Diet‐wide association study of 92 foods and nutrients and lung cancer risk in the European Prospective Investigation into Cancer and Nutrition study and the Netherlands Cohort Study

Abstract It is unclear whether diet, and in particular certain foods or nutrients, are associated with lung cancer risk. We assessed associations of 92 dietary factors with lung cancer risk in 327 790 participants in the European Prospective Investigation into Cancer and Nutrition (EPIC). Cox regression yielded adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) per SD higher intake/day of each food/nutrient. Correction for multiple comparisons was performed using the false discovery rate and identified associations were evaluated in the Netherlands Cohort Study (NLCS). In EPIC, 2420 incident lung cancer cases were identified during a median of 15 years of follow‐up. Higher intakes of fibre (HR per 1 SD higher intake/day = 0.91, 95% CI 0.87‐0.96), fruit (HR = 0.91, 95% CI 0.86‐0.96) and vitamin C (HR = 0.91, 95% CI 0.86‐0.96) were associated with a lower risk of lung cancer, whereas offal (HR = 1.08, 95% CI 1.03‐1.14), retinol (HR = 1.06, 95% CI 1.03‐1.10) and beer/cider (HR = 1.04, 95% CI 1.02‐1.07) intakes were positively associated with lung cancer risk. Associations did not differ by sex and there was less evidence for associations among never smokers. None of the six associations with overall lung cancer risk identified in EPIC were replicated in the NLCS (2861 cases), however in analyses of histological subtypes, inverse associations of fruit and vitamin C with squamous cell carcinoma were replicated in the NLCS. Overall, there is little evidence that intakes of specific foods and nutrients play a major role in primary lung cancer risk, but fruit and vitamin C intakes seem to be inversely associated with squamous cell lung cancer.

.03-1.14), retinol (HR = 1.06, 95% CI 1.03-1.10) and beer/cider (HR = 1.04, 95% CI 1.02-1.07) intakes were positively associated with lung cancer risk. Associations did not differ by sex and there was less evidence for associations among never smokers. None of the six associations with overall lung cancer risk identified in EPIC were replicated in the NLCS (2861 cases), however in analyses of histological subtypes, inverse associations of fruit and vitamin C with squamous cell carcinoma were replicated in the NLCS. Overall, there is little evidence that intakes of specific foods and nutrients play a major role in primary lung cancer risk, but fruit and vitamin C intakes seem to be inversely associated with squamous cell lung cancer.

K E Y W O R D S
cohort study, diet, foods, lung cancer, nutrients

What's new?
Lung cancer is the leading cause of cancer death worldwide, and while smoking is the most significant factor affecting lung cancer risk, other environmental and genetic factors may contribute. Here, the authors looked for an association between the foods people eat and their lung cancer risk. Using an approach they call a diet-wide association study, modelled after genome-wide association studies, they evaluated 92 individual food and nutrient intakes for association with lung cancer risk. No major associations were detected for lung cancer overall, but higher fruit and vitamin C intakes were associated with lower squamous cell lung cancer risk.

| INTRODUCTION
Lung cancer (including nonsmall cell and small cell lung cancer) is the most frequently diagnosed cancer in men and third most common cancer in women, and remains the leading cause of cancer death. 1 Tobacco smoking is overwhelmingly the main risk factor for lung cancer, but other factors such as genetic susceptibility, occupational exposures, air pollution, radon, lifestyle factors and diet might also play a role. 2 The World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR) Third Expert Report on diet, nutrition, physical activity and lung cancer suggested that certain dietary factors might contribute to the risk of primary lung cancer. 3 In particular, the report concluded there is some (albeit limited) evidence that drinking alcohol and consumption of red meat and processed meat might be associated with a higher risk of lung cancer, while consuming foods containing retinol, beta-carotene or carotenoids might be associated with a lower risk. 3 Since smoking is the predominant risk factor for lung cancer, diet-lung cancer associations might be confounded by or differ according to smoking status. 2,4 It is also possible that dietary factors modify the effects of tobacco exposure, and protective or harmful effects might be restricted to or greater in smokers. 3,5 There is some evidence of lower lung cancer risk associated with higher intake of fruit and vegetables in current and former smokers, 4 foods containing vitamin C in current smokers and isoflavones in never smokers. 3 There remains uncertainty about the role of diet in primary lung cancer risk, with the bulk of existing data stemming from case-control studies (which are prone to recall and selection biases) or studies that employed suboptimal assessment methods of dietary intake, were underpowered, and/or did not adequately adjust for smoking or examine associations by smoking status or for different histological subtypes of lung cancer.
We sought to evaluate a comprehensive list of individual foods and nutrients in relation to risk of lung cancer using a diet-wide association study approach. 6 A case-cohort approach was used in the NLCS for efficiency reasons. 15 A subcohort of 5000 participants was randomly sampled from the cohort immediately after recruitment, for whom vital status information was acquired biennially to estimate person-time at risk for the full cohort. Incident cancer cases in the entire cohort were identified by record linkage to cancer and pathology registries. This analysis excluded participants with cancer (other than nonmelanoma skin cancer) prior to recruitment as well as those with incomplete or missing dietary or confounder data.

| Outcome ascertainment
In EPIC, cancer and mortality data were obtained from populationbased cancer and mortality registries (in Denmark, Italy, the Netherlands, Norway, Spain, Sweden and the United Kingdom) or a combination of methods including cancer and pathology registries, health insurance records and active follow-up of participants or their next-of-kin (in France, Germany and Naples, Italy). 13

| Dietary assessment
In the EPIC study, usual diet during the preceding 12 months was assessed at enrolment using validated country-specific or study centre-specific dietary questionnaires or food records. 13 except Ragusa (Italy) and Spain, where interviewers were used. In Malmö (Sweden), a food record was used for cooked meals and a food frequency questionnaire was used for breakfast and foods consumed between the main meals. 17 Standardised nutrient intakes were calculated using the EPIC Nutrient Database. 18 The current analysis included 92 dietary factors (63 foods and 29 nutrients; Appendix S1) for which data were available in the centralised EPIC database for at least eight out of the nine countries. The dietary factors were not mutually exclusive, for example, apple/pear, bananas, berries, citrus fruit, grapes and stone fruit were investigated separately as well as total fruit which included all of these types of fruit. Likewise, total alcohol intake was investigated as well as individual alcoholic beverage types: beer/ cider, spirits, wine and fortified wine. In addition, the individual food items also contribute to nutrients.
Information on dietary intake in the NLCS was collected at recruitment using a 150-item semiquantitative food frequency questionnaire that estimated the average frequency and amounts of foods and beverages habitually consumed in the previous 12 months. The food frequency questionnaire has been validated and tested for reproducibility. 19,20 Nutrient intakes were calculated by multiplying the frequency of intake by the nutrient content of specified portions based on the Dutch food composition table. 21 Data were available for the same dietary factors as in EPIC, but replication analyses were only performed in the NLCS for foods and nutrients that were identified to be associated with lung cancer risk in EPIC.

| Statistical analysis
Cox proportional hazards regression models with age as the time scale were fitted to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for a 1 SD higher intake of each food or nutrient per day.
Age at recruitment was the entry time and age at cancer diagnosis (except nonmelanoma skin cancer), death, emigration or last followup, whichever occurred first, was the exit time. Intakes of foods and nutrients were adjusted for energy intake using the residual method, 22 Table 2. Compared to EPIC, the NLCS included a higher proportion of men and participants were on average older and more likely to be smokers.

| Diet-wide association study in EPIC
Of the 92 foods and nutrients that were evaluated in the EPIC study, six were associated with risk of lung cancer using the FDR threshold of 0.05 (Figure 1 and Appendix S1). There were inverse associations with lung cancer risk for consumption of fibre (HR for a 1 SD incre- Intakes of offal (HR for a 1 SD increment in intake per day = 1.08, 95% CI 1.03-1.14), retinol (HR = 1.06, 95% CI 1.03-1.10) and beer/ cider (HR = 1.04, 1.02-1.07) were positively associated with lung cancer risk.
The analyses in the nested case-control sample additionally adjusted for circulating cotinine concentrations supported these associations (albeit with wide CIs), with the exception of fibre intake which was not associated with lung cancer risk after adjustment for cotinine ( Figure S1).
The inverse associations for intakes of fibre, fruit and vitamin C, and positive associations for intakes of retinol and beer/cider, were  present in former and current smokers but not in never smokers; however, there were only 278 lung cancer cases among never smokers, and P-values for heterogeneity by smoking status were large, except for vitamin C (P = .06) (Figure 2). Associations did not differ substantially by sex (P ≥ .27) (Figure 3).  Figure 4 and Table S2). There were no substantial differences in the associations by histological subtype for intakes of the other foods and nutrients identified in the diet-wide association study (offal, retinol and beer/cider).
Pairwise correlations for the 92 foods and nutrients are displayed in Appendix S2. Fruit intake was correlated with intakes of vitamin C (0.71) and fibre (0.52). There were also notable correlations between intakes of fibre and vitamin C (0.58), and between offal and retinol (0.67).

| Replication in the NLCS
In the replication analysis for overall lung cancer, none of the associations for the six dietary factors that were identified in the EPIC study were confirmed in the NLCS (Figure 4 and Tables S3 and S4).
The positive association for beer/cider intake was of similar magnitude and in the same direction, while associations for intakes of offal and retinol were weaker in the NLCS, but estimates were accompanied by considerable uncertainty. Despite inverse associations in the EPIC study, intakes of fibre, fruit and vitamin C did not appear to be associated with risk of overall lung cancer in the NLCS. However, in analyses by histological subtype of lung cancer, intakes of fruit and vitamin C were inversely associated with squamous cell carcinoma (HR per 1 SD increment in intake/day = 0.86, 95% CI 0.77-0.95 and 0.88, 95% CI 0.80-0.98, respectively) ( Figure 4 and Table S2).

| DISCUSSION
In this systematic evaluation of 92 foods and nutrients and risk of lung cancer, higher intakes of fibre, fruit and vitamin C were associated with a 9% lower risk per SD increment in daily intake, while offal, retinol and beer/cider intakes were associated with a higher risk of lung cancer in the EPIC study. The inverse associations for fibre, fruit and vitamin C intakes, and positive association for retinol intake with lung cancer risk were evident in former and current smokers but not in never smokers. The associations with overall lung cancer were not replicated when assessed in the independent NLCS. However, when considering histological subtypes of lung cancer, inverse associations observed in EPIC were replicated in the NLCS for intakes of fruit and vitamin C with squamous cell carcinoma.
In a nested case-control sample of EPIC participants, associations for the six identified dietary factors were in the same direction and of ≤30, >30 to 60, >60 to 90, >90 min/day), highest level of education (in EPIC: none/primary school, technical/professional school, secondary school, longer education including university; in NLCS: primary school or lower vocational, secondary or medium vocational, higher vocational or university), family history of lung cancer (no, yes; NLCS only), history of diabetes (no, yes) and energy intake (kcal/day, continuous). For offal, only liver consumption was available in the NLCS similar magnitude after adjusting for circulating cotinine concentrations, apart from fibre intake, for which there was no association.
Nevertheless, the association for fibre was null in the NLCS, and this is consistent with the WCRF/AICR report which found a lack of evidence to suggest that dietary fibre intake is associated with primary lung cancer risk. 3 In EPIC, there was a higher risk of lung cancer associated with beer/cider consumption, but not with other alcoholic beverage types (wine, spirits) or total alcohol intake. Although a meta-analysis found a positive association between beer and lung cancer risk in those consuming an average of one or more drinks per day, 26 evidence from previous studies has suggested a higher risk of lung cancer for overall alcohol consumption but not beer, 3 however most studies have not distinguished between types of alcoholic drinks. Because smoking and alcohol consumption are strongly correlated, the relationship between alcohol and lung cancer risk reported in many studies, even after adjustment for smoking, might be biased by residual confounding. A meta-analysis found no association between alcohol consumption and lung cancer risk in never smokers. 27 In the current analysis, beer/cider intake was positively associated with lung cancer risk in current smokers (with an identical point estimate for former smokers), but not associated with lung cancer in never smokers.
Associations with intakes of other dietary factors in EPIC also appeared to differ according to smoking status, with inverse associations for fibre, fruit and vitamin C, and a positive association for retinol in former and current smokers but not in never smokers. The findings for fruit and vitamin C are consistent with the WCRF/AICR report, which identified some evidence of lower lung cancer risk for higher consumption of fruit in current and former smokers, and vitamin C in current smokers. 3 A meta-analysis similarly found an inverse association between fruit intake and lung cancer risk in smokers but not in never smokers. 4 Fruit is a source of vitamin C as well as other antioxidants and various phytochemicals which might ameliorate some of the effects of tobacco exposure in multiple pathways involved in lung carcinogenesis. 3 Antioxidant-rich foods might have greater benefits in smokers, and this might explain why the observed associations were restricted to ever smokers and squamous cell carcinoma. 5 The higher risk of lung cancer associated with higher intake of retinol in EPIC is inconsistent with previous studies suggesting no association for dietary retinol intake and an inverse association for serum retinol concentrations. 3 However, long-term use of retinol supplements was associated with a higher risk of lung cancer in the VITamins And Lifestyle (VITAL) cohort study, 28 and there is evidence that taking high-dose beta-carotene supplements (a retinol precursor) is associated with a higher risk of lung cancer in current and former smokers. 3,29,30 Our findings of a positive association between retinol intake and lung cancer risk in ever smokers but not in never smokers suggests that retinol potentially modifies the effects of tobacco exposure. It is thought that there is an interaction between smoking, betacarotene and glutathione-S transferase (GST) genetic variants such that high-dose beta-carotene supplementation-potentially leading to supra-physiological concentrations of beta-carotene-confers a higher risk of lung cancer mainly in heavy smokers without the GSTM1 The positive association between offal intake and lung cancer risk identified in EPIC was not replicated in the NLCS. A meta-analysis of four studies found no association between offal intake and lung cancer risk in nonsmokers, 34 but higher offal consumption was associated with a higher risk of lung cancer among heavy smokers participating in a lung cancer screening programme in Italy. 35 Previous studies have found a higher risk of lung cancer associated with higher red meat and processed meat consumption, 3,36-38 including among nonsmokers (for red meat), 34 whereas no association was found in the current study. Although there are plausible mechanisms linking meat consumption with carcinogenesis (albeit also often mediated by cooking methods used), 36 mechanisms specific to lung cancer have not been identified. 3 The lack of association using the systematic diet-wide association study approach suggests that red or processed meat consumption is unlikely to have a major effect on lung cancer risk.
Strengths of our study include the large study population with many lung cancer cases, long follow-up duration, wide variation in diet, extensive information on potential confounders and the systematic approach which assessed a comprehensive set of foods and nutrients while accounting for multiple comparisons, and replication of findings in an independent cohort. Further strengths were the ability to examine associations with further adjustment for circulating concentrations of cotinine and according to smoking status, sex and histological subtype. The main limitation was that analyses were based on a single assessment of diet at recruitment and changes in dietary habits were not considered. For consistency, analyses of each food and nutrient were adjusted for total energy intake, but this introduces a nonspecified substitution model for energy-providing foods and nutrients, which is not biologically optimal. The interpretation of results is therefore not straightforward; for example, for smoking antagonists, absolute intake may be more relevant. Because of the nature of the diet-wide association study approach, dietary patterns were not evaluated. In addition, mutual adjustment for dietary exposures was not performed and intercorrelations were not accounted for. Since many of the dietary items share common sources of intake, correlations of approximately 0.5 to 0.7 were found between several of the foods and nutrients associated with lung cancer risk in EPIC, and it is therefore difficult to disentangle their independent effects.
Although analyses were rigorously adjusted for smoking habits, resid-