The role of vegetable and fruit consumption in the aetiology of squamous cell carcinoma of the oesophagus: A case-control study in Uruguay

Authors


Abstract

There is strong evidence of a protection of high consumption of plant foods against squamous cell carcinoma of the oesophagus, but the respective role of fruits and vegetables is not well elucidated. In the time period 1996–2003, a case-control study designed in order to address this question was conducted in Uruguay. The study included 200 cases and 400 age- and sex-matched controls. High total vegetable and total fruit consumption was the strongest protective food group for this malignancy (OR 0.31, 95% CI 0.18–0.51). When this food group was dissected in total vegetables and total fruits, the latter item was more protective than the former one (OR for total fruits 0.48, 95% CI 0.35–0.66) (p-value for difference between both variables 0.04). Finally, when total fruits were dissected in citrus fruits and other fruits, citrus fruits were strongly associated with a reduction in risk (OR 0.47, 95% CI 0.34–0.67). The results of our study point towards an important role of fruits in the prevention of squamous cell carcinoma of the oesophagus. © 2005 Wiley-Liss, Inc.

The incidence of squamous cell carcinoma of the oesophagus is high in Uruguay, and there is evidence that it has declined by about 6% in the time period 1991–1997.1, 2 This trend has been observed also in another high-risk population in Shanghai, China.3 According to this Chinese registry, oesophageal cancer rates decreased more than 50% in a 18-years study period. Also mortality rates have been decreased in Uruguay, being the decline stronger in females compared to males.4

The declining incidence and mortality of oesophageal cancer observed in Uruguay could be due to a decrease in the prevalence of risk factors, to the increase in the prevalence of protective factors, or both. These may include a decreased consumption of salted meat and an increased intake of fresh vegetables and fruits. Although there are no published figures concerning salted meat consumption, according to our data about 50% of the control patients were consumers of this type of meat in their adolescence; this percentage have significantly declined to a figure of about 24% (Registro Nacional de Cáncer, unpublished data). On the other hand, there is evidence that Uruguayan population has increased its consumption of vegetables and fruits.5

Five Uruguayan studies conducted by our group have examined the associations between foods and nutrients and oesophageal cancer risk.6, 7, 8, 9, 10 According to these studies salted meat is a risk factor, whereas high intake of fresh fruits and vegetables was associated with strong reductions in risk. Studies conducted in other populations also reported a strong protective effect of plant foods.11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45 In spite of this consistency, the results concerning the respective role of vegetables and fruits have been somehow discrepant.6, 7, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45 Whereas most studies reported a higher protective effect of fresh fruits compared to vegetables,13, 14, 19, 22, 23 other reports suggest a higher protection associated with vegetables consumption.12, 15, 24, 25

Although our previous studies reported a stronger reduction in risk for high fruit consumption compared to high vegetable intake, the statistical power of these analyses was low and comparisons were conducted between categorical variables.6, 7, 9, 10 This is an important drawback since the absolute amount of consumption of vegetables and fruits could be different.46 To perform a direct comparison it is necessary to model these foods as continuous terms adding a fixed amount of intake. In order to 1) confirm the protection associated with vegetable and fruit consumption and 2) to disentagle the relative contribution of different types of plant foods in the aetiology of squamous cell oesophageal carcinoma we decided to conduct a new case-control study in Uruguay.

Material and methods

Selection of cases

In the period 1996–2003, 208 patients with newly diagnosed and microspically confirmed oesophageal carcinoma were identified in the 4 major public hospitals located in Montevideo, Uruguay. Since hospitals located outside the capital city of Montevideo lacked facilities for diagnosis and treatment of cancer, almost all patients are admitted in hospitals of Montevideo. Three patients were diagnosed as adenocarcinoma of the oesophagus and 5 patients refused the interview. These 8 patients were excluded from the study, leaving a final total of 200 cases with squamous cell carcinoma of the oesophagus (response rate 97.6%). The case series included 162 males (81%) and 38 females (19%).

Selection of controls

In the same time period and in the same hospitals, 723 patients with diseases not related to tobacco smoking or alcohol drinking and without recent changes in their diet were considered eligible for the study. Twenty patients refused the interview, leaving a final total of 703 potential controls (response rate 97.2%). From this pool, 400 controls were frequency matched to the cases on age (in 10-years intervals) and sex. The controls suffered from the following diseases: abdominal hernia (103 patients, 25.7%), eye disorders (88, 22.0%), diseases of the skin (48, 12.0%), acute appendicitis (37, 9.2%), urinary stones (29, 7.3%), varicose veins (29, 7.3%), injuries (26, 6.5%), hydatid cyst (20, 5.0%) and blood disorders (20, 5.0%).

Questionnaire

All participants were administered with a questionnaire shortly after admittance. The interviews were conducted by 2 trained social workers in the hospitals. No proxy interviews were accepted. The questionnaire included the following sections: sociodemographic characteristics, an occupational history based in the job titles and its duration, a family history of cancer among first-degree relatives, a complete tobacco smoking history (including age at start, age at quit, average number of cigarettes smoked per day, type of tobacco and type of cigarette), a history of alcohol drinking (including number of drinks per day or week for beer, wine and hard liquor), a complete history of maté drinking (a local herbal tea that is usually drunk hot or very hot), menstrual and reproductive events and a food frequency questionnaire (FFQ) on 64 food items. This FFQ allowed the calculation of total energy intake and was tested for reproducibility with good results.

Definition of food groups

All queries concerned the consumption of foods 5 years before the date of interview. The following plant foods were included in the FFQ: carrot, tomato, lettuce, onion, garlic, swiss chard, spinach, potato, sweet potato, beetroot, winter squash, cabbage, cauliflower, zucchini, red pepper, orange, tangerine, apple, pear, grape, peach, plum, banana, fig and fruit cocktail. These items were selected on the basis of the consumption of the Uruguayan population. From these individual items, the following food groups were created: raw vegetables (carrot, tomato, lettuce and onion), cooked vegetables (garlic, swiss chard, spinach, potato, sweet potato, beetroot, winter squash, cabbage, cauliflower, zucchini and red pepper), total vegetables (raw and cooked vegetables), citrus fruits (orange and tangerine), other fruits (apple, pear, grape, peach, plum, banana, fig, fruit cocktail), total fruits (citrus fruits and other fruits) and plant foods (total vegetables and total fruits). Average daily intake of each food item (g/day) was calculated by multiplying the reported daily frequency of consumption by the portion size for a middle age participant. In order to compare the odds ratios for the different food groups, each variable was treated as continuous and odds ratios were calculated for an increase of 25 g/day.

Statistical analysis

Relative risks of squamous cell oesophageal cancer for intake of different groups of plant foods, approximated by the odds ratios, were estimated by multiple unconditional logistic regression.47 Since all food groups were homogeneous for sex, OR's were calculated for both sexes together, after including a term for gender. The following variables were included in all models: age (categorical, 5 strata), sex, county of residence (dichotomized as Montevideo and other counties), urbal/rural status, birthplace (categorical, 3 strata), education (categorical, 3 strata), body mass index (categorical, 4 strata), average number of cigarettes smoked per day (categorical, 5 strata), years since quit (categorical, 5 strata), smoking status (categorical, 3 strata), alcohol drinking (categorical, 5 strata), maté temperature (categorical, 4 strata), total energy intake (continuous) and stewed meat (categorical, 4 strata). Tests for trend were performed after entering categorical variables as ordinal in the same model. Departure from the multiplicative model was determined by assessing the likelihood ratio test statistic. An alpha level of 0.05 was used as the indicator of statistical significance and, accordingly 95% CI s were reported. All p-values were derived from 2-sided statistical tests. All the calculations were done with the STATA programme.48

The distribution of cases and controls by sociodemographics and selected risk factors are shown in Table I. As expected for the frequency matched design, age and sex were identical. There was a higher proportion of dwellers living outside Montevideo, of rural residents and of subjects born in the Northern part of the country among cases than among controls. Cases were significantly less educated than controls (OR 0.6, 95% CI 0.4–0.9). Income (in US dollars) was rather similar in both series of participants. Cases were significantly leaner than controls (OR 0.5, 95% CI 0.3–0.8). The consumption of total energy was slightly higher in cases than in controls. Heavy smoking was associated with an increased risk of 4.8 (95% CI 2.6–9.1). There was no difference between smoking of black tobacco and of blond tobacco cigarettes (OR 1.2, 95% CI 0.7–1.8). On the other hand, smoking of hand-rolled cigarettes displayed an OR of 1.7 (95% CI 1.1–2.7) compared to commercial cigarettes (results not shown). Heavy drinking of alcohol was associated with a 4-fold increase in risk and drinking of very hot maté displayed an OR of 4.0 (95% CI 1.5–10.3) compared to drinking of warm maté.

Table I. Distribution of Controls and Cases for Sociodemographics and Selected Risk Factors
VariableCategoryCasesControlsOR95 % CI
Number%Number%
  • 1

    In militers of ethanol per day.

Age (years)40–49157.5307.5  
 50–594020.08020.0  
 60–696934.513834.5  
 70–796130.512230.5  
 80–89157.5307.5Not applicable
SexMales16281.032481.0  
 Females3819.07619.0Not applicable
ResidenceMontevideo7638.016741.8  
 Other counties12462.023358.2  
Urban/rural statusUrban14974.532180.3  
 Rural5125.57919.7  
BirthplaceMontevideo4723.513533.8  
 South6934.517142.7  
 North8442.09423.5  
Education (years)0–26934.512431.0  
 3–58442.013132.8  
 6+4723.514536.2  
Family history of oesophageal cancerNo19497.039197.8  
 Yes63.092.2  
Body mass index≤ 23.17537.510125.21.0 
 23.2–25.35226.010125.20.70.4–1.1
 25.4–27.33417.09824.60.50.3–0.8
 27.4+3919.510025.00.50.3–0.8
Total calories≤ 1,8513919.510025.01.0 
 1,852–2,2605628.010025.01.40.9–2.3
 2,261–2,6444723.510025.01.20.7–2.0
 2,645+5829.010025.01.50.9–2.4
Tobacco smokingNever smokers4020.014035.01.0 
Ex-smokers (years)20+157.5338.31.60.8–3.2
 10–19136.5328.01.40.7–2.9
 1–92914.5389.52.71.4–4.9
Current smokers (cigarettes/day)1–963.0205.01.10.4–2.8
 10–192110.56115.21.20.6–2.2.
 20–293618.04711.82.71.5–4.7
 30+4020.0297.24.82.6–9.1
Alcohol drinking1Never drinkers6231.019548.81.0 
 1–603718.58521.21.40.8–2.2
 61–1204120.56115.22.11.3–3.5
 121–2403417.0399.82.71.6–4.8
 241+2613.0205.04.12.1–8.0
Maté temperatureWarm115.8257.41.0 
 Hot14475.829586.81.10.5–2.5
 Very hot3518.4205.84.01.5–10.3
Number of patients 200100.0400100.0  

Results

Odds ratios of squamous cell oesophageal carcinoma for intake of plant foods categorised in quartiles are shown in Table II. Total plant food intake was inversely associated with the risk of squamous cell cancer of the oesophagus (OR for high intake 0.27, 95% CI 0.14–0.54, p-value for trend <0.0001). Also total vegetables and total fruits were negatively associated with risk of oesophageal carcinoma, with a stronger effect of the later food group (OR for high consumption of total fruits 0.21, 95% CI 0.10–0.43, p-value for trend <0.0001). The difference between total fruits and total vegetables was of borderline significance (p = 0.08). Cooked vegetables displayed a higher reduction in risk than raw vegetables (OR for cooked vegetables 0.51, 95% CI 0.26–0.99, p-value for trend = 0.02). On the other hand, whereas citrus fruits were strongly protective (OR for high consumption of citrus fruits 0.28, 95% CI 0.15–0.55, p-value for trend = 0.0001), the remaining fruits were inversely associated with risk of oesophageal carcinoma, but the trend was nonsignificant.

Table II. Odds Ratios of Squamous Cell Carcinoma for Total Plant Foods
Food groupsGrams/dayCases/controlsOR95 % CIOR95 % CI
  • 1

    Adjusted for age, sex, residence, urban/rural status, birthplace, education, smoking status, number of cigarettes smoked per day, years since quit, alcohol drinking, matè temperature, stewed meat and total energy intake.

  • 2

    Plus total fruits.

  • 3

    Plus total vegetables.

Total plant foods8.0–271.976/1001.01   
 272.0–384.856/1000.650.38–1.10  
 348.9–511.837/1000.370.21–0.67  
 511.9–1029.731/1000.270.14–0.54Not applicable
  p-value for trend<0.0001  
Total vegetables8.0–166.661/1001.0 1.02 
 166.7–229.652/1000.740.43–1.290.770.44–1.36
 229.7–308.043/1000.550.30–0.980.630.35–1.15
 308.1–752.244/1000.420.23–0.790.520.27–0.99
  p-value for trend0.0040.03
Total fruits0.0–74.785/1001.01.03  
 74.8–145.457/1000.720.43–1.200.760.45–1.27
 145.5–218.540/1000.490.28–0.860.520.29–0.91
 218.6–663.118/1000.190.09–0.380.210.10–0.43
  p-value for trend<0.0001<0.0001
Raw vegetables0.0–6.670/1011.01.02  
 6.7–17.751/920.960.57–1.631.030.61–1.77
 17.8–35.239/1070.540.31–0.960.600.34–1.06
 35.3–159.040/1000.610.34–1.070.830.45–1.51
  p-value for trend0.020.16
Cooked vegetables4.5–138.255/1001.0 1.02 
 138.3–201.752/1000.800.45–1.420.780.44–1.41
 201.8–275.347/1000.640.35–1.160.680.37–1.25
 275.4–716.546/1000.460.24–0.890.510.26–0.99
  p-value for trend0.010.04
Citrus fruits0.0–4.069/1001.0 1.03 
 4.1–25.463/1000.810.48–1.370.820.49–1.39
 25.5–60.845/1030.560.33–0.970.590.34–1.01
 60.9–319.223/970.270.14–0.510.280.15–0.55
  p-value for trend<0.00010.0001
Other fruits0.0–43.061/1001.0 1.03 
 43.1–88.065/1000.990.59–1.680.980.58–1.65
 88.1–170.348/1000.760.43–1.320.780.45–1.37
 170.4–379.528/1000.530.28–1.010.570.30–1.09
  p-value for trend0.04 0.08

Joint effects of total vegetables and total fruits are shown in Table III. Both terms were categorised in quartiles according the controls distribution. Both total vegetables and total fruits displayed independent effects (see marginals). The combination of high consumption of vegetables and fruits was associated with a strong reduction in risk of 0.09 (95% CI 0.03–0.3) and the results followed a multiplicative model.

Table III. Odds Ratios of Squamous Cell Carcinoma of the Oesophagus for the Interaction Between Total Vegetables and Total Fruits1
Total vegetablesTotal fruits
≤ 74.774.8–145.4145.5–218.5218.6+Fruit adjusted
OR95 % CIOR95 % CIOR95 % CIOR95 % CIOR95 % CI
  • 1

    Adjusted for age, sex, residence, urban/rural status, birthplace, education, smoking status, number of cigarettes smoked per day, years since quit, alcohol drinking, maté temperature, stewed meat, total energy and for each other.

  • 2

    Number of cases and controls.

  • 3

    Odds ratios and 95 % confidence intervals.

≤ 166.633/232 13/29 11/22 4/16   
 1.03 0.50.2–1.80.50.2–1.40.30.1–1.11.0 
166.7–229.624/31 13/27 11/24 4/18   
 0.60.3–1.50.40.2–1.20.40.1–1.10.20.05–0.70.80.4–1.3
229.7–308.018/21 15/23 6/24 4/32   
 0.60.2–1.60.60.2–1.50.20.06–0.60.080.02–0.30.60.3–1.1
308.1+10/15 16/21 12/30 6/34   
 0.30.1–1.10.40.1–1.10.20.1–0.70.090.03–0.30.50.3–0.9
Vegetables adjusted1.0 0.70.4–1.30.50.3–0.90.20.1–0.4  

Odds ratios for a fixed amount of 25 g/day of intake are shown in Table IV. The strongest reduction in risk was associated with total plant foods (OR 0.31, 95% CI 0.18–0.51). In Model 2, this term was dissected in total vegetables and total fruits. Whereas the effect of total vegetables was nonsignificant (OR 0.66, 95% CI 0.40–1.08), total fruits were inversely and significantly associated with risk of squamous cell carcinoma of the oesophagus (OR 0.48, 95% CI 0.35–0.66) (p-value for differences between total vegetables and total fruits = 0.04). In Model 3, total vegetables were further dissected in raw vegetables and cooked vegetables; in this model total fruits were retained. Although raw vegetables were more protective than cooked vegetables, neither variable was significant. In Model 4, total vegetables remained fixed, whereas total fruits were dissected in citrus fruits (OR 0.47, 95% CI 0.34–0.67) and other fruits (OR 0.67, 95% CI 0.48–0.92).

Table IV. Odds Ratios of Squamous Cell Carcinoma of the Oesophagus for Plant Foods as Continuous Terms1
ModelOR95 % CIchi-squarep-value trend
  • 1

    Adjusted for age, sex, residence, urban/rural status, birthplace, education, smoking status, years since quit, number of cigarettes smoked per day, alcohol drinking, matè temperature, stewed meat and total energy intake.

1 Total plant foods0.310.18–0.5120.22<0.0001
2 Total vegetables0.660.40–1.082.680.10
 Total fruits0.480.35–0.6620.21<0.0001
3 Raw vegetables0.650.40–1.053.070.08
 Cooked vegetables0.760.47–1.231.240.26
 Total fruits0.500.36–0.6917.53<0.0001
4 Citrus fruits0.470.34–0.6718.06<0.0001
 Other fruits0.670.48–0.925.910.01
 Total vegetables0.680.42–1.122.270.13

Odds ratios of squamous cell carcinoma of the oesophagus for individual plant foods were calculated using continuous variables (Table V). Each food item was entered into the model as a continuous term after adding the fixed amount of 25 g/day. Among vegetables, the following foods displayed significant inverse associations: tomato, lettuce, sweet potato and beetroot (4/15, 26.7%). Among fruits, orange, tangerine, apple, peach and banana were strongly protective (5/10, 50%). The stronger reduction in risk was associated with orange consumption (OR 0.46, 95% CI 0.32–0.67, p-value for trend <0.0001).

Table V. Odds Ratos of Squamous Cell Carcinoma Of The Oesophagus for Individual Plant Foods1
Food itemOR95 % CIp-value for trend
  • 1

    Adjusted for age, sex, residence, urban/rural status, birthplace, education, body mass index, smoking status, number of cigarettes smoked per day, years since quit, alcohol drinking, mate temperature, total energy intake and stewed meat.

Carrot0.350.10–1.280.11
Tomato0.500.28–0.900.02
Lettuce0.260.08–0.820.02
Onion0.250.06–1.110.07
Garlic0.840.40–1.780.66
Swiss chard0.930.46–1.870.84
Spinach0.620.28–1.380.24
Potato0.820.54–1.250.36
Sweet potato0.610.41–0.910.01
Beetroot0.280.08–0.910.03
Cabbage1.880.93–3.810.08
Cauliflower1.860.49–7.090.36
Zucchini0.670.22–2.030.48
Red pepper1.860.49–7.000.36
Winter squash0.750.50–1.130.17
Orange0.460.32–0.67<0.0001
Tangerine0.450.25–0.820.008
Apple0.690.50–0.950.02
Pear0.460.19–1.110.08
Grape0.370.12–1.160.09
Peach0.400.17–0.950.04
Plum1.330.21–8.320.76
Banana0.650.45–0.930.02
Fig1.660.46–5.920.44
Fruit cocktail0.480.17–1.310.15

Discussion

According to the results of our study, the joint consumption of vegetables and fruits was the strongest protective factor for oesophageal carcinogenesis. When total vegetables and total fruits were compared, total fruits were more protective than total vegetables (p-value for difference between high consumption of total fruits vs. total vegetables 0.08). When raw and cooked vegetables were analysed as categorical variables, cooked vegetables were significantly and inversely associated with risk of squamous cell cancer of the oesophagus, whereas raw vegetables were not protective. Although this finding is somehow unexpected, the controls of our study showed a median consumption of raw vegetables of 17 g/day, a very low value indeed. On the other hand, controls displayed a median of 202 g/day of cooked vegetables, that is, a ratio of cooked/raw vegetables of 12:1.This finding suggest that, in spite of the deleterious thermolabile effect, cooked vegetables represent a major part of total vegetables consumed by the Uruguayan population and are responsible for almost the entire effect of vegetables. Finally, citrus fruits appear to convey the highest protective effect when compared to the other plant foods. The 2 analytical approaches (quartiles of intake distribution and increase by a fixed amount) provided consistent results, with the exception of the findings related with different subsets of vegetables.

Thirty-eight observational studies reported information on the relationship between vegetables, fruits and oesophageal carcinoma6, 7, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45 (reviewed in 49–51). Two cohort studies displayed protective effects for green-yellow vegetables,12, 13 whereas the most recent contributions displayed significant inverse associations for fruits.13, 14 The prospective study of Tran et al.14 is particularly interesting due to the large follow-up of the participants and the significant differences between vegetables and fruits. Whereas vegetables were not associated with oesophageal cancer risk, fresh fruits displayed a reduction in risk of 20% (p-value for trend = 0.002). Among case-control studies, Uruguayan reports6, 7, 9, 10 consistently showed reductions in risk both for vegetables and fruits, but the effect of fruits was much more impressive. The results of these studies are reported in Table VI: when we performed a meta-analysis based on a random-effects model,52 the combined OR for high intake of fruits 0.27 (95% CI 0.19–0.38, p-value for heterogeneity = 0.08), and the corresponding OR for high intake of vegetables was 0.60 (95% CI 0.49–0.74, p-value for heterogeneity = 0.50), the 2 ORs being significantly different.

Table VI. Epidemiological Studies of Plant Foods and Risk of Oesophageal Cancer in Uruguay
Author, yearVegetablesFruits
OR95 % CIOR95 % CI
De Stefani et al.,19900.560.3–1.00.330.2–0.5
De Stefani et al.,19990.70.5–0.90.40.3–0.6
De Stefani et al., 20000.640.34–1.200.180.09–0.39
De Stefani et al.,20030.510.28–0.930.190.10–0.38
Present study0.520.27–0.990.210.10–0.43
Meta-analytic estimate0.600.49–0.740.270.19–0.38

Some studies suggested that vegetables were more protective than fruits11, 12, 20, 24, 26, 28, 30, 34, 37, 45 (10/38, 26.3%). In particular, the study of Launoy et al.24 displayed higher reductions in risk for vegetables compared to fruit intake. In fact, vegetables were included in their final model, whereas fruits were not. Other reports displayed lower risks for fruits compared to vegetables6, 7, 9, 10, 15, 16, 17, 18, 19, 23, 25, 33, 41, 43, 45 (15/38, 39.5%). Thus, the following question arises: is the overall body of evidence now sufficient to accept that fruits are the more protective plant foods? We performed a meta-analysis of almost all of available studies (32 studies reporting risk estimates and confidence intervals for comparable categories of intake of fruits and vegetables), but the high heterogeneity of the results for both fruits and vegetables (p < 0.001) precluded a meaningful interpretation of the combined risk estimates. The heterogeneity was not explained by factors such as geographic location or year of publication. We consider that a full exploration of the reasons of heterogeneity among results of non-Uruguayan studies goes beyond the scope of this report.

Also, in our study, we have employed the methodology developed for other authors.46, 53, 54 This approach allows the direct comparison of nested variables (in our study vegetables and fruits). To put the odds ratios on an equal footing of comparison, it is necessary to calculate them using the same fixed amount of consumption, as was emphasised by Kulldorff et al.46 Our results showed that total fruit intake was associated with a significantly higher protective effect than total vegetable consumption (p-value of difference between both subsets 0.04). Moreover, the inclusion of a quadratic term for total plant foods did not modify the results, which is compatible with the notion that both vegetables and fruits display a linear dose-response effect.

The mechanism(s) of the protective effect of plant foods in oesophageal carcinogenesis are not entirely clear. Bosetti et al.29 suggested that citrus fruits, a rich source of vitamin C and beta-cryptoxanthin, could play a central beneficial role in preventing the development of squamous cell oesophageal cancer. In fact, a rather recent study conducted by Block et al.55 demonstrated that a serum vitamin C is the main micronutrient resulting of a diet rich in vegetables and fruits. Since ascorbic acid is also an important nutrient in gastric carcinogenesis, blocking intragastric nitrosation, further sudies on ascorbic acid and oesophageal carcinoma are needed. In fact nitrosamines (endogenous or exogenous) could be determinants of oesophageal carcinogenesis.56 Other anti-oxidants such as lycopene, beta-cryptoxanthin, beta-sitosterol and flavonols may be responsible for the protective effect of plant foods in oesophageal carcinogenesis.9 The strong effect of high intake of citrus fruits found in our study supports a particularly important role of vitamin C.9, 18 Other mechanisms for cancer prevention have been suggested.51 Among them, adsorption of carcinogens and modulation of antioxidants enzyme systems could be of importance.

Residual confounding for tobacco smoking, alcohol drinking and hot beverages (maté drinking) is potential threat in oesophageal carcinoma. For this reason, we included in our model terms like smoking status, number of cigarettes smoked per day, years since quit smoking, alcohol drinking and maté temperature. Thus, residual confounding could be considered minimised.

As other hospital-based case-control studies, our study has some limitations. First, our study could not eliminate the problem of selection bias. We tried to minimise this bias by frequency matching cases and controls on age and sex. The high response rates also minimise the likelihood of this bias. Interviewer bias is unlikely, since the participants were interviewed by 2 trained social workers; the interviewers and the participants were mostly unaware of the protective effect of plant foods. Recall bias is always a central problem. This bias could be accentuated by the increased age of cases and controls. Nevertheless this bias could lead to results closed to the null, although differential misclassification remains as a possibility. Most oesophageal cancer patients have major difficulties in eating and have certainly modified their diet as compared to controls. Although the interview referred to diet in the past, differential recall bias remains a possibility. Moreover, our queries concerned 5 years before the date of the interview. This was done in order to avoid the preclinical stage of the malignancy. In fact, other authors like Terry et al.35 asked about usual frequency of vegetables and fruits 20 years prior to interview. According to Willett,57 the responses of distant past reflect more accurately the current diet. Studies conducted to date suggest that diet may be recalled with acceptable levels up to approximately 10 years.57 In our reproducibility study, total fruits displayed a correlation coefficient of 0.44, whereas total vegetables showed a coefficient of 0.61. On the other hand, our study has strengths, including the high response rate observed both in cases and controls, the ability to control for potential confounders and the absence of proxy interviews.

In summary, our study replicates previous reports showing that total vegetable and fruit consumption is associated with a strong reduction in risk of this malignancy. Also, total fruits were more protective than total vegetables in our population. Finally, among fruits, citrus fruits could play a central role in preventing oesophageal carcinogenesis. Our results suggest that high consumption of plant foods, and in particular citrus fruits, is strongly protective against squamous cell carcinoma of the oesophagus.

Ancillary