Sweets, sweetened beverages, and risk of pancreatic cancer in a large population-based case–control study

Details of this study have been published previously [10‐23]. Briefly, between 1995 and 1999, men and women with incident adenocarcinoma of the exocrine pancreas were identified in six counties of the San Francisco Bay Area (in-area cases) using rapid case ascertainment operated by the Northern California Cancer Center. Pancreatic cancer diagnoses were confirmed by contacting the participants’ physicians and using Surveillance, Epidemiology, and End Results (SEER) abstracts. Eligible in-area cases were 21–85 years of age, residents of one of the six counties, alive upon first contact, and could complete an in-person interview in English. Sixty-five eligible out-of-area cases also were interviewed. They were seen in the University of California San Francisco clinics and met the same criteria as in-area cases except that most were residents of counties adjacent to the six Bay Area counties. Of 798 eligible cases, 532 (67%) completed an interview for this study and 8% refused to participate. The main reason the remaining eligible cases did not participate was because they died prior to contact. Control participants were frequency-matched to cases by sex and age within five-year categories and were selected from the target population using random digit dial. Controls older than 65 years were supplemented by random selection from Health Care Finance Administration lists (now the Centers for Medicare & Medicaid Services). Of 2,525 eligible controls, 1,701 (67%) completed an interview for this study. No proxy interviews were conducted. This study was reviewed and approved by the University of California San Francisco Institutional Review Board and written informed consent was obtained from all the participants prior to interview.

Participants reported, via in-person interviews, their frequency of intake of individual food items that were listed on a 131-item food questionnaire for the time period of one year before their cancer diagnosis for cases, or interview for controls. Intake of seasonal food was averaged throughout the entire year. This semi-quantitative food-frequency questionnaire was obtained from the Harvard University, has been validated in a variety of populations [24‐26], and is considered to perform similarly to the National Cancer Institute Diet History Questionnaire and the Block food-frequency questionnaire [26]. In a validation study from the Nurses Health Study, the correlation coefficients between questionnaire and diet record responses for several of the sweet and beverage items of interest were: chocolate 0.41, sweets 0.63, coffee 0.77, tea 0.86, coke/pepsi 0.84, and non-cola carbonated beverages 0.40 [27].

The specified portion size of each food item was considered as one serving, and the food frequency responses were transformed into servings per day. The options for frequency of food intake were: never, <1/month (rarely), 1–3/month, 1/week, 2–4/week, 5–6/week, 1/day, 2–3/day, 4–5/day, and 6+/day. Consumption of total and individual sweets and beverage items was examined. Sweets included pure chocolate candy bars or packets of candy (e.g., M&Ms), other mixed candy bars (e.g., Snickers/Milky Way/Reeses), candy without chocolate (e.g., mints, lifesavers), and sweet condiments (i.e., jams/jellies/preserves/syrup/honey). Total carbonated beverages included those with sugar (i.e., Coke/Pepsi/other colas, caffeine-free Coke/Pepsi/other colas, and other) and sugar-free types (i.e., low-calorie colas, low-calorie caffeine-free colas, and other low-calorie carbonated beverages such as diet 7-up, Fresca, diet ginger ale, etc.). Other sweetened beverages assessed included punch/lemonade/non-carbonated fruit drinks (not juice). Other non-caloric or low-calorie beverages assessed for comparison included plain water, herbal tea, non-herbal tea, decaffeinated coffee, and caffeinated coffee. While we did not have specific information regarding sugar added to tea or coffee, we were able to examine overall teaspoons of sugar added to food or beverages. The range of the correlation coefficients among all of the sweets was 0.09–0.41. The strongest correlation was observed between pure chocolate candy bars or packets of candy and other mixed candy bars, and the weakest association was found between sweet condiments and candy without chocolate. The range of the correlation coefficients among all of the beverages was 0.00–0.32. The strongest correlations were observed for items such as Coke/Pepsi/other colas with sugar and other carbonated beverages (correlation coefficient = 0.32) and low-calorie colas with other low-calorie carbonated beverages (correlation coefficient = 0.24).

The completed food frequency questionnaires were sent to the Harvard School of Public Health, Department of Nutrition for assessment of nutrient intake using their previously published methods [28]. Of interest to this report, nutrients assessed included total calories, and three sugars—fructose, sucrose, and lactose. These were examined in quartiles based on control group cut-points.

Odds ratios (OR) and 95% confidence intervals (CI) were computed using unconditional logistic regression to estimate the relative risk (hereafter called risk) of pancreatic cancer. No substantial difference in ORs was observed when we included or excluded study participants with extremely low (<500 kcal/day) or extremely high (>3,500 kcal/day) caloric intake, and these participants were retained in the final analyses. The linear trend test was performed using the Wald procedure. Potential confounding effects were investigated for total calories, sex, body-mass index (BMI), cigarette smoking, race, education, history of diabetes, physical activity, and other food groups. Total caloric intake was divided into quartiles based on the consumption among control participants by sex (cutpoints: men: ≤1,545; 1,546–1,925; 1,926–2,364; ≥2,365 kcal/day; women: ≤1,333; 1,334–1,695; 1,696–2,113; ≥2,114 kcal/day). BMI was categorized as <25 (normal), 25.0–29.9 (overweight), and ≥30 (obese) kg/m², based on the World Health Organization criteria. Smokers were defined as participants who had smoked >100 cigarettes in their lifetime, or a pipe or cigar for at least once a month for ≥6 months. Participants were classified as: never smokers; former cigarette smokers who quit smoking >15 years ago; former cigarette smokers who quit smoking 1–15 years ago; former smokers who quit within one year prior to diagnosis or to interview, and current cigarette smokers; and pipe and/or cigar smokers. Participants provided self-reported data on race (white, black or African American, Asian or Pacific Islander, or “other”); education (<high-school graduate, high-school graduate, 1–4 years college, and graduate work); history of diabetes (yes/no); and frequency of non-occupational physical activity (i.e., 30 min intervals performed ≤1/month, 2–4/month, 2–6/week, or 1+/day). We also considered intake of other food groups in quartiles based on previous publications from this population (i.e., red meat, white meat, dairy, vegetable and fruit, eggs, fish, whole grain, and refined grain) [11, 19, 20].

Stepwise logistic regression modeling (using p = 0.05) was used to determine the variables that provided the best-fit among the potentially confounding factors. By these criteria, only total energy appeared to confound the relationships between sweets and beverages and pancreatic cancer risk. For comparability with other studies and our previous publications, we first present a parsimonious model in the tables that was adjusted for the matching factors of age and sex, and for total energy intake. We then present a full multivariable model additionally adjusted for the following putative confounding factors: BMI, smoking status, race, education, history of diabetes, physical activity, and other food groups. Adjustment for other food groups included mutual adjustment for all sweet and beverage risk factors of interest plus previously examined food groups [11, 19, 20] (e.g., total sweets were adjusted for all sweetened and unsweetened beverage intake, as well as consumption of red meat, white meat, vegetables and fruits, eggs, fish, dairy, whole grains, and refined grains). The results for men and women are presented combined and separately. Because there was little difference between the parsimonious model and the fully adjusted model, we present only the full model for the results stratified by sex. To further consider the effects of confounding or effect modification on these results, we examined total sweets and carbonated beverages (with or without sugar) in multivariable models within strata of BMI (normal, overweight, and obese), exercise (monthly vs. daily exercisers), diabetes (yes/no), and smoking history (current, former, never). We considered stratifying BMI further, however, limited numbers made this infeasible (e.g., BMI ≥ 35: n = 13 cases/31 controls; BMI ≥ 40: n = 3 cases/6 controls).