, but attenuated compared with 10 years. This is likely to be due to a form of measurement error—namely, regression dilution bias. Here the magnitude of any true effect size is underestimated the longer the interval between an initial measurement, such as diet and the development of a disease, owing to changes in participants' diet over time.
These nutritional epidemiological and serum data support the experimental work on biological mechanisms for the inhibition of carcinogenesis by dietary antioxidants, including the neutralisation of free radicals, genetic effects and stimulation of the immune system. Antioxidants scavenge free radicals, and smoking and diabetes, which are established risk factors for pancreatic cancer, induce oxidative stress and free radical production. Selenium is present in cereals, nuts, fish and meat, with the amount dependent on the mineral content in soil. Selenium is incorporated into selenoproteins, such as glutathione peroxidase, which catalyses the removal of hydroperoxides. Vitamin E, present in high amounts in vegetable oils, nuts, seeds, margarines and egg yolk, inactivates free radicals formed from polyunsaturated fatty acids in lipid cell membranes.20 Vitamin C, from fruits and vegetables, is a reducing agent which detoxifies hydroxyl or superoxide free radicals. A second mechanism for antioxidants is their effect on the inflammatory process, and chronic inflammation may play a role pancreatic carcinogenesis.21 Chronic pancreatitis is associated with the generation of reactive oxygen species and there is a gradual decrease in antioxidant enzyme expression in normal to inflamed to neoplastic tissue.22
The strengths of this cohort design were both its prospective nature and the accuracy of the method for measuring diet. A cohort study has the advantage over retrospective case–control work in that both recall and selection biases are minimised. In an aetiological study it is important that the diet before symptoms appear is reported as once symptoms develop dietary changes may occur. In a case–control study, patients may have difficulty in recalling their presymptomatic intake and report the current diet, altered by symptoms. This recall bias is removed in prospective work, where participants are recruited when they are well and report their current diet. Furthermore, in an aetiological study it is vital that the case and control populations are similar, and a cohort design reduces selection biases. A further strength was the use of the 7-DFDs to measure habitual diet. In large-scale cohort investigations, diet can be assessed using several methods, including diaries, FFQs and 24 h recalls. A validation study, comparing the accuracy of each method, against the standard of 16-day weighed food records, reported the superiority of the 7-DFDs.15 For example, for vitamin C intake the correlation coefficients between weighed records and FFQs was 0.54, whereas with food diaries it increased to 0.70.
The main potential weakness of the study was error in recording habitual diet, first in the use of a single measure of diet and second that changes may occur in people's diet over time. Variations in diet between recruitment and over subsequent years would occur in both those who subsequently develop the disease (cases) and those who do not (controls). This form of measurement error, regression dilution bias, would result in a spurious underestimate of the true effect size, rather than an inaccurate overestimate. To limit the magnitude of this potential error, the primary analysis included patients diagnosed within the first 10 years of follow-up. Similarly, measurement error due to the use of questionnaire-based assessments could be reduced by using repeated measures of dietary intake during follow-up. Although approximately half of the cohort have completed further diaries at 2 and 4 years of follow-up, these have not yet been coded. Importantly, however, the limited available evidence suggests that dietary antioxidant intake remains stable in adulthood. In a UK longitudinal study of repeated 5-day diaries, in adults aged 43 years and repeated at 53 years, the intakes of vitamin E and zinc, were similar, although vitamin C did increase.23 However, in a study from The Netherlands of adults aged 55–69 years, where diet was assessed annually over 14 years using a FFQ, there were moderate correlation coefficients for vitamin C between repeated measurements for men (r=0.63) and women (r=0.64).24
Another limitation was that only 35% of cases had histological confirmation of cancer, which might have introduced a misclassification error. Ideally, such information would be available, but for many of the years during follow-up, more recently introduced techniques for obtaining tissue were unavailable, such as endoscopic ultrasound. However, there were clinical similarities between cases with and without histology, including cancer stage and survival. For patients with no histological confirmation, all medical notes were thoroughly reviewed and supported a diagnosis of pancreatic cancer.
Although the study was generalisable in its demography and clinical aspects of the disease, it was conducted in just one county in England. However, the findings are generalisable in that the incidence, clinical characteristics of cancer staging, treatments and survival were similar to those expected.25
The associations may be explained by residual confounding, including other nutrients which are associated with the antioxidants we studied. Our work focused on particular dietary antioxidants investigated in previous reports to develop clarification of their role, although there are others antioxidants with such properties. These include the carotenes—namely vitamin A, β-carotene and lycopene, and also dietary iron. Our work will progress to investigate these, although some are present in only limited food sources and may have multiple complex anticancer effects, in addition to reducing oxidative stress. For example, the main dietary source of lycopene is tomatoes and this carotene may have several anticancer effects, including reducing oxidative stress and modulating carcinogen metabolism, immune status and angiogenesis.26 Iron, in addition to its antioxidant effects, can also have pro-oxidant ones.
A final limitation was we could not investigate that any genetic factors which might influence pancreatic cancer risk were not investigated. There are genetically programmed enzymes with antioxidant properties, such as superoxide dismutase, for which we had no data. Intracellular antioxidant enzymes, including superoxide dismutase and catalase, are lower in pancreatic cancer cells, than normal ones.22 One confounder, reported in other work, which was considered in this analysis, was cigarette smoking which provides pro-oxidants. The inverse association with total antioxidant intake was greater in smokers than non-smokers, although the difference was not statistically significant. This differential effect supports the protective effect of antioxidants, and suggests the need to investigate other factors linked with oxidative stress. Oxidative stress induces DNA lesions, including 8-hydroxy-2-deoxyguanine which promotes nucleotide base transversions in oncogenes and tumour suppressor genes leading to carcinogenesis.27 The investigation of pro-oxidants in the diet, including fat and red meat, would be a useful line of research.
To confirm that dietary antioxidants do have a causal role in preventing pancreatic cancer, supportive data from many epidemiological studies are required, ideally from prospective investigations or randomised controlled trials. To the best of our knowledge, only two such cohort investigations have used nutritional information from questionnaires. The Finnish male smokers cohort study of 27 111 participants reported no associations with the dietary intakes of vitamins C, E and selenium.11 However, higher serum levels of vitamin E nearly halved the risk,12 although the findings from other and much smaller studies on serum levels were inconsistent.28 ,29 In another cohort study of 13 979 residents of a retirement community, higher intakes of vitamin C were associated with a statistically non-significant reduced risk of pancreatic cancer.13 A randomised controlled trial of vitamin E supplements did not find any effect on the incidence of pancreatic cancer (RR=0.96, 95% CI 0.56 to 1.67), although this was in male smokers.30 Lower serum levels of selenium more than quadrupled the risk,29 but the association between serum vitamin C levels and pancreatic cancer risk has not previously been investigated. A meta-analysis of five cohort studies and four case–control studies reported an inverse association with citrus fruit intake, a rich source of vitamin C (summary OR=0.83, 95% CI 0.70 to 0.98), although there was a large degree of heterogeneity across investigations.31 However, two large prospective studies—namely, EPIC-Europe of 520 000 participants32 and a cohort of 81 922 from Sweden33 reported no associations with ‘total fruit’ or ‘total vegetable intake’. The current uncertainty about antioxidants may be due to small numbers in some work, measurement error in recording their intake, which is compounded as antioxidants are present in many foods, and finally, unrepresentative populations.
In summary, this prospective cohort study, using dietary antioxidant data, derived for the first time from food diaries, demonstrated inverse associations with total dietary antioxidant intake, selenium and vitamin E. A role for vitamin C was supported by the serum analysis with large effect sizes. The current evidence from aetiological epidemiological work is minimal and inconsistent, although our work using an accurate method of dietary assessment and other studies using biomarkers of antioxidants suggest that vitamins C and E and selenium may be protective. While null results from dietary antioxidant supplementation trials for other cancer end points have been discouraging, food sources of these nutrients may have different health effects from high-dose single supplements. If a causal association is confirmed by reporting consistent findings from other epidemiological studies, then population-based dietary recommendations may help to prevent pancreatic cancer.
The authors thank the EPIC-Norfolk participants who took part in this study and their general practitioners. The late Professor Sheila Bingham, a principal investigator of EPIC, whose work developed the nutritional analyses methodology.
Funding This analysis was funded by a grant from The Big C Cancer Charity, Norfolk, UK (08-10R). The EPIC-Norfolk Study is supported by research programme grant funding from The Medical Research Council, UK and Cancer Research, UK.
Competing interests None.
Patient consent Obtained.
Ethics approval Ethics approval was provided by Norwich District Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.