Cancer Causes & Control

, Volume 22, Issue 5, pp 725–736

Dietary habits and gastric cancer risk in north-west Iran

Authors

    • Department of Epidemiology and BiostatisticsSchool of Health, Yazd Shahid Sadoughi University of Medical Sciences
    • Nutritional Epidemiology Group, Centre for Epidemiology and BiostatisticsLeeds Institute of Genetics, Health and Therapeutics, University of Leeds
  • David Forman
    • Section of Cancer InformationInternational Agency for Research on Cancer
    • Cancer Epidemiology Group, Centre for Epidemiology and BiostatisticsLeeds Institute of Genetics, Health and Therapeutics, University of Leeds
  • Reza Malekzadeh
    • Digestive Diseases Research CenterTehran University of Medical Sciences
  • Abbas Yazdanbod
    • Ardabil University of Medical Sciences
  • Robert M. West
    • Division of Biostatistics, Centre for Epidemiology and BiostatisticsLeeds Institute of Genetics, Health and Therapeutics, University of Leeds
  • Darren C. Greenwood
    • Division of Biostatistics, Centre for Epidemiology and BiostatisticsLeeds Institute of Genetics, Health and Therapeutics, University of Leeds
  • Jean E. Crabtree
    • Section of Molecular GastroenterologyLeeds Institute of Molecular Medicine, University of Leeds
  • Janet E. Cade
    • Nutritional Epidemiology Group, Centre for Epidemiology and BiostatisticsLeeds Institute of Genetics, Health and Therapeutics, University of Leeds
Original paper

DOI: 10.1007/s10552-011-9744-5

Cite this article as:
Pakseresht, M., Forman, D., Malekzadeh, R. et al. Cancer Causes Control (2011) 22: 725. doi:10.1007/s10552-011-9744-5

Abstract

Objectives

North-west Iran is a high-risk area for gastric cancer (GC). Dietary practices may increase risk of GC. For the first time, the diet–GC association in this area was assessed using a validated food frequency questionnaire.

Methods

Cases and controls were recruited in a population-based study. In addition to collecting dietary data using a food frequency questionnaire, Helicobacter pylori antibody level was measured. Multiple logistic regression models were used to estimate odds ratios for associations between dietary factors and GC among 286 cases and 304 controls.

Results

A positive association was estimated for total fat intake (OR = 1.33/20 g, 95% CI: 1.12–1.57) and risk of GC. Inverse associations were observed for vitamin C, iron, and zinc intake and risk of GC and its subgroups (cardia, non-cardia). Fruits and vegetables consumption and refrigerator use showed inverse associations (OR = 0.72/100 g, 95% CI: 0.65–0.80 and OR = 0.75/10 years, 95% CI: 0.60–0.95, respectively). Positive association was observed among those who preferred fried food (OR = 2.21, 95% CI: 1.45–3.37) or consumed highly salted/roasted seeds (OR = 1.97, 95% CI: 1.13–3.43).

Conclusion

GC in north-west Iran is associated with dietary practices: foods, nutrients and food preparation habits.

Keywords

DietGastric cancerCase–control studyH. pyloriIran

Introduction

Gastric cancer (GC) is the fourth most common type of cancer in the world and the second most common cause of death from cancer [1]. Since the mid-1980s GC incidence and mortality rates have fallen substantially in most high-income countries, but remain high in parts of Asia (China, Japan and Korea), Eastern Europe and South America [1, 2].

A systematic review of the evidence-relating diet to stomach cancer risk found that fruit and certain types of vegetables were probably inversely associated and salt and salted foods probably positively associated with risk. However, there was limited evidence relating to other dietary factors and cooking practices [3]. Problems with measurement of exposures (e.g., qualitative instead of quantitative dietary data), inadequate sample size, and adjusting for different confounding factors may cause the lack of consistency among studies.

People living in the Ardabil Province of north-west Iran are at high risk of GC (age-standardized incidence rate of 51.8 and 24.7 per 100,000 populations per year between 2004 and 2006 for men and women, respectively) [4]. Consumption of some specific foods or snacks (e.g., black Halva, roasted/salted seeds) is common in Ardabil. Selenium deficiency and high prevalence of Helicobacter pylori (H. pylori) infection have also been reported in Ardabil [5, 6]. It is possible that such dietary and bacterial factors contribute to the high risk of GC among the Ardabil population. To date, however, only one published study [7] has assessed the environmental risk factors for GC in this area. These paper reports result from the first comprehensive study investigating diet and GC risk based in this high-risk population.

The primary objective of the Ardabil Case–Control Study (ACCS) reported here was to determine associations between certain nutrients, food groups, food items, and dietary habits and GC among the Ardabil population. As a potential risk factor for GC, H. pylori status was also investigated. All analyses were repeated for tumour subsite location and histological subtypes of GC.

Materials and methods

This population-based case–control study took place between August 2005 and August 2007. Patients with history of GC, without other malignancies, who had resided in Ardabil for over 20 years were recruited from hospitals and private clinics in Ardabil Province after microscopic assessment of gastric biopsy by a pathologist. The WHO/IARC guidelines [8] and the Pekka Lauren’s system [9] were used to classify tumours into cardia/non-cardia and the histologically defined intestinal/diffuse subtypes, respectively.

A stratified cluster random sampling technique was used to recruit controls in this study and was explained in detail elsewhere [10]. Briefly, controls were not individually matched to cases but were selected to be representative of the Ardabil population aged over 40 years living in rural and urban areas. The Ardabil Province Health System database including the household’s demographic information in rural and urban areas was used as a sampling frame. A table of random numbers was used to select 22 clusters from a list of the alphabetically ordered rural and urban health centres distributed around the Ardabil Province. The census track households in each cluster were investigated to find up to 15 subjects (only one control from each household) who met the inclusion and exclusion criteria (age > 40 years, resident of Ardabil Province for at least 20 years and no history of malignancy for cancer sites). This ensured a control group with a wide residential spread across the region.

Compared with dietary record and recall tools, a food frequency questionnaire is able to collect dietary information over a wider time period and from larger number of individuals. A quantitative food frequency questionnaire (QFFQ), already developed for use in another region of Iran [11], was modified to include all locally specific foods except those very low in energy and nutrients, such as condiments and spices. The final questionnaire included 117 food items and was validated specifically for the study population (n = 80) against a 24-h recall, a 4-day food diary and three biomarkers (i.e. serum zinc and iron and plasma vitamin C). Comparing the QFFQ with 24-h recall and the QFFQ with 4-day food diary, the average Spearman’s correlation coefficients (ρ) and weighted kappa indices (κw) for daily estimation of energy, macronutrients and three micronutrients (i.e. serum zinc and iron and plasma vitamin C) intake indicated a moderate agreement between the QFFQ and the reference tools (ρ = 0.38, κw = 0.29 and ρ = 0.41, κw = 0.30, respectively). These correlations were weaker between the QFFQ and three biomarkers (on average, ρ = 0.10, κw = 0.25) [10]. QFFQs were administered by three trained interviewers in subjects’ homes or during their hospital admission and for controls at the local health centre. All participants signed a consent form before starting the interview. Subjects were asked to recall their usual diet over the last year (for cases at 1 year prior to date of diagnosis). Frequency (daily, weekly, monthly and annually) and portion size of consumption were asked for each food item. To improve accuracy of obtained data on portion sizes, food models and photographs were used and one of the subjects’ family members was asked to help them to answer the questions. Amounts of dietary nutrients per day were calculated by multiplying the frequency of consumption of each food by the nutrient content of the indicated portion size. Except for a few local traditional food items, the UK food composition data set [12] was used for estimation of nutrients per 100 g of foods. Demographic, medical history and dietary habits information were summarized in an annex questionnaire.

Serum samples (0.5 ml) were extracted and stored at −70°C prior to measurements. Western blot assay (MP Diagnostics Helico Blot 2.1 kit) was used to detect IgG antibodies to H. pylori and to the specific CagA protein.

The study protocol and questionnaire were reviewed and approved by the ethical board of the Digestive Disease Research Center, Tehran, Iran.

Statistical method

Associations between dietary variables and the outcome were estimated by odds ratios as the appropriate measure of effect size estimation in case–control studies and as a good approximation of the risk ratio when the outcome is a rare disease.

Multiple logistic regression models were developed to estimate the association between the dietary exposures and GC controlling for the confounding effects of age and sex (model A) and age, sex, education, living area, smoking, gastric symptoms, income, refrigerator ownership, duration of using a refrigerator, seeds preparation method, frying,H. pylori infection and total energy intake (model B). Proteins, fats, carbohydrates and alcohol are four nutrients that contribute to calories, and as such energy intake would be an intervening factor in any association they have with the outcome. This situation does not exist for other nutrients even if there is correlation between them and calorie. Thus, in order to avoid overadjustment, energy intake was excluded when the main sources of energy (carbohydrate, fat and protein but not alcohol due to the low consumption in this population) were included in the model and vice versa. Any evidence of multicollinearity between explanatory variables was examined by assessing the extent to which the parameter estimates are impacted (via measurement of variance inflation factor) when they are included in a model together. The variables with variance inflation factors greater than 5.0 were included in the regression models separately to reduce the effect of collinearity.

When it was not clear whether the relationship between continuous variables and the outcome was linear or non-linear, the variable was entered into models alternately in continuous and categorical form and two-side likelihood ratio test (LRT) was carried out to compare the model with the categorical form to that with the continuous form. If the p value for the LRT was <0.05, it was interpreted that the assuming a non-linear relationship was a more realistic and consequently, the categorical form of the variable was included in the logistic regression model. If, however, the p value for LRT was >0.05, the continuous form of the variable was kept in the model.

The models were carried out for total GC risk and separately for risk of the subtypes cardia, non-cardia and diffuse and intestinal. All p values were two-sided and determined significant at <0.05. The statistical analyses were carried out using Stata MP version 10.1 (Stata Corp LP, College Station, TX, USA).

Results

A total of 286 cases and 304 controls were included in the study (high response rate of 95 and 97%, respectively, because of successful model of recruitment). Just over 70% of both the case and control groups were men (Table 1). On average, controls were 3 years younger than the cases. Almost all cases and controls were of Turkish ethnicity, and 59% resided in rural areas. Cases had a lower level of education (p = 0.007) and less income (p = 0.002) than controls. Current smoking was more common (p = 0.65) and with higher intensity (p = 0.06) among cases than controls. A few subjects in both groups reported alcohol consumption, although differences were not substantial.
Table 1

Descriptive characteristics of cases and controls

Characteristics

 

Cases (n = 286)

Controls (n = 304)

p value

Gender

 M/F

n(%)

210 (73.4)/76 (26.6)

217 (71.4)/87 (28.6)

0.58

Age (year)

Mean(SD)

66.3 (11.3)

62.9 (11.1)

<0.001

Education

 Illiterate

n(%)

228 (79.7)

206 (67.7)

 

 Primary school

n(%)

47 (16.4)

75 (24.7)

 

 High school

n(%)

8 (2.8)

19 (6.3)

 

 College

n(%)

3 (1.1)

4 (1.3)

0.007¥

Income

 Low

n(%)

93 (32.5)

69 (22.7)

 

 Medium

n(%)

159 (55.6)

172 (56.6)

 

 High

n(%)

34 (11.9)

63 (20.7)

0.002

Alcohol intake

 No

n(%)

278 (97.2)

297 (97.7)

 

 Yes

n(%)

8 (2.8)

7 (2.3)

0.79¥

Smoking pattern

 Never

n(%)

170 (59.4)

192 (63.2)

 

 Ex-smoking

n(%)

54 (18.9)

52 (17.1)

 

 Every day/occasion

n(%)

62 (21.7)

60 (19.7)

0.65

Smoking intensity (pack/year)

Mean(SD)

13.6 (24.8)

9.2 (21.0)

0.06

History of GIa cancer

 No

n(%)

261(91.3)

287 (94.4)

0.12

 Parent

n(%)

6 (2.1)

8 (2.6)

0.88

 Sibling

n(%)

19 (6.6)

9 (3.0)

0.04

H. pylori

 Seronegative

n(%)

3 (1.2)b

9 (3.0)c

 

 Seropositive

n(%)

247 (98.8)b

289 (97.0)c

0.16¥

CagA

 Seronegative

n(%)

4 (1.6)b

15 (5.0)c

 

 Seropositive

n(%)

246 (98.4)b

283 (95.0)c

0.04¥

 p value for Student’s t-test and Chi-squared test for continuous and categorical variables, respectively

¥ p value for Fisher’s exact test

aGastrointestinal tract

bIn total, 250 cases were evaluated for H. pylori and CagA infection

cIn total, 298 controls were evaluated for H. pylori and CagA infection

A greater proportion of cases than controls reported a history of GI tract cancer among siblings (p = 0.04) but not parents (p = 0.88). The majority (≥97%) of controls and cases had H. pylori infection, and the difference between the two groups was not significant (p = 0.16). In comparison with the control group, however, GC patients were more likely to be CagA IgG seropositive (p = 0.04).

Cases and controls had similar average daily intake of the main food groups with the exception of dairy, fruit and vegetables (F&V), and pickles, which were consumed more frequently by controls than cases (OR = 0.92/100 g, 95% CI: 0.85–0.99, OR = 0.77/100 g, 95% CI: 0.72–0.82 and OR = 0.84/10 g, 95% CI: 0.78–0.91, respectively) (Table 2). Except for fat, controls reported higher daily intake of most nutrients.
Table 2

Average daily intake of nutrients, food groups and food items of interest and frequency of dietary habits among cases and controls compared with relevant unadjusted odds ratio (OR) for gastric cancer

Variables

Cases n = 286

Controls n = 304

Unadjusted OR (95% CI)

Nutrients

Mean(SD)

 

Energy (per 100 kcal)

25.6 (8.8)

26.6 (7.1)

0.99 (0.97–1.01)

Carbohydrate (per 50 g)

7.5 (3.0)

8.1 (2.4)

0.93 (0.87–0.99)

Total fat (per 20 g)

4.3 (2.1)

4.1 (1.4)

1.08 (0.98–1.18)

Protein (per 10 g)

9.0 (3.2)

9.7 (2.7)

0.92 (0.87–0.97)

Vitamin C (per 10 mg)

6.4 (3.7)

9.3 (4.9)

0.85 (0.81–0.89)

Vitamin E (per 10 mg)

1.0 (0.9)

1.1 (0.6)

0.77 (0.59–1.01)

Iron (per 5 mg)

3.0 (1.3)

3.4 (1.1)

0.73 (0.62–0.84)

Zinc (per 5 mg)

2.9 (1.1)

3.3 (1.0)

0.70 (0.60–0.83)

Selenium (per 50 μg)

2.7 (1.3)

2.7 (1.0)

0.96 (0.83–1.12)

Food groups and item

Mean(SD)

 

Bread, cereals, & potatoesa (per 100 g)

5.1 (2.2)

5.2 (1.8)

0.98 (0.90–1.06)

Fat & sugara (per 100 g)

1.5 (1.0)

1.4 (0.9)

1.11 (0.92–1.34)

Dairya (per 100 g)

3.6 (2.1)

4.0 (2.5)

0.92 (0.85–0.99)

Meat & fisha (per 100 g)

2.5 (1.4)

2.7 (1.2)

0.93 (0.82–1.06)

Fruit & vegetablesa (per 100 g)

4.2 (2.8)

6.9 (3.9)

0.77 (0.72–0.82)

Raw leafy vegetables (per 100 g)

0.2 (0.3)

0.4 (0.5)

0.09 (0.04–0.21)

Pickles (per 10 g)

1.0 (2.1)

2.0 (3.0)

0.84 (0.78–0.91)

Salted fish (per 10 g)

0.5 (0.6)

0.6 (0.9)

0.77 (0.61–0.97)

Mixed nuts (per 10 g)

0.9 (1.1)

0.8 (0.8)

1.03 (0.86–1.22)

Black Halva (per 10 g)

0.1 (0.2)

0.1 (0.3)

0.94 (0.36–2.47)

Garlic (per 10 g)

0.1 (0.3)

0.3 (0.4)

0.30 (0.16–0.58)

Onions (per 100 g)

0.2 (0.2)

0.3 (0.3)

0.04 (0.02–0.09)

Salt (per 1 g)

2.1 (1.8)

1.9 (1.8)

1.06 (0.97–1.16)

Dietary habits

Cooking method preference

n(%)

 

 Boiling

72 (25.2)

131 (43.1)

0.82 (0.59–1.13)

 Frying

132 (46.2)

116 (38.2)

2.36 (1.68–3.31)

 Barbecuing

5 (1.7)

6 (1.9)

0.80 (0.41–1.57)

Seeds preparing common method

n(%)

 

 Low riskb

102 (36.0)

156 (51.7)

0.66 (0.44–0.98)

 High riskc

101 (35.7)

66 (21.8)

1.53 (0.99–2.38)

Using refrigerator

275 (96.2)

301 (99.0)

0.25 (0.07–0.91)

 

Mean(SD)

 

Duration of using refrigeratord (per 10 year)

1.7 (1.1)

1.9 (1.0)

0.82 (0.70–0.95)

aMain food groups

bLow salted and roasted in low temperature and for a short time

cHighly salted or roasted in high temperature and for a long time

d275 cases and 301 controls provided information for this variable

In comparison with controls, a smaller proportion of cases owned a working refrigerator (OR = 0.25, 95% CI: 0.07–0.91). Among refrigerator owners, controls had used their refrigerators an average of 2 years longer than cases (OR = 0.82/10 year, 95% CI: 0.70–0.95). Dietary habits such as frying and consumption of highly salted seeds or seeds roasted at high temperature were more common among cases (OR = 2.36, 95% CI: 1.68–3.31 and OR = 1.53, 95% CI: 0.99–2.38, respectively).

Total gastric cancer risk

An inverse dose–response association with GC was observed for protein, vitamin C, iron and zinc in both the limited adjusted (model A, adjusted for age and sex) and fully adjusted (model B, adjusted for age, sex, education, living area, smoking, gastric symptoms, income, owning refrigerator, duration of using refrigerator, seeds preparing method, frying, H. pylori infection and total energy intake) models of regression analysis (Table 3). A positive dose–response association was observed for total fat consumption in both models. Among food items, F&V, raw vegetables, garlic, onions and pickles were inversely associated, while fat & sugar were positively associated with GC in both models (Table 4). Meat & fish intake showed an imprecise positive association with GC only in model B. Consumption of mixed nuts only in model B (Table 4) and consumption of highly salted/roasted seeds and frying foods in models A and B (Table 5) also had clearly positive associations with GC risk. An inverse dose–response relationship with the risk of GC was observed in both models for duration of using a refrigerator.
Table 3

Odds ratio (OR) for gastric cancer (GC) and its anatomical and histological subgroups related to selected nutrients

 

Total GC

Cardia GC

Non-cardia GC

Intestinal GC

Diffuse GC

Model A

Model B

OR (95% CI)a

OR (95% CI)b

OR (95% CI)b

OR (95% CI)b

OR (95% CI)b

OR (95% CI)b

Frequency of cases

284

284

53

197

134

68

Nutrients

 Energy (per 100 kcal)

0.99 (0.97–1.01)

0.99 (0.97–1.02)

0.91 (0.86–0.97)

1.01 (0.98–1.04)

0.98 (0.95–1.02)

0.93 (0.88–0.98)

 Carbohydrate (per 50 g)

0.94 (0.88–1.00)

1.00 (0.88–1.13)c

0.92 (0.71–1.16)c

1.04 (0.91–1.18)c

1.05 (0.89–1.23)c

1.02 (0.82–1.26)c

 Total fat (per 20 g)

1.11 (1.00–1.23)

1.33 (1.12–1.57)c

1.13 (0.78 -1.64)c

1.33 (1.12–1.58)c

1.15 (0.92–1.43)c

1.18 (0.86–1.62)c

 Protein (per 10 g)

0.93 (0.87–0.99)

0.87 (0.76–0.99)c

0.77 (0.59–1.01)c

0.87 (0.76–1.00)c

0.85 (0.71–1.02)c

0.71 (0.55–0.90)c

 Vitamin C (per 10 mg)

0.85 (0.81–0.89)

0.82 (0.76–0.87)

0.73 (0.61–0.86)

0.82 (0.76–0.88)

0.77 (0.69–0.85)

0.77 (0.67–0.87)

 Vitamin E (per 10 mg)

0.83 (0.64–1.08)

0.67 (0.44–1.03)

0.26 (0.08–0.83)

0.67 (0.45–1.10)

0.51 (0.27–0.97)

0.66 (0.31–1.42)

 Iron (per 5 mg)

0.74 (0.64–0.87)

0.37 (0.25–0.56)

0.44 (0.22–0.89)

0.36 (0.23–0.55)

0.41 (0.25–0.67)

0.21 (0.10–0.46)

 Zinc (per 5 mg)

0.72 (0.61–0.86)

0.47 (0.32–0.70)

0.64 (0.33–1.24)

0.46 (0.30–0.70)

0.47 (0.29–0.76)

0.40 (0.21–0.76)

 Selenium (per 50 μg)

1.00 (0.86–1.17)

1.11 (0.80–1.54)

1.48 (0.80–2.74)

1.10 (0.77–1.55)

1.52 (1.00–2.31)

1.00 (0.58–1.76)

aAdjusted for age and sex

bAdjusted for age, sex, education, living area, smoking, gastric symptoms, income, owning refrigerator, duration of using refrigerator, seeds preparing method, frying, H. pylori infection and total energy intake

cNot adjusted for total energy intake. Carbohydrate, fat and protein were included all together in one statistical model

Table 4

Odds ratio (OR) for gastric cancer (GC) and its anatomical and histological subgroups related to selected food groups and food items

 

Total GC

Cardia GC

Non-cardia GC

Intestinal GC

Diffuse GC

Model A

Model B

OR (95% CI)a

OR (95% CI)b

OR (95% CI)b

OR (95% CI)b

OR (95% CI)b

OR (95% CI)b

Frequency of cases

284

284

53

197

134

68

Foods and food groups

 Bread, cereals, & potatoes (per 100 g)

1.07 (0.97–1.18)

1.11 (0.98–1.25)c

1.05 (0.82–1.33)c

1.15 (1.01–1.31)c

1.21 (1.03 -1.41)c

1.08 (0.87–1.34)c

 Fat & sugar (per 100 g)

1.18 (0.98–1.44)

1.47 (1.12–1.93)c

1.27 (0.70–2.30)c

1.45 (1.09–1.92)c

1.42 (0.95–2.12)c

1.12 (0.67–1.88)c

 Dairy (per 100 g)

0.91 (0.85–0.98)

1.01 (0.90–1.13)c

0.92 (0.72–1.18)c

1.03 (0.92–1.16)c

0.94 (0.80–1.12)c

0.95 (0.77–1.18)c

 Meat & fish (per 100 g)

0.95 (0.84–1.09)

1.21 (1.00–1.46)c

1.11 (0.75–1.64)c

1.18 (0.97–1.44)c

1.19 (0.93–1.53)c

0.90 (0.64–1.28)c

 Fruit & vegetables (per 100 g)

0.77 (0.72–0.82)

0.72 (0.65–0.80)c

0.54 (0.42–0.70)c

0.74 (0.67–0.82)c

0.61 (0.52–0.71)c

0.67 (0.55–0.80)c

 Raw vegetables (per 100 g)

0.11 (0.05–0.24)

0.12 (0.04–0.33)

0.01 (0.0007–0.16)

0.15 (0.05–0.40)

0.02 (0.003–0.11)

0.02 (0.002–0.20)

 Garlic (per 10 g)

0.33 (0.17–0.64)

0.28 (0.12–0.63)

0.15 (0.02–1.04)

0.27 (0.11–0.65)

0.18 (0.06–0.59)

0.11 (0.02–0.74)

 Onions (per 100 g)

0.04 (0.02–0.10)

0.07 (0.02–0.20)

0.007 (0.0004–0.12)

0.08 (0.03–0.25)

0.05 (0.01–0.20)

0.01 (0.001–0.12)

 Pickles (per 10 g)

0.85 (0.79–0.93)

0.83 (0.74–0.93)

0.61 (0.43–0.88)

0.86 (0.76–0.96)

0.82 (0.71–0.94)

0.78 (0.61–0.99)

 Salted fish (per 10 g)

0.80 (0.63–1.01)

0.86 (0.63–1.18)

0.75 (0.34–1.64)

0.88 (0.63–1.22)

0.93 (0.61–1.40)

0.89 (0.47–1.69)

 Mixed nuts (per 10 g)

1.11 (0.93–1.33)

1.41 (1.08–1.83)

1.31 (0.81–2.10)

1.38 (1.05–1.82)

1.30 (0.91–1.86)

1.26 (0.79–2.01)

 Salt (per 1 g)

1.06 (0.97–1.16)

1.07 (0.95–1.20)

1.03 (0.78–1.35)

1.06 (0.94–1.21)

1.02 (0.87–1.21)

1.10 (0.91–1.33)

aAdjusted for age and sex

bAdjusted for age, sex, education, living area, smoking, gastric symptoms, income, owning refrigerator, duration of using refrigerator, seeds preparing method, frying, H. pylori infection and total energy intake

cMain food groups (5 first variables in the list) were included all together in one statistical model. The model was not adjusted for total energy intake

Table 5

Odds ratio (OR) for gastric cancer (GC) and its anatomical and histological subgroups related to selected dietary habits and infection with Helicobacter pylori

 

Total GC

Cardia GC

Non-cardia GC

Intestinal GC

Diffuse GC

Model A

Model B

OR (95% CI)a

OR (95% CI)b

OR (95% CI)b

OR (95% CI)b

OR (95% CI)b

OR (95% CI)b

Frequency of cases

284

284

53

197

134

68

Dietary habits

Frying (yes vs. no)

2.49 (1.77–3.52)

2.21 (1.45–3.37)

4.91 (2.19–11.06)

2.17 (1.34–3.50)

3.33 (1.90–5.82)

3.96 (1.83–8.56)

Seeds—usual preparation method (versus none)

 Low riskc

0.67 (0.45–1.00)

0.75 (0.46–1.22)

1.25 (0.52–3.01)

0.62 (0.35–1.08)

0.76 (0.41–1.40)

0.55 (0.24–1.27)

 High riskd

1.83 (1.16–2.89)

1.97 (1.13–3.43)

2.77 (1.00–7.64)

1.70 (0.93–3.11)

1.62 (0.81–3.24)

1.99 (0.83–4.73)

Using refrigeratore (yes vs. no)

0.32 (0.09–1.17)

0.79 (0.20–3.13)g

1.60 (0.13–19.12)g

0.71 (0.17–2.89)g

0.66 (0.14–3.11)g

0.93 (0.15–5.66)g

Duration of using refrigeratore (per 10 years)

0.83 (0.71–0.97)

0.75 (0.60–0.95)

0.68 (0.46–1.02)

0.75 (0.59–0.96)

0.63 (0.47–0.84)

0.85 (0.60–1.20)

H. pylori

H. pylori (positive vs. negative)f

2.31 (0.61–8.75)

1.35 (0.27–6.73)

1.73 (0.10–31.16)

1.39 (0.22–8.90)

0.88 (0.14–5.61)

All positive

CagA (positive vs. negative)f

2.98 (0.97–9.20)

1.75 (0.46–6.66)h

2.47 (0.17–35.13)h

1.72 (0.38–7.84)h

1.05 (0.23–4.91)h

All positive

aAdjusted for age and sex

bAdjusted for age, sex, education, living area, smoking, gastric symptoms, income, owning refrigerator, duration of using refrigerator, seeds preparing method, frying, H. pylori infection, and total energy intake

cLow salted AND roasted in low temperature and for a short time

dHighly salted AND/OR roasted in high temperature and for a long time

e275 cases provided information for this variable

f250 cases were evaluated for H. pylori and CagA infection

gNot adjusted for duration of using refrigerator

hNot adjusted for H. pylori infection

GC subgroups

All variables with precisely estimated inverse associations with total GC risk in regression model B showed similar associations with one or two of the GC anatomical and histological subgroups. Vitamin E showed a clear inverse relationship only with cardia and intestinal type of GC (Table 3). Clearly positive associations were observed between total fat intake (Table 3), the fat & sugar food group and mixed nuts (Table 4) and risk for non-cardia GC but not cardia or either of the histologically defined GC types. Bread, cereals and potatoes showed positive associations only for non-cardia and the intestinal type of GC (Table 4). In addition, it was found that preference for fried food had a positive relationship with the risk of GC at both tumour subsites, particularly cardia (Table 5). An imprecisely estimated positive association was observed between highly salted/roasted seeds intake and risk of cardia GC.

H. pylori infection

Infection with H. pylori increased the risk of GC by an estimated 35% after excluding for the effect of confounders (Table 5, model B). This association was stronger for the CagA-positive serotype. Neither of these H. Pylori infection markers showed clear associations with GC subgroups in multivariable models.

After developing appropriate models, some degree of effect modification was observed between H. pylori infection and F&V consumption (p value for LRT = 0.04) as well as between CagA seropositivity and F&V intake (p value for LRT = 0.02). However, an interpretation of observed ORs was impossible due to a very small number in the H. pylori-negative category (data were not shown). There was no evidence of any effect modification by H. pylori status of the association between salt intake and GC, when formally tested by including a statistical interaction term in the model; however, power for this test was very low.

Discussion

This is the first study to explore the diet and GC association in Ardabil based on data extracted from a culturally appropriate QFFQ. The study indicates that certain nutrients, foods and food preparation habits in Ardabil are statistically associated with the risk of GC. To our knowledge, this is the first study reporting a positive association between consumption of highly roasted/salted seeds and risk of GC.

Our data showed a positive association between fat intake and GC, which was seen in other studies [13, 14]. Dietary fat may modulate carcinogenesis by modifying the responsiveness of hormone receptors of the tumour cells [15] or by accelerating formation of arachidonic acid and subsequently of prostaglandins, which may promote tumour growth [16]. Although participants in ACCS were not asked specifically about the source of dietary fat (animal or vegetable), it is possible that high consumption of animal fat in an animal husbandry area like Ardabil contributed to the observed positive association of fat with GC [13]. Saturated vegetable oils are still very popular in Iran and form a large part (47%) of the daily oil utilized in households in Ardabil [17]. Two studies reported increased risk of GC associated with saturated fat intake in Italy [13] and France [18]. In GC subgroups analysis, only non-cardia GC showed a clear positive link with total fat. One possibility could be related to higher saturated fat intake in non-cardia than cardia GC patients [13, 18, 19]. A positive relationship between fat-rich foods and GC was also found, presumably for similar reasons.

The observed inverse association between protein intake and risk of GC in ACCS is in agreement with some studies [18, 20], but not others [21, 22]. While some researchers [13, 22] differentiated between animal proteins and proteins from plant sources regarding the GC risk, this detailed information was not available in this study.

Only risk of diffuse GC was clearly inversely related to higher consumption of protein. In this relatively low-income rural population, a higher protein intake may indicate a generally healthier lifestyle. Nevertheless, even after adjusting for refrigeration use and other income and education-related factors, the relationship with protein was still negative.

Higher intakes of dietary iron were strongly associated with a reduction in GC risk, as has been seen in other studies [14, 18]. The important role of iron as an enzyme-bound co-factor (in both heme and non-heme forms) in the detoxifying of free radicals and prevention of oxidative damage to DNA has been well described [23]. An inverse association was also observed between iron intake and the GC subgroups in ACCS. Other studies [24, 25], except one [19], estimated imprecise associations in an inconsistent direction.

Zinc has a structural role in superoxide dismutase, an antioxidant enzyme; and also for ZNRD1 (zinc ribbon domain containing 1), a transcription-associated gene, with a significantly suppressive effect on cell proliferation of stomach cancer cells in vitro and in vivo [26]. A strong inverse dose–response association was observed for zinc intake, similar to large reductions in risk observed in a large American cohort study [27]. Our findings also showed a strong inverse association between zinc intake and non-cardia GC and its histological subtypes (intestinal and diffuse). Findings from other studies [19, 24, 25] are not consistent.

The antioxidant activity of vitamin C has been well described [28]. The inverse dose–response association for vitamin C that was observed in the ACCS is stronger than the pooled relative risk (0.85 for every 30 mg increment in vitamin C intake) reported in a comprehensive meta-analysis [3]. A consistent inverse relationship was also found for all GC subgroups and is in agreement with many other studies [24, 29].

As the main source for antioxidants (vitamins and minerals), F&V are very likely to play an important preventive role in tumourigenesis [30]. The magnitude of the observed inverse association between F&V and risk of GC in the ACCS (OR = 0.72, 95% CI: 0.65–0.80) is very similar to the reported pooled odds ratio (0.79) in a comprehensive meta-analysis [3]. An inverse association was also found with F&V for all GC subgroups in this study. Similar to many other studies [29, 31], with the exception of garlic and cardia GC, clear inverse associations were observed for raw leafy vegetables, garlic and onions in all regression models. Induction of enzymatic detoxification systems; [32] inhibition of the bacterial conversion of nitrate to nitrite in the stomach; [33] and antibacterial properties against H. pylori [34] are three suggested mechanisms for allium vegetables like garlic and onions to suppress tumourigenesis in the gastric mucosa.

We observed an inverse association for pickles intake in all of the regression models, which is similar to some studies [35, 36], but not all [37]. Pickles are theoretically a risk factor for GC due to containing high amounts of salt and N-nitroso compounds. Pickles vary considerably from country to country, in terms of the vegetables used, amount of salting, degree of fermentation and acidity; thus, it is unlikely that all pickles would have the same association with GC risk. The mutagenic activities of pickles seem to be closely related to the amount of quercetin present [38].

The lack of association between salt intake and risk of GC in ACCS may be due to the limitations in measurement as only salt added during cooking was recorded.

An OR of 1.97 was estimated for highly roasted and salted seeds consumption. Technically roasting seeds for a long time and at high temperatures could burn both the shell and bean. The salt crystals also appear on the shell when water in the added brine is vaporised. It is very traditional in Iran to use one’s teeth to break the seeds’ shell. This method of consumption lets the salt and burnt parts of seeds be ingested. Therefore, the plausible mechanism of pathogenesis for high-risk seeds could be linked to salt [39] and the toxic ingredients in burnt shell and bean. As a toxic chemical and probable carcinogen [40], acrylamide, has been identified in roasted almonds, roasted sunflower seeds and nuts in general [41]. In Iran, mixed nuts are prepared by salting and roasting. So, the observed positive association may be explained as above but requires further confirmatory studies.

There was a high OR for GC associated with frying reported as the main cooking technique. Frying, particularly pan-frying and at high temperature, is more likely than boiling to produce toxicants and potential mutagens (i.e. heterocyclic amines and acrylamide) [40, 41].

The estimated risk reduction of 25% for every 10 years of using a refrigerator is exactly the same as the pooled relative risk reported in a meta-analysis [3]. It was suggested that access to a refrigerator early in life is more important in relation to GC development [42] due to an increased possibility of fresh food and F&V consumption, and a decreased chance of bacterial contamination and lower consumption of preserved foods. The clearly inverse association for long-term refrigerator use was observed only for the intestinal type of GC, plausibly due to intestinal GC being more dependent on environmental factors than the diffuse type [19].

A Western blot assay was chosen to detect systemic immune responses to H. pylori rather than a H. pylori ELISA because Western blotting assays have greater sensitivity in patients with gastric cancer where H. pylori antibody titres can fall with reduced density of colonization with increasing intestinal mataplasia [43]. In addition, Western blotting allows assessment of immune recognition of CagA.

The prevalence of H. pylori infection and its cagA genotype was reported to be up to 90% in Iran [4446]. The causal relationship between H. pylori and GC is widely believed to have been established by a series of prospective studies [47], and the cagA strain has been recognised to be more virulent [48]. Due to the overall high rate of infection, however, the contrast in exposure was small among cases and controls and may be a reason for the weak association observed in ACCS as reported by another [49]. The other possibility for the non-significant association is the seroreversion phenomenon [47] that is a positive to negative change in serology, which can occur during the period between the time at which H. pylori exerts a carcinogenic effect (several years before a malignant tumour appears) and the time of diagnosis of cancer (recruiting the patient for a case–control study and serum sampling). Gastric mucosal atrophy following H. pylori infection or due to age sometimes causes the bacillus to disappear from the mucosa followed by decrease in antigen and consequently, the antibody titre. It is expected that this phenomenon happens more frequently in GC patients than control subjects.

Individually, fitting the regression models for cardia and non-cardia GC showed similar findings to GC overall. The ORs for cardia GC were stronger than the ORs for non-cardia GC for both H. pylori and CagA seropositivity. A Chinese study [50] has also found this relationship and argued that the null or inverse relationship between H. pylori infection and cardia GC observed in studies from Western countries is related to oesophageal malignant tumours, which are located in the distal oesophagus and very close to the cardia area of stomach and are very common in the West. Given that H. pylori infection was associated with decreased risk of distal oesophageal cancer in some studies [51, 52], categorising a fraction of these cancers as cardia GC by mistake, which is possible technically, can lead to an inverse correlation between H. pylori infection and cardia cancer being reported.

The ACCS recruited a large random sample in a high-risk area for GC, characterized the diet using a comprehensive validated QFFQ and measured H. pylori status. Although a fair to moderate agreement between the QFFQ and the reference tools was observed in general, a good agreement was shown between the QFFQ and 4-day food diary for carbohydrate, vitamin E, vitamin C, iron and zinc intake (data was not shown). Nonetheless, the study has some limitations. There was possibility of misclassification of subjects in this study into an incorrect level or category of exposure. This potential misclassification may occur for cases and controls similarly (non-differential misclassification) or differently (differential misclassification). For dichotomous variables, non-differential misclassification biases the expected odds ratio towards the null value when the misclassification is independent of error in measuring other study variables, but differential misclassification may bias the expected odds ratio either towards or away from the null value. For continuous variables or variables with more than two categories, the expected odds ratio may be biased in either direction by either differential or non-differential misclassification [53]. Because cases may alter their dietary pattern in symptomatic stage of the disease and the recall of intake during 1 year before diagnosis of GC may not reflect the intake relevant to the hypothesis under study, the possibility of differential misclassification for daily nutrients intake or dietary habits (e.g., type of consumed seeds, frying) should be considered in this study. There is also possibility of non-differential misclassification in nutrient intake due to recall bias among both cases and controls. Because we used a very sensitive diagnostic method (Western blot) for finding evidence of previous H. pylori infection among cases, we believe that the chance for differential misclassification of H. pylori status was low. Interviewers were not blind regarding the disease status of study participants; however, they were instructed to follow the study standardized procedures. Three-dimensional food models and photographs were used during interview to support the description of portion sizes; nevertheless, some error of assessment of portions could remain [54]. Using a UK database for nutrient analysis may not give an accurate nutrient estimate at the group level; this might lead to bias in this study from non-differential misclassification because the degree of error would be similar in cases and controls, but the bias may be negligible if measured nutrient levels across foods are roughly proportional to their true values.

Case–control design, while having well-known limitations for etiologic studies of cancer with respect to estimating effects of and controlling for confounding by exposures that can change over time, is an efficient approach for describing relationships of interest in a setting where a more elaborate design is not feasible. It is not possible to rule out the possibility that recall bias affects the results in the case–control studies, and confounding could be producing artefactual associations in both cohort and case–control studies. Moreover, non-differential, independent measurement error, inadequate consideration of exposure timing and inadequate intake range would tend to underestimate the magnitude of associations. As reported in a big meta-analysis on diet-cancer association [3], it is notable that associations between dietary factors and risk of GC extracted from case–control studies are generally greater, more consistent and more statistically significant in comparison with cohort studies.

Innovative data from a high-risk area are valuable to complete the missed parts of the puzzle of diet–GC association. This study presents the first detailed data on dietary exposures in the Ardabil population and their relationship with GC risk.

Acknowledgments

Thanks to Dr Hamid Jafarzadeh and Dr Alireza Sadjadi for facilitating the field work and all Ardabil Aras Clinic staff, particularly Ms Sareh Mohammadi and Mrs Suzan Seifi, for their collaboration. Also thanks to the Nutritional Epidemiology and Cancer Epidemiology groups in University of Leeds for permission to access the literature review data set. Special thanks to the laboratory of Dr Jean E Crabtree which is supported by the Sixth Research Framework Programme of the European Union, Project INCA. This study was not possible without support from the Digestive Diseases Research Center, the Ardabil University of Medical Sciences and the World Bank Nutrition Project with the Ministry of Health and Medical Education, Iran.

Copyright information

© Springer Science+Business Media B.V. 2011