Introduction
Cancers of the esophagus (sixth place) and stomach (third place) were amongst the most common causes of cancer-related death in the world in 2012 [
1]. Two histologic types of esophageal cancer can be distinguished, namely esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma (EAC) [
2]. Based on anatomic location, gastric cancers are subdivided into gastric cardia adenocarcinomas (GCA) and gastric non-cardia adenocarcinomas (GNCA) [
3]. Different etiologies have been suggested for these subtypes [
2,
3]. In the past decades, incidence rates of EAC and GCA have been rising in many European countries and the United States (US) [
4,
5].
The traditional Mediterranean diet (MD) is characterized by a high consumption of vegetables, fruits, whole grains, and other plant foods, with olive oil as the principal source of fat. Foods from animal origin are consumed in low amounts in the MD, whereas alcohol intake is moderate [
6‐
8]. The relation between
a priori defined MD adherence and the incidence of esophageal and/or gastric cancer (subtypes) has been the topic of a limited number of studies [
9‐
13]. In these studies, higher MD adherence has been associated with reduced risks of ESCC, GCA, GNCA, and total gastric cancer (GC), but results were not always significant and sometimes inconsistent, primarily with respect to the gastric cancer subtypes [
9‐
13].
Information bias due to reversed causation is a major concern when investigating relations between dietary factors and gastrointestinal cancer risk, because preclinical disease symptoms may cause patients with gastrointestinal tumors to alter their dietary habits already before clinical diagnosis. Another concern is recall bias, which could particularly be a problem in case–control studies. For these reasons, the effects of dietary factors on gastrointestinal cancers should be investigated prospectively, if possible. So far, associations between MD adherence and risks of esophageal and/or gastric cancer subtypes have been prospectively investigated in only two cohort studies [
10,
11]. Therefore, more prospective evidence on this topic is desired.
This study prospectively investigated the association of MD adherence with the risk of esophageal and gastric cancer subtypes (ESCC, EAC, GCA, and GNCA) in the Netherlands Cohort Study (NLCS). We assessed MD adherence using various a priori defined MD scores, with and without alcohol, and examined associations for men and women separately.
Results
Sex-specific baseline characteristics of cases and subcohort members are presented in Table
1. Male ESCC, GCA, and GNCA cases had lower MD adherence than subcohort members. No clear differences in MD adherence were observed between female cases and subcohort members. Concerning potential confounding factors, cases were older (except for male ESCC and EAC cases) and less often never smokers (except for female EAC cases) than subcohort members. In addition, alcohol consumption was higher in ESCC and GCA (men only) cases, but lower in EAC cases (women only). Finally, the mean BMI was lower in ESCC cases, but higher in EAC and GCA cases.
Table 1
Baseline characteristics of the NLCS subcohort, ESCC, EAC, GCA, and GNCA cases
aMEDr | 3.9 | (1.6) | 3.2 | (1.5) | 4.0 | (1.6) | 3.7 | (1.5) | 3.6 | (1.5) | 4.0 | (1.6) | 4.0 | (1.5) | 3.9 | (1.5) | 3.8 | (1.7) | 3.7 | (1.7) |
mMEDr | 4.0 | (1.5) | 3.6 | (1.4) | 4.0 | (1.5) | 3.8 | (1.5) | 3.9 | (1.4) | 4.0 | (1.5) | 4.2 | (1.5) | 3.8 | (1.5) | 4.1 | (1.4) | 4.0 | (1.5) |
Age (years)a | 61 | (7) | 61 | (7) | 61 | (6) | 62 | (7) | 62 | (7) | 61 | (7) | 63 | (6) | 62 | (7) | 62 | (7) | 62 | (7) |
Never cigarette smokers (%) | 12.8 | 9.5 | 8.4 | 8.7 | 9.4 | 57.7 | 35.6 | 62.2 | 47.1 | 51.5 |
Cigarette smoking frequency (cig/day)ab | 15 | (10) | 20 | (5) | 20 | (15) | 15 | (15) | 15 | (10) | 10 | (13) | 15 | (10) | 10 | (14) | 10 | (15) | 10 | (15) |
Cigarette smoking duration (years)ab | 36 | (17) | 40 | (15) | 35 | (17) | 37 | (18) | 40 | (15) | 30 | (20) | 39 | (15) | 35 | (16) | 29 | (25) | 30 | (20) |
Higher vocational education or university (%) | 19.3 | 21.7 | 19.9 | 19.0 | 12.0 | 9.5 | 5.1 | 6.7 | 14.7 | 6.4 |
Alcohol consumption (g/day)a | 9.7 | (20.9) | 24.2 | (29.8) | 9.3 | (23.6) | 12.4 | (21.6) | 10.0 | (21.9) | 1.6 | (7.8) | 5.9 | (20.7) | 1.2 | (12.4) | 1.6 | (5.4) | 1.5 | (8.2) |
Daily energy intake (kcal) | 2162 | (501) | 2057 | (453) | 2163 | (474) | 2132 | (563) | 2190 | (504) | 1687 | (392) | 1755 | (384) | 1631 | (410) | 1813 | (338) | 1666 | (386) |
Body mass index (kg/m2) | 24.9 | (2.6) | 24.5 | (3.2) | 25.6 | (2.5) | 25.5 | (2.6) | 24.8 | (2.8) | 25.0 | (3.5) | 24.0 | (3.6) | 26.9 | (4.3) | 26.5 | (4.3) | 25.3 | (3.9) |
Non-occupational physical activity > 60 min/day (%) | 50.7 | 41.5 | 53.6 | 51.4 | 51.5 | 44.3 | 37.3 | 42.2 | 41.2 | 39.0 |
Family history of esophageal cancer (%) | 0.7 | 0.0 | 1.1 | 1.1 | 1.2 | 0.9 | 3.4 | 6.7 | 0.0 | 1.5 |
Family history of gastric cancer (%) | 6.8 | 9.5 | 7.3 | 9.2 | 11.2 | 6.7 | 5.1 | 6.7 | 5.9 | 11.2 |
Tables
2 (men) and
3 (women) show fully adjusted associations between MD adherence, assessed using various MD scores, and the risk of esophageal and gastric cancer subtypes. Not all eligible study participants could be included in the Cox models because of missing values in covariates. Results of the age-adjusted analyses are presented in Online Resource 1.
Table 2
Fully adjusted associations of aMED and mMED (including and excluding alcohol) with the risk of esophageal and gastric cancer subtypes in male NLCS participants
aMEDr |
0–3 | 11889 | 46 | 1.00 | 59 | 1.00 | 71 | 1.00 | 190 | 1.00 |
4–5 | 12569 | 25 | 0.56 (0.33–0.94) | 70 | 1.18 (0.81–1.72) | 74 | 1.00 (0.71–1.41) | 155 | 0.79 (0.62–1.01) |
6–8 | 4792 | 5 | 0.35 (0.14–0.89) | 28 | 1.33 (0.81–2.18) | 13 | 0.48 (0.26–0.89) | 45 | 0.65 (0.45–0.94) |
Ptrend | | 0.013 | | 0.248 | | 0.019 | | 0.016 |
Continuous, per 2 pts | 29250 | 76 | 0.57 (0.41–0.80) | 157 | 1.19 (0.96–1.47) | 158 | 0.86 (0.70–1.06) | 390 | 0.82 (0.71–0.95) |
aMEDb |
0–3 | 9233 | 38 | 1.00 | 46 | 1.00 | 57 | 1.00 | 153 | 1.00 |
4–5 | 13035 | 29 | 0.61 (0.36–1.01) | 71 | 1.15 (0.77–1.72) | 71 | 0.91 (0.63–1.31) | 166 | 0.78 (0.61–1.01) |
6–9 | 6983 | 9 | 0.38 (0.18–0.80) | 40 | 1.28 (0.81–2.02) | 30 | 0.74 (0.46–1.18) | 71 | 0.68 (0.49–0.93) |
Ptrend | | 0.004 | | 0.291 | | 0.241 | | 0.011 |
Continuous, per 2 pts | 29250 | 76 | 0.62 (0.47–0.83) | 157 | 1.11 (0.91–1.37) | 158 | 0.90 (0.74–1.10) | 390 | 0.85 (0.74–0.98) |
mMEDr |
0–3 | 10920 | 39 | 1.00 | 61 | 1.00 | 66 | 1.00 | 156 | 1.00 |
4–5 | 13549 | 29 | 0.57 (0.32–0.99) | 68 | 0.93 (0.64–1.36) | 75 | 0.92 (0.64–1.31) | 182 | 0.92 (0.72–1.18) |
6–8 | 4782 | 8 | 0.53 (0.24–1.19) | 28 | 1.07 (0.66–1.73) | 17 | 0.57 (0.33–1.00) | 52 | 0.78 (0.55–1.12) |
Ptrend | | 0.097 | | 0.791 | | 0.047 | | 0.174 |
Continuous, per 2 pts | 29250 | 76 | 0.65 (0.46–0.90) | 157 | 1.04 (0.82–1.31) | 158 | 0.82 (0.66–1.03) | 390 | 0.93 (0.80–1.09) |
mMEDb |
0–3 | 7782 | 25 | 1.00 | 45 | 1.00 | 50 | 1.00 | 109 | 1.00 |
4–5 | 14091 | 39 | 0.87 (0.51–1.51) | 69 | 0.82 (0.54–1.22) | 72 | 0.75 (0.51–1.11) | 199 | 0.95 (0.73–1.25) |
6–9 | 7377 | 12 | 0.52 (0.26–1.07) | 43 | 1.02 (0.65–1.60) | 36 | 0.73 (0.46–1.16) | 82 | 0.78 (0.56–1.08) |
Ptrend | | 0.065 | | 0.725 | | 0.262 | | 0.111 |
Continuous, per 2 pts | 29250 | 76 | 0.78 (0.58–1.04) | 157 | 1.03 (0.82–1.30) | 158 | 0.86 (0.70–1.07) | 390 | 0.93 (0.81–1.08) |
Table 3
Fully adjusted associations of aMED and mMED (including and excluding alcohol) with the risk of esophageal and gastric cancer subtypes in female NLCS participants
aMEDr |
0–3 | 12254 | 21 | 1.00 | 19 | 1.00 | 15 | 1.00 | 92 | 1.00 |
4–5 | 15123 | 27 | 1.18 (0.62–2.25) | 18 | 0.80 (0.42–1.54) | 12 | 0.65 (0.30–1.42) | 71 | 0.70 (0.50–0.98) |
6–8 | 6278 | 9 | 1.13 (0.48–2.66) | 6 | 0.71 (0.27–1.88) | 6 | 0.85 (0.33–2.18) | 33 | 0.85 (0.56–1.31) |
Ptrend | | 0.756 | | 0.480 | | 0.726 | | 0.410 |
Continuous, per 2 pts | 33655 | 57 | 1.09 (0.76–1.57) | 43 | 0.98 (0.67–1.44) | 33 | 0.82 (0.51–1.33) | 196 | 0.83 (0.67–1.01) |
aMEDb |
0–3 | 10734 | 18 | 1.00 | 15 | 1.00 | 14 | 1.00 | 84 | 1.00 |
4–5 | 14697 | 24 | 1.10 (0.56–2.15) | 19 | 1.00 (0.50–1.99) | 11 | 0.54 (0.24–1.24) | 73 | 0.69 (0.49–0.97) |
6–9 | 8224 | 15 | 1.41 (0.65–3.03) | 9 | 0.92 (0.38–2.21) | 8 | 0.75 (0.30–1.91) | 39 | 0.73 (0.49–1.10) |
Ptrend | | 0.373 | | 0.851 | | 0.653 | | 0.153 |
Continuous, per 2 pts | 33655 | 57 | 1.09 (0.77–1.53) | 43 | 1.05 (0.73–1.51) | 33 | 0.80 (0.51–1.27) | 196 | 0.83 (0.69–1.01) |
mMEDr |
0–3 | 11675 | 17 | 1.00 | 17 | 1.00 | 9 | 1.00 | 75 | 1.00 |
4–5 | 16510 | 31 | 1.58 (0.83–3.03) | 19 | 0.75 (0.37–1.54) | 18 | 1.43 (0.63–3.27) | 91 | 0.92 (0.66–1.28) |
6–8 | 5470 | 9 | 1.62 (0.68–3.87) | 7 | 0.89 (0.35–2.26) | 6 | 1.39 (0.49–3.94) | 30 | 0.95 (0.60–1.50) |
Ptrend | | 0.231 | | 0.793 | | 0.538 | | 0.797 |
Continuous, per 2 pts | 33655 | 57 | 1.33 (0.89–2.00) | 43 | 0.89 (0.58–1.36) | 33 | 1.03 (0.66–1.61) | 196 | 0.98 (0.80–1.21) |
mMEDb |
0–3 | 9693 | 15 | 1.00 | 16 | 1.00 | 9 | 1.00 | 63 | 1.00 |
4–5 | 16579 | 30 | 1.34 (0.68–2.66) | 17 | 0.59 (0.29–1.22) | 14 | 0.86 (0.36–2.05) | 98 | 0.95 (0.67–1.33) |
6–9 | 7383 | 12 | 1.35 (0.58–3.10) | 10 | 0.83 (0.36–1.92) | 10 | 1.37 (0.53–3.51) | 35 | 0.80 (0.51–1.26) |
Ptrend | | 0.511 | | 0.793 | | 0.429 | | 0.327 |
Continuous, per 2 pts | 33655 | 57 | 1.28 (0.89–1.83) | 43 | 0.97 (0.65–1.45) | 33 | 0.99 (0.64–1.54) | 196 | 0.97 (0.79–1.18) |
High MD adherence according to aMEDr was associated with significantly reduced risks of ESCC, GCA, and GNCA in men [HRhigh vs. low (95% CI): ESCC = 0.35 (0.14–0.89), GCA = 0.48 (0.26–0.89), and GNCA = 0.65 (0.45–0.94)] with significant tests for trends. In women, associations of aMEDr with GCA and GNCA risk were also inverse, but did not reach statistical significance [HRper two−point increment (95% CI): GCA = 0.82 (0.51–1.33), GNCA = 0.83 (0.67–1.01)]. In contrast to men, aMEDr was not inversely associated with ESCC risk in women. EAC risk was not associated with aMEDr in both sexes. Heterogeneity tests showed that the association of aMEDr with ESCC risk differed significantly between the sexes (pheterogeneity = 0.008). Associations of similar directions were observed when MD adherence was assessed using mMEDr in men, whereas we did not observe associations with any of the subtypes in women using mMEDr.
Overall, vegetable and fruit intakes were strong contributors to the inverse associations observed in male NLCS participants. High nut intake was associated with a significantly reduced GNCA risk (p = 0.008) in men, but did not contribute to the inverse association with ESCC risk. Furthermore, a low intake of red and processed meats contributed considerably to the inverse association with ESCC risk in men. Concerning the non-significant inverse associations with GCA and GNCA risk observed in women, intakes of nuts, whole grains, fish, and the MUFA:SFA ratio particularly contributed. In women, a high fish intake was associated with a significantly reduced GCA risk (p = 0.046) (data not shown).
Based on AIC values, models containing aMEDr performed similarly or better than mMEDr-containing models for all esophageal and gastric cancer subtypes considered. Inclusion of alcohol in aMED resulted in a clearly worse model fit when considering ESCC risk. Similar model performances were observed when aMED variants with and without alcohol were compared for EAC, GCA, and GNCA risks.
In the stratified analyses (Table
4), men and women were combined to increase the statistical power. Associations of aMEDr with all esophageal and gastric cancer subtypes were similar across strata of cigarette smoking status, alcohol consumption, and BMI. A significant interaction between aMEDr and level of education was observed for GNCA risk (
Pheterogeneity = 0.0073) with a significant inverse association only being present in the lowest category. Inverse associations were also most apparent among those in the lowest education category when considering ESCC and GCA, but interaction tests were not significant for these subtypes. Finally, exclusion of the first 2 years of follow-up did not relevantly change the results (data not shown).
Table 4
Fully adjusted associations of aMEDr with the risk of esophageal and gastric cancer subtypes for various subgroups in the NLCS
Overall | 133 | 0.77 (0.61–0.98) | 200 | 1.14 (0.94–1.37) | 191 | 0.86 (0.71–1.04) | 586 | 0.83 (0.73–0.93) |
Cigarette smoking statusc |
Never smoker | 27 | 0.96 (0.54–1.68) | 40 | 0.96 (0.64–1.44) | 31 | 1.07 (0.64–1.80) | 139 | 0.93 (0.74–1.17) |
Former smoker | 41 | 0.73 (0.47–1.12) | 98 | 1.38 (1.04–1.84) | 96 | 0.72 (0.55–0.96) | 240 | 0.77 (0.64–0.93) |
Current smoker | 65 | 0.78 (0.55–1.10) | 62 | 0.98 (0.68–1.39) | 64 | 1.02 (0.73–1.42) | 207 | 0.83 (0.67–1.03) |
Pheterogeneityd | | 0.7359 | | 0.1553 | | 0.3049 | | 0.4399 |
Alcohol consumptione |
0 g/day | 21 | 0.84 (0.44–1.60) | 36 | 1.43 (0.87–2.33) | 27 | 1.11 (0.66–1.85) | 124 | 0.76 (0.57–0.99) |
> 0–<15.0 g/day | 47 | 1.05 (0.71–1.54) | 90 | 0.99 (0.74–1.32) | 97 | 0.83 (0.63–1.09) | 296 | 0.90 (0.76–1.06) |
≥ 15.0 g/day | 65 | 0.60 (0.42–0.84) | 74 | 1.33 (0.97–1.80) | 67 | 0.75 (0.53–1.08) | 166 | 0.73 (0.58–0.92) |
Pheterogeneityd | | 0.1293 | | 0.1793 | | 0.2517 | | 0.3890 |
Body mass indexf |
≥ 18.5–< 25.0 kg/m2 | 80 | 0.85 (0.62–1.15) | 84 | 1.01 (0.75–1.37) | 82 | 0.73 (0.55–0.96) | 314 | 0.83 (0.71–0.98) |
≥ 25.0 kg/m2 | 50 | 0.58 (0.38–0.90) | 115 | 1.25 (0.98–1.61) | 109 | 0.97 (0.74–1.27) | 264 | 0.82 (0.68–0.98) |
Pheterogeneityd | | 0.3690 | | 0.2248 | | 0.1509 | | 0.9198 |
Level of educationg |
Primary school or lower vocational | 66 | 0.54 (0.36–0.80) | 103 | 0.94 (0.70–1.27) | 96 | 0.71 (0.51–0.97) | 356 | 0.72 (0.61–0.85) |
Secondary school or medium vocational | 49 | 1.00 (0.67–1.49) | 63 | 1.28 (0.96–1.72) | 56 | 0.89 (0.65–1.23) | 169 | 0.89 (0.73–1.09) |
Higher vocational or university | 18 | 1.08 (0.68–1.72) | 34 | 1.48 (0.87–2.54) | 39 | 1.30 (0.86–1.97) | 61 | 1.35 (0.93–1.95) |
Pheterogeneityd | | 0.1135 | | 0.2762 | | 0.0842 | | 0.0073 |
Discussion
In the NLCS, higher aMEDr values were associated with significantly reduced risks of ESCC, GCA, and GNCA in men. In women, we observed non-significant inverse associations between aMEDr and risks of GCA and GNCA, but not ESCC. MD adherence was not associated with EAC risk. Associations of aMEDr with ESCC risk significantly differed between the sexes. Compared to mMEDr-containing models, aMEDr-containing models had similar or better performances. Model performances were generally comparable for aMED variants with and without alcohol, except for ESCC, where models containing aMED (including alcohol) clearly fitted worse.
Results of previously conducted prospective cohort studies [
10,
11] investigating the association of
a priori defined MD adherence with the risk of esophageal and/or gastric cancer subtypes were partially in accordance with our observations in the NLCS. In the US National Institutes of Health-American Association of Retired Persons Diet and Health study, high MD adherence (aMED) was associated with a significantly reduced risk of ESCC, but not EAC, GCA, and GNCA [
11]. Although not statistically significant, the association with GNCA was also inverse. In the European Prospective Investigation into Cancer and Nutrition, inverse associations with MD adherence (relative Mediterranean diet score including alcohol, rMED) were suggested for total GC, GCA, and GNCA, but only reached statistical significance for total GC and GCA [
10]. In accordance with the cohort evidence, an Italian case–control study also observed a significant inverse association between
a priori defined MD adherence and ESCC risk [
9]. Case–control studies focusing on gastric cancer risk found higher MD adherence (
a priori defined) to be associated with significantly reduced risks of GCA, GNCA, and total GC [
12,
13]. Finally, adherence to an
a posteriori defined MD pattern was inversely associated with total GC risk in a Spanish case–control study [
33]. Subtype-specific analyses showed that the inverse association was only statistically significant for GNCA. Combining male and female participants in the present study, HRs (95% CI) per two-point increment in aMEDr were 0.77 (0.61–0.98) for ESCC, 1.14 (0.94–1.37) for EAC, 0.86 (0.71–1.04) for GCA, and 0.83 (0.73–0.93) for GNCA.
For ESCC, associations with aMEDr significantly differed between male and female NLCS participants. Prior studies by Bosetti et al. [
9] and Li et al. [
11] did not observe a significant interaction between sex and MD adherence for ESCC risk. However, the inverse association in the latter study also seemed to be restricted to men [
11]. Residual confounding by smoking behavior could potentially have caused the inverse association between aMEDr and ESCC risk that we observed in men. Tobacco smoking is strongly associated with an increased risk of ESCC [
2,
34]. In our study, subjects in the highest aMEDr category were less likely to be current smokers. Since male participants were more likely to smoke than female participants, the effect of residual confounding by smoking behavior would be larger in men. This could potentially explain why we only observed an inverse association between aMEDr and ESCC risk in men. However, additional subgroup analyses for smoking status restricted to men showed that the inverse association between aMEDr and ESCC risk was strongest in men who had never smoked (data not shown), making it less likely that the observed differences between men and women in the NLCS were solely due to residual confounding by smoking behavior. Therefore, potential male–female differences in the association of MD adherence with ESCC risk deserve attention in future studies. For EAC, GCA, GNCA, and total GC, there was no evidence of heterogeneity between the sexes in neither the present study nor the literature [
10‐
13].
In the present study, inverse associations between aMEDr and ESCC, GCA, and GNCA risk were most pronounced in subjects in the lowest education category with a significant interaction being observed for GNCA. Similarly, Li et al. [
11] reported that aMED was only significantly inversely associated with EAC risk in the lowest education category (
pinteraction = 0.02). However, there was no evidence of effect modification by educational level in the study by Praud et al. [
12]. Although the interaction between aMEDr and educational level that we observed might be a chance finding due to the large number of tests performed, it should be investigated in future studies.
In their Third Expert Report, the World Cancer Research Fund/American Institute for Cancer Research suggested that high intakes of vegetables and fruits, and low intakes of processed meat and grilled (broiled) or barbecued (charbroiled) meat and fish might be associated with reduced risks of ESCC, EAC, and/or GC [
26]. This is in correspondence with our observations that high intakes of vegetables, fruits and nuts, and a low intake of red and processed meats (ESCC only) particularly contributed to the inverse associations with ESCC, GCA, and GNCA risks in men in the NLCS. In women, nuts, whole grains, fish, and the MUFA:SFA ratio were important aMEDr components contributing to the non-significant inverse associations observed for GCA and GNCA risk. Inverse associations (not all significant) between esophageal and/or gastric cancer subtypes and intakes of nuts (ESCC and GNCA), vegetables (ESCC and EAC), and fruits (ESCC) were also documented in previous NLCS analyses [
35,
36]. In addition, high intakes of red (non-significant) and processed meats were associated with an increased ESCC risk in men [
37]. Although the above-mentioned aMEDr components were important contributors to the inverse associations that we observed in the present analysis, none of the individual components seemed to be the sole driver. This supports our pattern-based approach, which accounts for synergistic and antagonistic interactions between dietary components, and solves collinearity and confounding issues associated with the evaluation of individual components. Moreover, weak effects of single dietary components may only emerge when combined in dietary patterns [
22,
38,
39].
Different etiologies have been suggested for the subtypes of esophageal and gastric cancer based on differences in risk factors and incidence trends [
2,
3,
34,
40]. This stresses the importance of considering ESCC, EAC, GCA, and GNCA as separate outcomes as we did in this study. ESCC and EAC were clearly differently associated with MD adherence in our analysis. While MD adherence was significantly inversely associated with ESCC risk in men, no association was observed with EAC risk. In contrast to the esophageal cancer subtypes, GCA and GNCA seemed to have roughly comparable associations with MD adherence.
MD adherence might reduce cancer risk by decreasing oxidative stress, reactive oxygen species-induced DNA damage and inflammation [
39,
41]. The MD is rich in antioxidants (e.g., vitamins and polyphenols) from plant foods and olive oil, and has been associated with higher total antioxidant capacity and lower levels of oxidized low-density lipoprotein cholesterol [
26,
39,
41,
42]. Moreover, polyphenols (e.g., flavonoids) may reduce inflammation and MD adherence has been inversely associated with inflammatory biomarker concentrations [
24,
43]. Finally, dietary fiber possibly acts as a nitrite scavenger, counteracting the carcinogenic effects of N-nitroso compounds [
39,
44]. Low meat intake may also contribute to the cancer-protective effects of the MD. Nitrates and nitrites in processed meat can form N-nitroso compounds in the stomach. Besides, heme iron in red meat also stimulates the endogenous formation of N-nitroso compounds and causes oxidative stress and DNA damage. Finally, carcinogenic heterocyclic amines and polycyclic aromatic hydrocarbons are formed during high-temperature cooking of meat [
26].
The use of data from a large prospective cohort with a long duration of follow-up enabled us to perform subtype- and sex-specific analyses. However, case numbers for the individual subtypes were low, necessitating us to combine men and women in the stratified and sensitivity analyses to increase the statistical power. Another strength of our study was the availability of high-quality dietary data. The NLCS-FFQ was validated using 9-day dietary records completed over three different seasons, showing an adequate performance [
16]. Furthermore, a reproducibility study demonstrated that the single baseline measurement of this FFQ performed relatively well in ranking subjects according to their nutrient intake for over at least 5 years [
19]. Nonetheless, changes in dietary habits and potential confounders during follow-up might still have attenuated the observed associations. Instead of using self-reporting based methods (e.g., FFQs and dietary records), dietary intake could be assessed by the measurement of biomarker concentrations in blood. However, there are currently no biomarkers available that assess adherence to all the aspects of the MD. Furthermore, biomarker levels in blood are also influenced by, e.g., absorption and excretion rates and reflect only short-term dietary intake.
Despite the fact that we adjusted for a large number of potential confounders, residual confounding by unmeasured factors may still exist. For example, we did not obtain data regarding
Helicobacter pylori infection, which might have confounded our results in particular for GNCA. Furthermore, we cannot exclude the possibility of errors in the outcome measurements. However, it was reported that the histology and topography information from the Netherlands Cancer Registry, which we used to define the tumor subtypes, is of high accuracy [
45]. Reversed causation due to the presence of preclinical disease symptoms in cases is another concern, particularly when investigating relations between dietary factors and gastrointestinal cancers, as we discussed previously [
46]. Prospective cohort studies are less sensitive to this type of bias than case–control designs and we obtained similar results when excluding the first 2 years of follow-up.
Data on the reliability of
a priori scores used in the assessment of MD adherence are limited [
47]. The reliability of ten indexes measuring MD adherence, including mMED and rMED, has been evaluated by assessing correlations with a hidden common factor (obtained by factor analysis) understood as “MD adherence” [
48]. Both mMED and rMED were amongst the four indexes that showed high correlations with the “MD adherence” factor (mMED: 0.83, rMED: 0.80) [
48]. Regarding the validity of the MD scores used, subjects with higher MD score values in our study consumed, as expected, more plant foods (e.g., vegetables, legumes, fruits, and nuts) and less foods from animal origin (e.g., meat and dairy products). A similar pattern was previously observed for tMED [
22]. However, the validity of tMED and its variants has mainly been established by showing inverse associations with various adverse health outcomes including all-cause mortality and risks of and mortality from cardiovascular diseases and cancer (e.g., [
22,
23,
25,
49,
50]). Several studies have compared associations between various index-based dietary patterns, including tMED (variants), and cancer risk. A review published in 2018 showed that 3 out of 4 studies investigating the association of tMED (variants) with postmenopausal estrogen receptor-negative breast cancer risk observed a significant inverse relation. However, associations were inconsistent for other dietary pattern scores [e.g., Dietary Inflammatory Index (DII), Healthy Eating Index (HEI) and Alternate Healthy Eating Index (AHEI)] [
51]. Considering prostate cancer risk, tMED (variants) and DII showed relatively consistent associations across studies, whereas the evidence was inconsistent and/or insufficient for the other indexes [
52]. Furthermore, healthier diets according to tMED (variants), HEI/AHEI and DII have all been associated with lower risks of colorectal cancer [
53]. More studies evaluating associations of various index-based dietary patterns with cancer risk are required to identify the preferred dietary pattern(s) in the perspective of cancer prevention. A final limitation of our study was the population-dependent assignment of scores in the assessment of MD adherence, which we have elaborately discussed previously [
46].
In conclusion, high MD adherence was associated with reduced risks of ESCC, GCA, and GNCA in the NLCS. However, the inverse association with ESCC risk seemed to be restricted to men. So far, results for esophageal cancer generally were consistent with high MD adherence being associated with a reduced risk of ESCC, but not EAC. Findings concerning GCA and GNCA were more diverse, but, generally, inverse associations (not always significant) were observed for at least one of the subtypes. The potential differences in associations between men and women, particularly for ESCC, require further attention.