Since the early report of Segi et al. (1954), starch intake has been associated with an increased risk of gastric cancer in several studies (Graham et al.,1972; Wynder et al.,1963; Trichopoulos et al.,1985; La Vecchia et al.,1987). Furthermore, La Vecchia et al., (1987) reported increased risks of gastric cancer associated with ingestion of traditional starchy foods, i.e., polenta and maize. Thus, diets high in starch possibly increase the risk of stomach cancer, whereas high fiber intake may decrease the risk of this malignancy (World Cancer Fund, 1997). Sugars had been associated directly with gastric cancer risk (Kaaks et al.,1998) after controlling for dietary fiber intake. Several starchy foods, such as white bread and tubers, are consumed frequently in Uruguay. Therefore, we decided to carry out a case-control study on the relationships between carbohydrate intake and stomach cancer in Uruguay.
In the period April 1997 to June 1998, 96 patients with newly diagnosed and microscopically confirmed stomach cancer were identified in the 4 major Hospitals in Montevideo, Uruguay. Eight (8) patients refused the interview, leading to a final number of 88 cases (62 men and 26 women) that were included in the study (response rate 91.7%). The distribution of tumors in the stomach was as follows: (1) cardia (18.2%), (2) corpus (13.6%), and (3) antrum (68.2%).
In the same time period and in the same Hospitals, 371 patients admitted for non-neoplastic conditions were considered eligible for the study. Twenty (20) refused the interview, leading to a final total of 351 patients (response rate 94.6%). The control series was distributed by diagnostic category as follows: eye disorders (26.2%), abdominal hernia (23.8%), acute appendicitis (13.4%), fractures and injuries (10.5%), diseases of the skin (8.8%), varicose veins (7.9%), hydatid cyst (5.7%) and anemia (2.6%). Controls were frequency matched on age (10 years intervals), sex, residence (Montevideo, other counties) and urban/rural status to cases.
Both cases and controls were interviewed shortly after to admittance by 2 trained social workers. The questionnaire included sections on (1) socio-demographic characteristics, (2) a complete tobacco history, (3) a complete alcohol consumption history, (4) an occupational history, (5) anthropometric variables, (6) a complete reproductive section for women and (7) a food-frequency questionnaire including queries on 64 food items.
This FFQ was not validated but it has been previously tested for reproducibility (Comisión Honoraria de Lucha contra el Cáncer, 1996) and the Pearson correlation coefficients for food items ranged from 0.56 for barbecued meat to 0.66 for raw vegetables. For each food, a commonly used unit or portion size was specified, and participants were asked how often, on average over the previous 2 years or in the period previous to the onset of symptoms, they consumed that amount of each food. The responses were open-ended allowing to treat each food as a continuous variable. The responses were coverted to times per year, multiplying them by the corresponding units. Food-group consumption was calculated by summing intake of all individual foods in each group. The following food groups were created for this study (1) starchy foods: rice, pasta, polenta, white bread, potatoes and sweet potatoes; (2) sugar-rich foods: “dulce de leche,” rice pudding, marmalade, custard and cakes. The major sucrose-rich foods in the Uruguayan population of this low socioeconomic strata were “dulce de leche,” rice pudding and marmalade. The first 2 desserts are folk foods prepared adding a substantial amount of refined sugar. The consumption of these desserts represents over 80% of the refined sugar in the diet.
Macro- and micro-nutrients were calculated using local tables of chemical composition of foods (Mazzei et al.,1995). Nutrient intake was computed by multiplying the frequency of intake of each unit of food by the nutrient content of the standard average portion. All nutrients were energy-adjusted using the residuals method (Willett and Stampfer, 1986).
Food items and groups were categorized according to approximated tertiles of consumption, following the controls distribution. Similarly, nutrients were categorized in tertiles. Relative risks of stomach cancer, approximated by the odds ratio (OR), were calculated by multiple unconditional logistic regression (Breslow and Day, 1980). Tests for linear trend were calculated entering to the full model categorical variables as continuous terms with one degree of freedom. Since OR's for men and women were homogeneous, the risks were calculated for both sexes together. All models included terms for matching variables (age, sex, residence and urban/rural status) and study variables (starchy foods, sugar-rich foods, starch, dietary fiber and sugars). Potential confounders were included in the multivariate models. Several important variables, i.e., sodium, protein and fat, were not confounders when statistical criteria were applied.
Cases and controls were very similar regarding age, sex, residence, urban/rural status and monthly income, whereas cases were less educated than controls. Cases showed a significantly greater intake of meat and a lower consumption of vegetables and fruits than controls.
Mean intake of calories, protein, total fat, total carbohydrate, starch and lactose was significantly greater in cases when compared with controls. On the other hand, cases consumed less glucose and fructose than controls. No differences were observed for dietary fiber and total non-starch polysaccharides.
Odds ratios of stomach cancer for starchy foods, desserts and related nutrients are shown in Table I. Among individual food items, white bread and potato intakes were directly associated with stomach cancer risk (OR of white bread 3.3, 95% CI 1.7–6.2). No association was observed for rice, polenta, pasta and sweet potato intakes. All food groups (cereals, tubers and all starchy foods) were significantly associated with gastric cancer risk after controlling for the matching variables, total energy, dietary fiber intake, and sugar-rich foods (OR for all starchy foods 4.7, 95% CI 2.0–10.8). Finally, the food group corresponding to all desserts was associated with an increased risk of gastric cancer (OR 2.4, 95% CI 1.2–5.1), after controlling for the matching variables, total energy, starch and dietary fiber intake.
|Food/nutrient||Tertiles||p-value for trend|
|OR (95% CI)||1.0||0.91 (0.42–1.98)||3.67 (1.67–8.05)||<.001|
|OR (95% CI)||1.0||1.09 (0.50–2.33)||3.28 (1.73–6.19)||<.001|
|OR (95% CI)||1.0||2.10 (1.04–4.24)||2.24 (1.09–4.58)||0.04|
|OR (95% CI)||1.0||2.16 (0.98–4.75)||2.76 (1.43–5.33)||.002|
|All starchy foods1||Cases/controls||17/129||19/127||52/95|
|OR (95% CI)||1.0||1.24 (0.57–2.71)||4.67 (2.01–10.8)||<.001|
|OR (95% CI)||1.0||2.40 (1.23–4.70)||2.43 (1.17–5.07)||0.01|
|OR (95% CI)||1.0||0.84 (0.45–1.56)||1.21 (0.66–2.24)||0.55|
|OR (95% CI)||1.0||1.49 (0.73–3.02)||4.26 (2.11–8.60)||<.001|
|OR (95% CI)||1.0||0.36 (0.19–0.69)||0.34 (0.17–0.65)||<.001|
|OR (95% CI)||1.0||0.41 (0.22–0.77)||0.27 (0.14–0.52)||<.001|
|OR (95% CI)||1.0||0.61 (0.33–1.10)||0.54 (0.30–1.02)||0.05|
|OR (95% CI)||1.0||0.52 (0.28–0.95)||0.34 (0.16–0.74)||.004|
|OR (95% CI)||1.0||0.64 (0.35–1.18)||0.40 (0.19–0.86)||0.02|
Total carbohydrate intake was not associated with gastric cancer risk. On the other hand, starch intake showed a strong positive association with stomach cancer (OR 4.3, 95% CI 2.1–8.6 for the highest tertile of intake) after controlling for the matching variables, total energy, sugar-rich foods and dietary fiber intakes. Dietary fiber intake was inversely associated with stomach cancer risk (OR for dietary fiber 0.3, 95% CI 0.2–0.7). When fiber intake was examined by its source, fiber from fruits displayed a strong reduction in risk (OR 0.3, 95% CI 0.1–0.5), but also cereal fiber was associated with a protective effect (OR 0.5, 95% CI 0.3–1.0). Insoluble NSPs and cellulose were more protective than soluble NSPs.
The effect of mono- and disaccharides is also partially shown in Table I. Total sugar intake was associated with a slight and non-significant risk of gastric cancer (OR 1.3, 95% CI 0.7–2.6). Both glucose and fructose intakes showed a strong reduction in risk, after controlling for total energy, starch and fiber (OR for glucose 0.3, 95% CI 0.2–0.7). A moderate increase in gastric cancer risk was associated with high lactose intake (OR 1.7, 95% CI 0.9–3.2). Finally, sucrose intake was associated with a moderate increased risk of stomach cancer (OR 1.8, 95% CI 0.9–3.8).
Joint effects of starch and sugar-rich foods intakes are shown in Table II. As is shown in marginals, both starch and sugar-rich foods displayed independent effects. On the other hand, the effect of starch was maximal at high intake of sugar-rich foods, reaching an OR of 12.0 (95% CI 2.7–52.3). Dessert intake was more evident at high levels of starch consumption. In the same table, joint effects of starch and dietary fiber intakes are shown. Dietary fiber was associated with a strong reduction in risk at low intake of starch. On the other hand, starch showed its greater effect at high levels of dietary fiber intake. Interactions for both tables were non-significant.
|Sugar-rich foods (tertiles)|
|I (low)||II||III (high)|
|I (low)||1.0||2.9 (0.6–13.5)||1.2 (0.3–4.8)||1.02|
|II||0.8 (0.2–4.2)||0.8 (0.6–13.5)||1.2 (0.3–4.8)||1.5 (0.7–3.0)|
|III (high)||2.8 (0.7–13.5)||5.2 (1.3–20.4)||12.0 (2.7–52)||4.2 (2.1–8.6)|
|1.02||2.4 (1.2–4.7)||2.4 (1.2–5.0)|
|Dietary fiber (tertiles)|
|I (high)||II (medium)||III (low)|
|I (low)||1.0||0.9 (0.2–4.5)||4.3 (1.1–17.6)||1.0 (b)|
|II||1.4 (0.3–6.5)||1.1 (0.2–5.5)||7.4 (1.7–31.3)||1.4 (0.7–2.9)|
|III (high)||5.5 (1.3–23.3)||8.1 (1.9–33.9)||7.8 (1.8–34.5)||4.0 (2.0–8.2)|
|1.02||1.1 (0.6–2.1)||3.0 (1.5–5.8)|
The lack of association between total carbohydrate intake and gastric cancer risk concealed opposite effects for starch and dietary fiber. These results are in line with the recommendations of the World Cancer Fund (1997), which strongly suggested separate analysis for starch, fiber and sugars when dealing with carbohydrate effect.
Starch intake has been the subject of several previous studies. Some of these have examined starchy foods, rather than starch as such (Haenszel et al.,1972; Modan et al.,1974; Correa et al.,1985; Risch et al.,1985; Trichopoulos et al.,1985; La Vecchia et al.,1987; Kono et al.,1988; Buiatti et al.,1989; Kolonel et al.,1981; Tuyns et al.,1992; Hansson et al.,1993; Ramón et al.,1993). Seven of 8 studies have found increases in risk of gastric cancer with high intakes of grains or starchy foods (Graham et al.,1972; Bjelke, 1974; Modan et al.,1974; Risch et al.,1985; Trichopoulos et al.,1985; You et al.,1988; Boeing et al.,1991; Tuyns et al.,1992; Ramón et al.,1993). As happened in our study, white bread was a risk factor in a case-control conducted by Tuyns et al. (1992). Concerning tubers, the evidence is limited. Graham et al. (1972) reported an increased risk of gastric cancer associated with high consumption of potatoes. On the other hand, starch have been estimated as separate from total carbohydrates in 3 studies (Buiatti et al.,1990; Graham et al.,1990; Ramón et al.,1993); in these studies, starch was associated with an moderate increase in risk of stomach cancer. According to La Vecchia et al. (1995), attributable risk of gastric cancer for starchy foods was 39.5%, whereas our estimate is of 37.0%. Mechanisms through which starch could increase gastric cancer risk are mostly unknown (Kono and Hirohata, 1996; World Cancer Fund, 1997). It has been suggested that starch could cause damage to the mucosa and that diet high in starchy-foods could increase nitrosation due to the their low protein content (World Cancer Fund, 1997). On the other hand, since diets high in starch are relatively low in vitamins and antioxidants, starch could be a marker of deficiencies in ascorbic acid and other protective micronutrients.
Our results also show an increase in stomach cancer risk associated with sugar-rich foods and, to a lesser extent, with sucrose intake. Several studies reported positive associations between sugars or sugar-rich foods and gastric cancer (Correa et al.,1985; Trichopoulos et al.,1985; Risch et al.,1985; Nomura et al.,1990; Cornée et al.,1995; Kaaks et al.,1998). Our results also display strong protective associations with glucose and fructose, even after controlling for total vegetable and fruit intake (the major sources of these nutrients). This result is, to our knowledge, a new finding. Mechanisms by which sugar-rich foods increase or decrease gastric cancer risk are unknown. Kaaks et al. (1998) suggested an hyperglycemic state eliciting high levels of insulin which might promote tumor development.
In our study, dietary fiber intake was associated with a strong reduction in risk. Six case-control studies have reported on fiber intake. Three studies reported a reduction in risk of 60–70% (Risch et al.,1985; González et al.,1994; Kaaks et al.,1998), whereas the 3 remaining ones found no association with gastric cancer risk (Buiatti et al.,1990; Ramón et al.,1993; Hansson et al.,1994). We further examined the effect of fiber by its source of origin; fiber from fruits was associated with the strongest reduction in risk. The mechanisms through which dietary fiber may protect gastric mucosa are unknown. It is possible that dietary fiber could be a marker of vegetable and fruit intake, but further studies both in epidemiological and experimental settings are necessary.
Our study has a number of limitations. The small number of cases is against its power. As for other case-control studies, there are also problems related with recall of the diet. Since both cases and controls were hospitalized, a similar recall could be expected. There is no reason to suspect a differential misclassification; thus, the results could be biased to the null. The similar socio-demographic characteristics of cases and controls are against selection bias. Another limitation is related with the lack of information on Helicobacter pylori infection. This infection is not related with sex and no consistent association with any particular diet has been reported (Hansson et al.,1993). Nevertheless, this limitation precludes the study of interactions between H. pylori and diet. We also lacked information on histologic classification of tumors in intestinal and diffuse. This precluded any analysis stratified on histology as has been done by others (Buiatti et al.,1991). The study has a number of strengths. The high participation rate of both series is also against selection bias. The separate study of carbohydrate intake in starch, fiber and sugars is also a strength of our study.
In conclusion, we show that total carbohydrate intake conceals opposite effects of starch (and sugars) and dietary fiber.