点击显示 收起
1 From the Cancer Research Center of Hawaii, University of Hawaii, Honolulu, HI (UN, SPM, LRW, and LNK); the University of Southern California, Los Angeles, CA (BEH); and the German Institute of Human Nutrition Potsdam-Rehbruecke, Nuthetal, Germany (UN)
2 The contents of this report are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute. 3 Supported by grant no. R37 CA054281 from the National Cancer Institute. 4 Address reprint requests to U Nöthlings, Department of Epidemiology, German Institute of Human Nutrition Potsdam-Rehbruecke, Arthur-Scheunert-Allee 114-116, 14558 Nuthetal, Germany. E-mail: ute.noethlings{at}dife.de.
ABSTRACT
Background: Because elevated blood glucose concentrations have been shown to be associated with greater risk of pancreatic cancer, a high dietary glycemic load, which is based on an empirical measure of blood glucose response after food consumption, has been hypothesized as a pancreatic cancer risk factor. However, results so far are scarce and inconsistent.
Objective: We analyzed data for 162 150 participants in the Hawaii-Los Angeles Multiethnic Cohort Study to investigate associations between glycemic load, dietary carbohydrates, sucrose, fructose, total sugars, and added sugars and the risk of pancreatic cancer.
Design: Dietary intake was assessed at baseline by using a quantitative food-frequency questionnaire. During 8 y of follow-up, 434 incident pancreatic cancer cases occurred.
Results: Glycemic load and added sugars were not significantly associated with pancreatic cancer risk. The risk increased with higher intakes of total sugars, fructose, and sucrose, and the association with fructose was significant when the highest and lowest quartiles were compared (relative risk: 1.35; 95% CI: 1.02, 1.80; P for trend = 0.046). A significant association was found with fruit and juices intake (1.37; 1.02, 1.84; P for trend = 0.04) but not with soda intake. Statistical evidence of a significant interaction with body mass index was present only for sucrose intake (P = 0.04). A comparison of the highest and lowest quartiles of sucrose intake in overweight or obese participants gave a relative risk of 1.46 (0.95–2.25; P for trend = 0.04), but the comparison was not significant in normal-weight participants.
Conclusions: High fructose and sucrose intakes may play a role in pancreatic cancer etiology. Conditions such as overweight or obesity in which a degree of insulin resistance may be present may also be important.
Key Words: Glycemic load • pancreatic cancer • added sugars • cohort studies • diet • soft drinks
INTRODUCTION
Pancreatic cancer is the most fatal cancer in adults, with a 5-y survival rate of <5% (1). More than 33 000 new pancreatic cancer cases were expected in the United States in 2006 (2). Because of the poor prognosis and the minimal effect of conventional treatment methods (3), it is important to focus on prevention of this disease. Cigarette smoking is the most important etiologic factor yet identified—current smokers have a pancreatic cancer risk approximately double that of nonsmokers (4). The risk attributable to smoking has been estimated at 25% (5). Obesity and a family history of pancreatic cancer have also been associated with the disease (3, 6-13). Other risk factors include increasing age, male sex, and Native Hawaiian or African American race-ethnicity (14). Dietary factors may also be important. We recently reported a positive association between the intakes of red meat and processed meat and pancreatic cancer in our cohort (15).
Elevated fasting (16) and postload (17) glucose concentrations and diabetes mellitus (18) have been associated with greater risk of pancreatic cancer. The hypothesized mechanism involves insulin (19-21), and this hypothesis has recently been supported by findings linking higher insulin concentrations and insulin resistance to pancreatic cancer risk (22).
Postprandial blood glucose concentrations are influenced by food consumption. Different foods cause different absolute peaks in blood glucose and different rates of change in blood glucose concentrations during the period after consumption (23). An empirical measure of blood glucose response after consumption of a specific food, the glycemic index (GI), has been developed to classify foods according to their postprandial glycemic effects and, hence, according to their effects on blood insulin concentrations (24). Because the amount of carbohydrate in a diet is a major determinant of blood glucose concentrations, the GI of a food item is multiplied by its carbohydrate content to derive the glycemic load (GL) per 100-g intake of the food. The GL of a diet can then be calculated from the amounts and types of foods consumed.
On the basis of the assumption that glucose metabolism plays a role in the development of pancreatic cancer, we hypothesized that a high dietary GL is positively associated with the risk of pancreatic cancer. Both GI and GL have been investigated with respect to pancreatic cancer in 3 prospective studies (25-27), and the results have been inconsistent. Therefore, we analyzed 8-y prospective data from the Multiethnic Cohort Study to investigate associations between dietary GL and pancreatic cancer risk. Furthermore, we examined various carbohydrates and sugars, especially added sugars, to fully investigate the associations.
SUBJECTS AND METHODS
Study design and population
The Multiethnic Cohort Study in Hawaii and Los Angeles was established to investigate lifestyle exposures, especially diet, in relation to cancer outcomes. The design of the Multiethnic Cohort Study was detailed elsewhere (28). In brief, the cohort is composed of >215 000 men and women who were 45–75 y old at cohort creation and who were enrolled in the study between 1993 and 1996. African Americans, Japanese Americans, Latinos, Native Hawaiians, and whites were the 5 targeted racial-ethnic groups. All study participants initially completed a self-administered questionnaire including a detailed dietary assessment and sections on body weight and height, physical activity, smoking behavior, history of medical conditions, reproductive history, and family history of cancer.
We excluded study participants who did not belong to 1 of the 5 targeted racial-ethnic groups, those with implausible diets as described previously (15), those whose body mass index (BMI; in kg/m2) information was missing or implausible (ie, <15 or >50), those with missing information on smoking status or intensity or duration of smoking, and those with prevalent pancreatic cancer at cohort entry. We also excluded participants with self-reported prevalent diabetes mellitus because they may have reduced their consumption of added sugars and carbohydrates. These exclusions left 162 150 participants (72 966 men and 89 184 women) for this analysis.
As explained in the invitation to participate, return of a completed questionnaire indicated a subject's written informed consent. The institutional review boards of the University of Hawaii and the University of Southern California approved the study proposal.
Dietary assessment
The quantitative food-frequency questionnaire (FFQ) was especially designed and validated for use in this multiethnic population (28, 29). In brief, food items for the quantitative FFQ were selected from 3-d measured dietary records in 60 subjects in each racial-ethnic group. The minimum set of food items contributing 85% of the intake of a specific list of nutrients for each racial-ethnic group was chosen and supplemented by the inclusion of food items that were common in the diet of each particular racial-ethnic group. The quantitative FFQ asks about consumption frequencies and portion sizes. In the quantitative FFQ calibration study (29), average correlation coefficients for carbohydrate intake as a percentage of energy intake were 0.57 for men and 0.58 for women between three 24-h recalls and the quantitative FFQ.
The Cancer Research Center of Hawaii food composition table contains both GL and added sugars. GL values are calculated with the use of published GI values (23). Added sugars include all sugars used as ingredients in processed and prepared foods, such as breads, cakes, soft drinks, jam, and ice cream, and sugars eaten separately or added to foods at the table (30). The food composition table includes a large recipe database and many unique foods consumed by a multiracial and multiethnic population. For FFQ items covering >1 food, nutrient profiles of the items were calculated by using a weighted average of the specific foods based on the frequency of use in the 24-h recalls obtained as part of the calibration study (29). Food mixtures were disaggregated into their ingredients by using a customized recipe database before food group intake was calculated. The intakes of total carbohydrates, sucrose, fructose, total sugars, added sugars, and GL and the intakes of nondiet sodas, fruit and juices, and subgroups of citrus fruit and yellow-orange fruit were analyzed in this study.
Case ascertainment
Incident exocrine pancreatic cancer cases (ICD-02 codes C25.0-C25.3 and C25.7-C25.9) were identified by record linkages to the Hawaii Tumor Registry, the Cancer Surveillance Program of Los Angeles County, and the California State Cancer Registry. All 3 registries are members of the National Cancer Institute's Surveillance, Epidemiology and End Results Program. Case ascertainment was complete through 31 December 2002. Linkages to the National Death Index and death certificate files in Hawaii and California provided information on vital status and causes of death.
Statistical analysis
Differences across quartiles of GL were tested with the Cochran-Armitage test for trend for categorical variables and with the t test for slope in linear regression models of mean values on GL for continuous variables. Cox proportional hazards models using age as the time metric were calculated to derive relative risks (RRs). Person-times were determined by beginning with the date of cohort entry, defined as the date of questionnaire completion, and ending at the earliest of the following dates: date of pancreatic cancer diagnosis, date of death, or 31 December 2002, the closure date of the study. Tests based on Schoenfeld residuals showed no evidence that proportional hazards assumptions were violated for any analysis. We present models including both sexes, because there was no evidence of interaction by sex, after adjustment for sex and follow-up time on study from baseline (2 y, 2–5 y, or >5 y) as strata variables, to allow for different baseline hazard rates. A separate analysis suggested some differences between RRs across 3 follow-up time strata, although the differences were not statistically significant (data not shown). Quartiles of nutrient and food intakes were based on the distribution of the variable in the overall cohort. Median values for quartiles by sex and race-ethnicity were used in the respective models to test for trend. Race-ethnicity, age at cohort entry, smoking status, pack-years of smoking, family history of pancreatic cancer, energy intake (logarithmically transformed), intake of red and processed meats, and BMI were used as adjustment factors in the disease risk models. The latter 3 factors have been identified in the Multiethnic Cohort Study as risk factors for pancreatic cancer (15, 31).
To reduce measurement error in the dietary assessments, nutrients and foods were analyzed in terms of densities—ie, by 100 or 1000 kcal/d. Our group (29) and others (32) have shown that correlations with reference measurements were higher for densities than for absolute intake. Given the strong evidence for the hypothesis of a different association by overweight or obesity that can be associated with insulin resistance (26, 33, 34), we also calculated all models separately for normal-weight (BMI < 25) or overweight or obese (BMI 25) participants. The likelihood ratio test was used to determine the significance of the interaction between BMI and main exposure variables with respect to pancreatic cancer. The test compares a main effects, no-interaction model with a fully parameterized model containing all possible interaction terms for the variables of interest.
RESULTS
Participant characteristics are given in Table 1. The percentage of men increased across quartiles of dietary GL. Mean BMI was slightly elevated in the last quartile of GL only. All dietary variables of interest were positively associated with GL.
View this table:
TABLE 1. Characteristics of participants in the Multiethnic Cohort Study by quartile (Q) of dietary glycemic load1
Because all of these dietary variables are likely to be highly associated with each other, we calculated Spearman correlation coefficients between GL and the intakes of total carbohydrates, sucrose, fructose, total sugars, and added sugars. Correlation coefficients were 0.93, 0.40, 0.45, 0.46, and 0.23, respectively, which indicates that the strongest correlate of GL was the intake of total carbohydrates.
During follow-up, 434 incident pancreatic cancer cases occurred in the cohort. GL was not associated with pancreatic cancer risk in the overall cohort (Table 2). However, the RR for fructose was significantly elevated in the highest quartile (P for trend = 0.046). Although the risks for the intakes of total sugars as well as sucrose were highest in the fourth quartile and were suggestive of an association, neither the point estimates nor the test for trend were significant.
View this table:
TABLE 2. Multivariate-adjusted relative risks (RRs) (and 95% CIs) for glycemic load and intakes of carbohydrates in relation to pancreatic cancer in the Multiethnic Cohort Study1
Because obesity can be a determinant of insulin resistance, we stratified our analysis to examine whether the effects of GL or carbohydrate intakes varied by BMI. Higher (but not significantly higher) risks of pancreatic cancer were seen in the overweight and obese group (BMI 25) than in the normal-weight group (BMI < 25) in the top quartiles of intakes of all dietary variables, and there were stronger trends across quartiles (data not shown). However, a significant (P = 0.04) interaction was evident for sucrose only (Table 3). Among overweight and obese participants, the RR (95% CI) for the fourth quartile of sucrose intake compared with the first was 1.46 (0.95, 2.25; P for trend = 0.04); among normal-weight participants, the respective values were 1.07 (0.71, 1.60; P for trend = 0.85).
View this table:
TABLE 3. Multivariate-adjusted relative risks (RRs) (and 95% CIs) for sucrose and pancreatic cancer stratified by BMI in the Multiethnic Cohort Study1
The use of calibration-adjusted nutrient intakes gave similar results (data not shown). Additional adjustment for total physical activity did not alter the findings, although a further stratification by level of physical activity among the overweight or obese participants suggested a higher RR for those with a higher level of physical activity than for those with a low level of physical activity (data not shown).
Fruit and juices combined were the largest contributor to fructose intake in the cohort, followed by nondiet sodas. To confirm our findings for fructose, we analyzed the intakes of several food groups—soda, fruit juices, total fruit, citrus fruit, and yellow-orange fruit—in disease risk models (Table 4). The combined intake of fruit and juices and the intake of fruit alone were associated with a 37–42% increase in the risk of pancreatic cancer in the overall cohort. Among overweight or obese participants, total fruit and juice intake was associated with a 51% higher risk, but weaker associations were seen when fruit and fruit juices were separated (data not shown). When total fruit intake was divided into citrus fruit or yellow-orange fruit, the intake of yellow-orange fruit was positively associated with pancreatic cancer risk among normal-weight participants. However, no statistical evidence of an interaction of any fruit or juice variable with BMI was present. A high intake of regular sodas was not associated with pancreatic cancer risk in the overall cohort or in the subgroup of obese or overweight participants. To determine whether the association was limited to fruit, we also investigated jointly the trends for fructose from fruit, soda, and other sources. All trends were positive and nonsignificant. Therefore, the association with fruit may be due to the fructose content of fruit or another factor.
View this table:
TABLE 4. Multivariate-adjusted relative risks (RRs) (and 95% CIs) for sodas, juices and fruit in relation to pancreatic cancer in the Multiethnic Cohort Study1
Because smoking is the most well-established risk factor for pancreatic cancer, we stratified our analysis according to smoking status. Overall, there was little evidence of effect modification by smoking status (never, former, or current), although the greater risk associated with fruit intake was most apparent among never smokers (data not shown).
DISCUSSION
In the Multiethnic Cohort Study, high sugar intake—specifically, fructose intake—was associated with a greater risk of pancreatic cancer. This association was not reflected in the consumption of sodas, but we did observe a greater risk with a higher intake of fruit and juices. Statistical evidence for an interaction of sucrose intake with BMI was present, and it showed a higher risk of pancreatic cancer in overweight or obese study participants with higher sucrose consumption.
To date, 4 prospective studies have investigated dietary GL and various carbohydrates in relation to pancreatic cancer risk (25-27, 35). Two of these studies found no association for GL, GI, total carbohydrates, total sugar, sucrose, or fructose (25, 27), and one study found a greater risk with higher dietary GL, GI, and fructose intake in sedentary women with a BMI 25, but not in the overall cohort (26). In the present study, fructose intake was associated with the highest risk of pancreatic cancer. A subsequent analysis of high consumption of soft drinks in this same study population showed significantly greater risks in women but not in men (33). The present study also found a positive association between fructose intake and pancreatic cancer risk, but we did not see a greater risk with higher intake of sodas; these results did not differ significantly between men and women. A significant inverse association with the intake of carbohydrates was reported among smokers in Finland in the fourth prospective study (35). A recent prospective study in Sweden reported high consumption of sugar added to coffee, tea, cereals and other foods, and soft drinks to be significantly associated with a greater risk of pancreatic cancer (36), but the study did not confirm an effect modification by BMI or physical activity. Our results supported a modifying effect of BMI on the association between the intake of sucrose and pancreatic cancer risk; the association was positive in overweight and obese participants, but there was no significant association in normal-weight participants. We previously reported that physical activity was not associated with pancreatic cancer in the Multiethnic Cohort Study (31), and stratification on this variable did not support the hypothesis that low physical activity associated with a high BMI will lead to an even larger RR. In fact, the opposite was true—the RRs were greater in overweight or obese participants with higher physical activity. Differences in energy metabolism may explain this observation, and, if that observation is confirmed in other studies, more detailed investigations should be undertaken.
The finding of a greater risk with higher fruit and juices intake was surprising, and it merits some discussion. Fruit, most often in combination with vegetables, generally is thought to have beneficial effects in terms of cancer prevention at various sites, including the pancreas (37). To date, 7 prospective studies have reported on fruit intake and pancreatic cancer risk, and none detected a significant association, either positive or inverse (10, 35, 38-42). A recent study from Sweden included 135 pancreatic cancer cases in a cohort of 81 922 men and women (38). When the highest and lowest quartiles were compared, overall fruit intake was not significantly associated with pancreatic cancer risk (RR: 1.10; 95% CI: 0.64, 1.88), nor was citrus fruit intake. Participants in the highest quartile consumed 2.5 servings of fruit/d. A study in Finnish smokers found a nonsignificant RR of 0.85 for the highest consumption of all fruit and berries in an analysis of 163 cases and 26 948 noncases (35). Median consumption for cases and noncases was 100 g/d. Four studies with mortality as the outcome did not find any associations (39-42). One of these studies (39) included the largest number of pancreatic cancer deaths thus far (n = 3751), although the dietary assessment was very limited and included only one question about consumption of citrus fruit or juices. Our study, in contrast to these previously published studies, examined the largest number of incident pancreatic cancer cases by using a comprehensive and detailed dietary assessment, which enabled us to detect statistically significant associations of smaller magnitude. Because the intake of fruit and juices combined is the largest contributor to fructose intake in our cohort, the increase in risk with high fruit intake may be explained by fructose and total sugars, both of which are highly correlated with fruit intake (0.69 and 0.65, respectively). However, models with fructose separated by sources showed similar associations for all sources. An effect of fruit other than through sugars, therefore, cannot be ruled out at this point. Furthermore, although the null association with soda does not seem to support this conjecture, underreporting of the intake of low-nutrient-density beverages such as soda may attenuate the true association (43).
The hypothesized mechanism linking an impaired glucose metabolism or diabetes mellitus to pancreatic cancer involves insulin. Insulin can promote tumor development by inhibiting apoptosis and stimulating cell proliferation, and it has been argued that insulin acts as a promoter for pancreatic carcinogenesis (19-21). Our findings support this hypothesis to some extent, because slightly elevated risks, especially those with sucrose, were seen in overweight or obese participants who may already have an underlying degree of insulin resistance (21). We reported a higher pancreatic cancer risk in obese men but not in obese women in the Multiethnic Cohort Study, and evidence in the literature for a positive association between BMI and pancreatic cancer risk is fairly consistent (31).
As discussed in the literature, the GI is a controversial concept (44). It has been pointed out that the GIs of foods vary by types, by processing or preparation, and by combinations of foods consumed together, which renders the actual determination of a GI value for a specific food difficult. In addition, the glycemic response after ingestion of a food does not necessarily predict the insulin response (44). Most important for our study, it has been debated whether the GI or GL is accurate when using dietary intakes collected with an FFQ, because some questionnaire items may group foods with differing GL values (44). However, the quantitative FFQ used for the Multiethnic Cohort Study is lengthy (almost 200 items) and thus less likely than most to group dissimilar foods.
In our study, we used GL rather than GI because the blood glucose response after ingestion of a food is determined by both its carbohydrate content and its GI, and GL combines the 2. We further explored the relations between variables by calculating correlation coefficients. Indeed, the correlation between GL and carbohydrate intake was very high (>0.9), and we found almost identical RRs for pancreatic cancer. We suggest that dietary GL in general may not add important new information about the quality of carbohydrates in the diet of our cohort participants.
Several limitations have to be taken into account in interpreting this analysis. Our population was from Hawaii and California only. However, the study was population-based in design to maximize generalizability to the US population (28). Furthermore, dietary measurement error is certainly present in our data. However, we attempted to minimize the error by analyzing nutrients and foods as densities—ie, per 100 or 1000 kcal/d (29). Finally, GI values have not been determined for many local food products, and values had to be estimated from similar foods.
Because of the rapid fatality of pancreatic cancer, case-control studies have generally relied heavily on proxy interviews. Recall bias, which is already a problem in case-control studies, is of even more concern in the case of proxy interviews. Hence, results from case-control studies of pancreatic cancer must be interpreted with great caution. In contrast, the relatively low incidence rate of pancreatic cancer leads to low numbers of cases in prospective studies unless the observation period is long. In addition to the large sample size, the heterogeneity of participants in the Multiethnic Cohort Study—and therefore the variability in dietary intake—is a strength of the present study, because it facilitates the detection of meaningful associations. Our FFQ was especially designed for the study population, and it was comprehensive, which allowed us to adjust for energy intake.
In conclusion, the present study adds to the evidence that intakes of fructose and sucrose play a role in the development of pancreatic cancer. The stronger association with a high intake of sucrose in those who were overweight or obese implies that such conditions, in which an underlying degree of insulin resistance may be present, are also important. We found some evidence for a greater pancreatic cancer risk with a high intake of fruit and juices but not with a high intake of sodas.
ACKNOWLEDGMENTS
The authors' responsibilities were as follows—UN, SPM, and LNK: study concept and design; LNK, LRW, and BEH: data collection; UN and LW: statistical analysis; UN, SPM, and LNK: interpretation of results; UN: writing of the manuscript; UN, SPM, LRW, LNK, and BEH: critical review of the manuscript; and all authors: review and approval of the final manuscript. None of the authors had a financial or personal conflict of interest.
REFERENCES