Dietary Carbohydrates, Fiber, and Breast Cancer Risk

American Journal of Epidemiology Copyright 2004 by the Johns Hopkins Bloomberg School of Public Health All rights reserved Vol. 159, No. 8 Printed in U.S.A. DOI: 10.1093/aje/kwh112 Dietary Carbohydrates, Fiber, and Breast Cancer Risk Michelle D. Holmes 1, Simin Liu 2,3, Susan E. Hankinson 1,3, Graham A. Colditz 1,3, David J. Hunter 1,3,4, and Walter C. Willett 1,3,4 1 Channing Laboratory, Department of Medicine, Brigham and Women s Hospital and Harvard Medical School, Boston, MA. 2 Division of Preventive Medicine, Department of Medicine, Brigham and Women s Hospital and Harvard Medical School, Boston, MA. 3 Department of Epidemiology, Harvard School of Public Health, Boston, MA. 4 Department of Nutrition, Harvard School of Public Health, Boston, MA. Received for publication June 24, 2003; accepted for publication November 25, 2003. Dietary fiber, fiber fractions, carbohydrate, glycemic index, and glycemic load were prospectively assessed five times over 18 years with a validated food frequency questionnaire in relation to breast cancer risk among 88,678 women (aged 34 59 years at baseline) in the Nurses Health Study. Incident breast cancer occurred in 4,092 of these women between 1980 and 1998. The authors observed no material association between carbohydrate intake, glycemic index and glycemic load, total dietary fiber intake, and breast cancer risk. The relative risks for the highest versus the lowest quintile of intake were 0.97 (95% confidence interval (CI): 0.87, 1.08) for carbohydrates, 1.08 (95% CI: 0.97, 1.19) for glycemic index, 0.99 (95% CI: 0.89, 1.10) for glycemic load, and 0.98 (95% CI: 0.87, 1.11) for fiber. The relative risk comparing those in the highest 0.7% of fiber intake (>30 g/day) with those in the lowest 10% of fiber intake ( 10 g/day) was 0.68 (95% CI: 0.43, 1.06). Analyses stratified by menopausal status and body mass index also showed no clear risk pattern. In this cohort of middle-aged women, no overall association was found for dietary carbohydrates, glycemic index and glycemic load, and breast cancer risk. This study also confirmed the lack of an overall association between intake of fiber and fiber types and breast cancer risk observed in other prospective studies. breast neoplasms; dietary carbohydrates; dietary fiber; glycemic index; prospective studies; questionnaires; risk factors Abbreviations: BMI, body mass index; CI, confidence interval; RR, relative risk. More than 20 years ago, Goldin et al. (1) reported that 10 premenopausal vegetarian women consumed more fiber, excreted more estrogens, and had lower estrogen levels than 10 omnivorous women. Since then, the hypothesis that dietary fiber could lower breast cancer risk by lowering circulating estrogens has received much attention. The proposed mechanisms include alteration of the intestinal flora to increase estrogen excretion, competition for estrogen-binding sites by phytoestrogens, direction of estrogen metabolism toward less active forms, and reduction in sex hormone-binding globulin (2). However, prospective studies of dietary fiber and breast cancer risk have in general shown no association (3 6). To our knowledge, only one recently published prospective study has examined fiber fractions (7). Lately, some have hypothesized that dietary carbohydrate quality could affect circulating insulin levels and thus promote cancer growth (8, 9). Because insulin levels are higher in the presence of insulin resistance and obesity (10, 11), we hypothesized that high carbohydrate or glycemic load intake would increase breast cancer risk primarily in overweight women. We therefore prospectively examined the association of dietary fiber, fiber fractions, carbohydrate, glycemic index, and glycemic load with the risk of breast cancer during 18 years of follow-up in the Nurses Health Study. The extended follow-up with repeated measures of diet mini- Correspondence to Dr. Michelle D. Holmes, Channing Laboratory, 181 Longwood Avenue, Boston, MA 02115 (e-mail: michelle.holmes@channing.harvard.edu). 732

Dietary Carbohydrates, Fiber, and Breast Cancer Risk 733 mized the possibility that a modest, but still potentially important effect would be missed. MATERIALS AND METHODS The Nurses Health Study cohort In 1976, the Nurses Health Study cohort was established when 121,700 female registered nurses from across the United States, aged 30 55 years, answered a mailed questionnaire on risk factors for cancer and cardiovascular disease. Every 2 years since, we have sent follow-up questionnaires to Nurses Health Study participants. In 1980, we added a 61-item food frequency questionnaire designed to assess dietary intake. In 1984, 1986, 1990, and 1994, an expanded food frequency questionnaire was used. This analysis is based on women who answered the 1980 diet questionnaire. Women were excluded if their dietary information was incomplete ( 10 blank items in 1980 and 70 blank items after 1980) or if their scores were implausible (<500 kcal or >3,500 kcal/day) for total energy intake (approximately 2 percent of returned diet questionnaires). Women were also excluded if they had been diagnosed with cancer other than nonmelanoma skin cancer (3,093 women) prior to 1980. Data for 88,678 women were included in the analysis. The semiquantitative food frequency questionnaires The food frequency questionnaires have been described in detail previously (12). A commonly used portion size was specified for each food (e.g., one slice of bread or one egg). Participants were asked to describe how frequently over the past year they had consumed that portion of that particular food. The nine prespecified responses ranged from never to six or more times per day. The validity and reproducibility of the food frequency questionnaires for nutrients and foods have been documented elsewhere (12 14). In this analysis, we primarily examined the association of breast cancer risk with intakes of the following nutrients: carbohydrates; total fiber; and cereal, fruit, and vegetable fiber. We also examined the associations with glycemic index and glycemic load. The glycemic index ranks foods on the basis of their relative postprandial blood glucose response per gram of carbohydrate (15). We calculated a food s glycemic load by multiplying the carbohydrate content of each food by its glycemic index value (16); we then multiplied this value by frequency of consumption and summed over all food items to produce the dietary glycemic load. Dietary glycemic load thus represents the quality and quantity of carbohydrates and their interaction, with a higher glycemic index having a greater impact at higher carbohydrate intakes. Each unit of dietary glycemic load represents the glycemic equivalent of 1 g of carbohydrate from white bread. Dietary glycemic index and glycemic load, as assessed by this food frequency questionnaire, do have biologic validity because they have been associated with plasma lipid levels (11). Cereal, fruit, and vegetable fiber were determined by summing the Association of Analytical Communities (AOAC) total dietary fiber for all of the cereal (or grain) foods, all of the fruits, and all of the vegetables on the questionnaire, respectively. If dishes were mixed (e.g., pie = crust (cereal) + apples (fruit)), the food was evaluated as the specific weight of the cereal, fruit, and vegetable component and the total dietary fiber from that food source added to the sum of that specific component fiber. Identification of breast cancer cases In each biennial questionnaire, participants were asked whether they had been diagnosed with breast cancer in the previous 2 years. Deaths were identified by a report from a family member, the postal service, or the National Death Index; ascertainment was estimated to be 98 percent complete. Follow-up of the initial cohort for this analysis was 96 percent complete through 1998. Medical records were obtained for breast cancer cases identified by either self-report or vital records, and over 99 percent of these records confirmed the self-report. We included 4,092 invasive breast cancer cases in this analysis. Women developing carcinoma in situ were excluded at the time of diagnosis. Statistical analysis Each participant accumulated person-time beginning with the return of the 1980 questionnaire and ending with her cancer diagnosis, her death, or June 1, 1998, whichever came first. Cox proportional hazards models with age in months as the underlying time variable were used to calculate relative risks of breast cancer adjusted for other breast cancer risk factors. To provide more stable estimates and to take into account dietary changes over time, we calculated the cumulative average intake of foods and nutrients from all available dietary questionnaires up to the start of each 2-year interval. In this calculation, the 1980 diet was related to 1980 1984 breast cancer incidence; the average of the 1980 and 1984 diets was related to 1984 1986 breast cancer incidence; the average of the 1980, 1984, and 1986 diets was related to 1986 1990 breast cancer incidence; the average of the 1980, 1984, 1986, and 1990 diets was related to 1990 1994 breast cancer incidence; and the average of all five diets was related to 1994 1998 breast cancer incidence. For the baseline analyses, the 1984 dietary assessment was used. It was significantly expanded from the 1980 version, and follow-up was from 1984 to 1998. Dietary intakes were energy adjusted by using the residual method (13) and were categorized into quintiles. The lowest quintile of intake was the reference category. Total energy intake was included in each regression model. Because exposures may affect risk of pre- and postmenopausal breast cancer differently, results were stratified by menopausal status at the time of diagnosis. We additionally stratified the results by a body mass index (BMI) of less than 25 kg/m 2 and greater than or equal to 25 kg/m 2 and by physical activity. The following nondietary covariates were updated every 2 years: age, history of benign breast disease, menopausal status, age at menopause, use and duration of use of postmenopausal hormones, parity, age at first birth, and BMI. Age at menarche and height were determined at baseline,

734 Holmes et al. TABLE 1. 1990 levels of selected variables for 88,678 women in the US Nurses Health Study * SD, standard deviation. Variable Age (years) 56.5 (7.1) Body mass index (kg/m 2 ) 25.8 (4.9) Mean (SD*) (10th 90th percentile) Energy intake (kcal/day) 1,751 (509) Alcohol intake (g/day) 5.1 (9.5) Fiber intake (energy-adjusted g/day) 18.1 (5.5) (12.0 24.8) Glycemic index (energy-adjusted units) 75.2 (5.3) (68.8 81.4) Carbohydrate intake (energy-adjusted g/day) 200 (32) (160 240) Age (years) at first birth (parous women only) 24.8 (3.3) Family history of breast cancer 10 Personal history of benign breast disease 40 Parous 92 Postmenopausal 68 Current use of hormone replacement therapy (postmenopausal women only) 38 % and information on family history of breast cancer was sought in 1976, 1982, 1988, 1992, and 1996. Waist-hip ratio was ascertained in 1986. Women of uncertain ovulatory status (mainly those who had undergone hysterectomy but had intact ovaries) were excluded from analyses that included stratification by menopausal status until they reached the age at which 90 percent of the cohort had become postmenopausal; at that point, they were considered postmenopausal. In tests for linear trend across quintiles of food and nutrient intake, an ordinal rank was assigned to each category. Interaction terms were generated by multiplying continuous nutrient intakes by continuous BMI. RESULTS All women Table 1 shows the characteristics of the population in 1990, approximately halfway through the follow-up period. We have previously reported associations between several factors in table 1 and risk of breast cancer: a decreased risk of parity (17) and increased risks of benign breast disease (18); alcohol intake (19); and, among postmenopausal women, use of hormone replacement therapy (19, 20) as well as overweight and weight gain (21). During 18 years of follow-up, we documented 4,092 incident cases of breast cancer in the 88,678 women. We found no significant overall association between intake of carbohydrates, dietary glycemic index and glycemic load, total fiber, and risk of incident breast cancer across all menopausal categories. The relative risks for the highest versus the lowest quintile of intake were 0.97 (95 percent confidence interval (CI): 0.87, 1.08) for carbohydrates, 1.08 (95 percent CI: 0.97, 1.19) for glycemic index, 0.99 (95 percent CI: 0.89, 1.10) for glycemic load, and 0.98 (95 percent CI: 0.87, 1.11) for fiber. We performed a number of sensitivity analyses on the association between fiber intake and breast cancer risk. We found no association between fiber intake and breast cancer risk when we used as the dietary exposure the baseline (1984) diet, the simply updated diet (i.e., updated at each follow-up questionnaire, not cumulatively averaged over time), and follow-up time lagged 8 years after dietary assessment. We examined breast cancer risk associated with fiber intakes more extreme than the highest quintile. The highest decile of fiber intake was not associated with a decreased risk of breast cancer compared with the lowest decile of intake (relative risk (RR) = 0.97, 95 percent CI: 0.83, 1.14). When absolute cutpoints were used, women who had cumulatively averaged a fiber intake of more than 30 g/day (highest 0.7 percent of the cohort) had a nonsignificantly decreased risk of breast cancer compared with women consuming 10 g/day or less (lowest 10 percent of the cohort) (RR = 0.68, 95 percent CI: 0.43, 1.06); this finding was also evident when only the baseline (1984) dietary exposure was used (RR for >30 g/day of fiber (1 percent of the cohort) vs. 10 g/day of fiber (6 percent of the cohort) = 0.73, 95 percent CI: 0.53, 1.00). The associations between intake of cereal fiber and diabetes and between glycemic load and diabetes negatively confound each other (22, 23). We found no association between cereal fiber intake and breast cancer risk, nor glycemic load and breast cancer risk, when controlling for the other simultaneously (RR for the highest vs. the lowest quintile of cereal fiber = 1.08, 95 percent CI: 0.97, 1.20) and for glycemic load (RR = 0.96, 95 percent CI: 0.86, 1.07). In addition, intake of cereal fiber did not modify the association with glycemic load. We categorized glycemic load and cereal fiber intake into tertiles and cross-classified participants. Compared with those having the highest fiber intake and the lowest glycemic load, those with the lowest fiber intake and the highest glycemic load did not have an

Dietary Carbohydrates, Fiber, and Breast Cancer Risk 735 increased risk of breast cancer (RR = 0.95, 95 percent CI: 0.79, 1.14). We similarly categorized total fat and total fiber intake into tertiles and cross-classified participants. Compared with those having the highest fiber intake and the lowest fat intake, those having the lowest fiber intake and the highest fat intake did not have an increased risk of breast cancer (RR = 1.03, 95 percent CI: 0.92, 1.16). We also examined the association of breast cancer risk with other fiber types (soluble and insoluble fiber, legume fiber, cruciferous vegetable fiber, cellulose, hemicellulose, and lignans), simple sugars (sucrose, fructose, and lactose), and carbohydrate-containing foods: candy, pie, cake, cookies, cereal, bread, rice, potatoes, punch, and soda pop. Each nutrient and food was examined separately. We found no clear pattern of association between intake of these dietary factors and breast cancer risk. In addition, we repeated the above analyses by excluding women with diabetes, again with no substantial change in results. Although we adjusted for BMI as a continuous variable, distribution of body fat may further modify these relations. We performed an analysis stratified by waist-hip ratio with follow-up beginning in 1986, when this information was available, with no evidence of effect modification. Controlling for physical activity and stratifying by physical activity also did not change the results. Premenopausal women Table 2 shows the multivariate association between dietary factors and breast cancer risk for premenopausal women, overall and stratified by BMI. We chose a BMI cutpoint of 25 kg/m 2, the definition of overweight according to standard guidelines (24). For the 852 cases of breast cancer in premenopausal women, there was no significant association overall with intake of carbohydrate. For heavier women (BMI, 25 kg/m 2 ), we observed a borderline inverse association with carbohydrate intake (RR for the highest compared with the lowest quintile of intake = 0.72, 95 percent CI: 0.48, 1.07; p for linear trend = 0.04) and with glycemic load (RR = 0.68, 95 percent CI: 0.45, 1.03; p for linear trend = 0.02). However, the analysis of premenopausal women whose BMI was 25 kg/m 2 was based on only 292 breast cancer cases. In addition, we found no association of dietary glycemic index or glycemic load; total fiber; and cereal, fruit, or vegetable fiber with breast cancer risk among premenopausal women. Results for premenopausal women adjusted only for age and energy intake were very similar to the multivariate findings. Postmenopausal women Among postmenopausal women, we documented 2,924 cases of breast cancer (table 3). We found no association between intake of carbohydrate and postmenopausal breast cancer risk. This result did not vary when data were stratified by BMI of <25 kg/m 2 or BMI of 25 kg/m 2. We also examined the association between dietary carbohydrate and breast cancer risk among postmenopausal women who were even more overweight and found no such association for those whose BMI was 29 kg/m 2 or those whose BMI was 32 kg/m 2. We found a borderline positive association between glycemic index (but not glycemic load) and postmenopausal breast cancer risk (RR for the highest compared with the lowest quintile of intake = 1.15, 95 percent CI: 1.02, 1.30; p for linear trend = 0.02). This association was stronger among postmenopausal women whose BMI was <25 kg/m 2 (RR = 1.28, 95 percent CI: 1.08, 1.53; p = 0.003) and was not present among postmenopausal women whose BMI was 25 kg/m 2. In secondary analyses, we further examined the association between glycemic index and breast cancer risk among postmenopausal women whose BMI was even less than 25 kg/m 2. No association was found between glycemic index and breast cancer risk among those whose BMI was <23 kg/m 2 nor among those whose BMI was <21 kg/m 2. We also examined the association of dietary glycemic index and of glycemic load with breast cancer risk among postmenopausal women with even greater degrees of overweight than a BMI of 25 kg/m 2. We also found no association between these factors and breast cancer risk among postmenopausal women with a BMI of 29 kg/m 2 nor among women with a BMI of 32 kg/m 2. In addition, we found no association of intakes of total fiber and of cereal, fruit, and vegetable fiber with postmenopausal breast cancer risk. These results did not vary by BMI. Results for postmenopausal women adjusted only for age and energy intake were very similar to the multivariate findings. DISCUSSION An early meta-analysis of case-control studies that found a positive association of energy and fat intake with breast cancer found no association for carbohydrate intake (25). Much research effort then focused further on the fat hypothesis. A high carbohydrate intake was seen primarily as necessary to maintain caloric intake in low-fat diets investigated in trials to reduce markers of breast cancer risk, such as mammographic density (26), and plasma hormone levels (27, 28) or breast cancer risk itself (29). Similar to previous studies, we found no association between carbohydrate intake and overall breast cancer risk in pre- or postmenopausal women. However, recent studies have raised the hypothesis that carbohydrate quality rather than absolute quantity of intake may be important in breast cancer risk, particularly for premenopausal women. In a case-control study of 140 premenopausal bilateral breast cancer cases from California and Quebec, Canada, authors reported an odds ratio of 2.3 (95 percent CI: 1.0, 5.2) for the highest versus the lowest quartile of carbohydrate intake (30). In this same study, no association was found for intake of fruits, vegetables, and grains, but an elevated risk was reported for intake of sweetened beverages (odds ratio = 2.6, 95 percent CI: 1.2, 5.8) for the highest versus the lowest quartile of intake (30). In a large (2,569 cases) Italian population-based casecontrol study, the breast cancer cases had a higher mean dietary glycemic index, a measure of carbohydrate quality based on postprandial blood glucose. When the highest quin-

736 Holmes et al. TABLE 2. Multivariate* association between dietary factors and breast cancer risk for premenopausal women in the United States, overall and stratified by body mass index, 1980 1998 Dietary factor 1990 median value Quintile All premenopausal women (n = 53,891; cases = 852) Body mass index of <25 kg/m 2 (n = 39,403; cases = 527) Body mass index of 25 kg/m 2 (n = 25,129; cases = 292) p for interaction RR 95% CI RR 95% CI RR 95% CI Carbohydrate (energy-adjusted g/day) 159 1 1.00 1.00 1.00 183 2 0.91 0.74, 1.13 1.05 0.79, 1.40 0.77 0.54, 1.11 200 3 0.92 0.74, 1.14 1.04 0.78, 1.39 0.81 0.57, 1.15 215 4 0.83 0.66, 1.05 1.00 0.74, 1.35 0.62 0.42, 0.92 240 5 0.98 0.78, 1.23 1.20 0.89, 1.61 0.72 0.48, 1.07 p = 0.61 p = 0.37 p = 0.04 p = 0.46 Glycemic index 69 1 1.00 1.00 1.00 73 2 0.90 0.71, 1.13 0.96 0.72, 1.30 0.72 0.49, 1.06 75 3 0.89 0.71, 1.12 1.00 0.75, 1.34 0.69 0.47, 1.02 78 4 0.93 0.75, 1.17 1.00 0.75, 1.33 0.75 0.51, 1.09 81 5 1.02 0.82, 1.28 1.06 0.79, 1.42 0.83 0.57, 1.22 p = 0.68 p = 0.64 p = 0.48 p = 0.17 Glycemic load 116 1 1.00 1.00 1.00 136 2 0.95 0.77, 1.18 0.92 0.69, 1.22 1.04 0.73, 1.47 150 3 0.97 0.78, 1.20 1.01 0.76, 1.34 0.85 0.59, 1.24 164 4 0.90 0.72, 1.12 1.02 0.77, 1.36 0.74 0.50, 1.10 186 5 0.87 0.70, 1.12 1.01 0.75, 1.35 0.68 0.45, 1.03 p = 0.26 p = 0.70 p = 0.02 p = 0.70 Total fiber (energy-adjusted g/day) 12.1 1 1.00 1.00 1.00 15.1 2 1.11 0.90, 1.36 1.16 0.90, 1.51 1.15 0.80, 1.66 17.4 3 1.08 0.87, 1.34 1.03 0.78, 1.37 1.23 0.84, 1.80 20.1 4 1.15 0.91, 1.45 1.15 0.86, 1.55 1.29 0.86, 1.96 24.8 5 0.99 0.75, 1.29 1.10 0.78, 1.54 0.85 0.51, 1.42 p = 0.79 p = 0.62 p = 0.94 p = 0.93 Cereal fiber (energy-adjusted g/day) 2.4 1 1.00 3.4 2 0.94 0.75, 1.17 4.4 3 1.04 0.83, 1.29 5.8 4 1.21 0.97, 1.50 8.4 5 0.99 0.78, 1.25 p = 0.29 Fruit fiber (energy-adjusted g/day) 1.3 1 1.00 2.4 2 0.94 0.77, 1.15 3.5 3 1.00 0.82, 1.24 4.8 4 1.09 0.88, 1.36 7.1 5 0.86 0.67, 1.10 p = 0.78 Vegetable fiber (energy-adjusted g/day) 3.6 1 1.00 5.0 2 0.99 0.80, 1.22 6.1 3 0.97 0.77, 1.21 7.5 4 0.95 0.75, 1.21 10.2 5 0.95 0.72, 1.25 p = 0.65 * All relative risks were adjusted for age (continuous); body mass index (kg/m 2, continuous); 2-year time period; total energy intake (quintiles); alcohol intake (0, <5, 5 14.9, 15 g/day); parity and age at first birth (nulliparous, 1 2 births and age at first birth of <25 years, 1 2 births and age at first birth from 25 to <30 years, 1 2 births and age at first birth of 30 years, 3 4 births and age at first birth of <25 years, 3 4 births and age at first birth from 25 to <30 years, 3 4 births and age at first birth of 30 years, 5 births and age at first birth of <25 years, 5 births and age at first birth from 25 to <30 years, missing); height in inches (1 inch = 2.54 cm; <63, 63 63.9, 64 65.9, 66); family history of breast cancer (yes/no); history of benign breast disease (yes/no); and age at menarche in years ( 12, 13, 14). According to quintile of cumulatively averaged and updated carbohydrate and fiber intake and to glycemic index and glycemic load. RR, relative risk; CI, confidence interval.

Dietary Carbohydrates, Fiber, and Breast Cancer Risk 737 TABLE 3. Multivariate* association between dietary factors and breast cancer risk for postmenopausal women in the United States, overall and stratified by body mass index, 1980 1998 Dietary factor Quintile All postmenopausal women (n = 76,200; cases = 2,924) Body mass index of <25 kg/m 2 (n = 47,051; cases = 1,343) Body mass index of 25 kg/m 2 (n = 46,111; cases = 1,344) p for interaction RR 95% CI RR 95% CI RR 95% CI Carbohydrate 1 1.00 1.00 1.00 2 0.97 0.86, 1.10 1.00 0.83, 1.20 0.98 0.82, 1.16 3 0.99 0.87, 1.11 1.06 0.88, 1.28 0.91 0.76, 1.10 4 1.02 0.91, 1.16 1.02 0.85, 1.24 1.01 0.84, 1.22 5 0.96 0.84, 1.09 0.95 0.78, 1.15 0.96 0.80, 1.17 p = 0.82 p = 0.65 p = 0.91 p = 0.08 Glycemic index 1 1.00 1.00 1.00 2 0.99 0.88, 1.10 1.00 0.85, 1.19 1.02 0.87, 1.21 3 0.99 0.88, 1.11 0.93 0.77, 1.11 1.08 0.91, 1.27 4 1.05 0.93, 1.18 1.12 0.95, 1.34 1.08 0.91, 1.29 5 1.15 1.02, 1.30 1.28 1.08, 1.53 1.05 0.87, 1.26 p = 0.02 p = 0.003 p = 0.46 p = 0.22 Glycemic load 1 1.00 1.00 1.00 2 1.02 0.90, 1.14 1.05 0.87, 1.25 0.98 0.83, 1.17 3 1.01 0.89, 1.14 1.01 0.84, 1.22 0.96 0.81, 1.15 4 1.06 0.94, 1.20 1.12 0.93, 1.35 1.02 0.85, 1.22 5 1.03 0.90, 1.16 1.06 0.87, 1.28 0.97 0.80, 1.18 p = 0.51 p = 0.42 p = 0.95 p = 0.20 Total fiber 1 1.00 1.00 1.00 2 0.99 0.87, 1.12 1.04 0.86, 1.26 0.94 0.78, 1.13 3 1.02 0.89, 1.16 1.08 0.89, 1.31 0.92 0.76, 1.11 4 0.93 0.82, 1.07 0.93 0.76, 1.14 0.88 0.72, 1.07 5 0.96 0.83, 1.10 1.04 0.84, 1.27 0.85 0.69, 1.05 p = 0.35 p = 0.83 p = 0.11 p = 0.10 Cereal fiber 1 1.00 2 0.98 0.87, 1.11 3 0.99 0.88, 1.13 4 1.05 0.93, 1.19 5 1.08 0.96, 1.22 p = 0.09 Fruit fiber 1 1.00 2 1.02 0.90, 1.16 3 0.95 0.84, 1.08 4 0.94 0.83, 1.07 5 0.92 0.81, 1.04 p = 0.08 Vegetable fiber 1 1.00 2 1.00 0.88, 1.13 3 0.93 0.82, 1.06 4 0.94 0.82, 1.08 5 0.94 0.82, 1.08 p = 0.27 * All relative risks were adjusted for the same factors as those specified in table 2 and also for age at menopause and hormone replacement therapy (HRT) use categories (age at menopause of <45 years and no HRT use, age at menopause of <45 years and past HRT use, age at menopause of <45 years and current HRT use for <5 years, age at menopause of <45 years and current HRT use for 5 years, age at menopause of 45 52 years and no HRT use, age at menopause of 45 52 years and past HRT use, age at menopause of 45 52 years and current HRT use for >5 years, age at menopause of 45 52 years and current HRT use for 5 years, age at menopause of 53 years and no HRT use, age at menopause of 53 years and past HRT use, age at menopause of 53 years and current HRT use for <5 years, age at menopause of 53 years and current HRT use for 5 years) and duration of menopause (continuous). According to quintile of cumulatively averaged and updated carbohydrate and fiber intake and to glycemic index and glycemic load. The ranges of intakes are the same as those listed in table 2. RR, relative risk; CI, confidence interval.

738 Holmes et al. tile of glycemic index intake was compared with the lowest quintile, the odds ratio was 1.36 (95 percent CI: 1.14, 1.64); in this study, results were similar for pre- and postmenopausal women (31). In a prospective study with 5.5 years of follow-up and 144 incident cases also from Italy, the risk of breast cancer was nearly threefold higher when the highest quartile of fasting blood glucose was compared with the lowest (32). However, in a Swedish prospective study of dietary patterns with 1,328 cases, no association was found between a Western dietary pattern including refined grains and sweets and the risk of breast cancer, even when stratified by age (40 49 years vs. 50 years) (33). Because of the association between insulin resistance and obesity, we hypothesized that high carbohydrate or glycemic load intake would increase breast cancer risk primarily in overweight women. Few studies have reported results additionally stratified by measures of obesity, including the Nurses Health Study II, a cohort similar to the Nurses Health Study but younger (aged 26 44 years at baseline), with 714 premenopausal breast cancer cases during 8 years of follow-up. A positive association between carbohydrate intake, glycemic load, and breast cancer risk was observed for overweight premenopausal women in the Nurses Health Study II, although it was not statistically significant; the relative risks for the highest versus the lowest quintile of intake were 1.47 (95 percent CI: 0.84, 2.59) for carbohydrate and 1.46 (95 percent CI: 0.89, 2.39) for glycemic load (34). We found a borderline reduced breast cancer risk for the highest intakes of carbohydrate and glycemic load for only overweight (BMI, 25 kg/m 2 ) premenopausal women, our smallest subgroup (292 cases). This result is in contrast to results from the Nurses Health Study II, the younger companion cohort to the Nurses Health Study (34). For premenopausal women in the Nurses Health Study II, Cho et al. (34) found a reduced breast cancer risk for the highest intakes of carbohydrate (quintile 5 vs. quintile 1 RR = 0.62, 95 percent CI: 0.40, 0.97; p for linear trend = 0.02) and glycemic load (RR = 0.83, 95 percent CI: 0.56, 1.24; p for linear trend = 0.19) among women whose BMI was less than 25 kg/m 2 (422 cases). In the Nurses Health Study II, an increased risk of breast cancer was seen for a higher intake of carbohydrate and glycemic load among overweight premenopausal women; the test for the interactions (carbohydrate intake BMI) and (glycemic load BMI) were statistically significant (p = 0.02 for each). Although the distribution of carbohydrate intake was similar between the Nurses Health Study in 1990 and the Nurses Health Study II in 1991, the upper range of dietary glycemic load was higher in the Nurses Health Study II. The median glycemic loads for the highest quintile were 186 for the 1990 Nurses Health Study and 211 for the 1991 Nurses Health Study II. We found a slightly elevated breast cancer risk for the highest intake of glycemic index among lowerweight (BMI, <25 kg/m 2 ) postmenopausal women. Given that our results are opposite those found in the Nurses Health Study II, are contrary to the hypothesis that adverse associations would be observed for heavier women, and show opposite trends depending on menopausal status, it is likely that the results observed for subgroups may be due to chance. However, even for premenopausal women, Nurses Health Study II participants were substantially younger, so biologic reasons for these differences are possible. Although plausible mechanisms have been proposed for how intake of fiber could lower breast cancer risk (2), most prospective studies of fiber intake and breast cancer risk have not shown a reduced risk (3 6). Recently, Mattisson et al. (35) reported an interaction between fat and fiber intakes and breast cancer risk in a Swedish cohort, which we were not able to replicate in our study. It is possible that particular fiber fractions are more important than total fiber intake. Recently, Terry et al. (7) reported on breast cancer risk associated with dietary fiber fractions. They found no associations between dietary intake of fiber fractions (total, soluble, insoluble, cereal, fruit, and vegetable fiber, as well as lignin and cellulose) and breast cancer risk among 89,835 women in the Canadian National Breast Screening Study. We have confirmed these findings in a cohort of similar size (88,678 women), with a slightly longer follow-up period (18 years) and 60 percent more cases (n = 4,092). In addition, we extended these findings by measuring diet five times over the follow-up period, which reduces random error in long-term dietary measurement (36). We also performed secondary analyses stratified by menopausal status and BMI, and sensitivity analyses looking at the exposure in many different ways, and were similarly unable to find an association between fiber intake and breast cancer risk. We did find a suggestion of reduced risk, although not statistically significant, when we compared those consuming >30 g of fiber per day with those consuming 10 g/day (RR = 0.68, 95 percent CI: 0.43, 1.06). However, only 0.7 percent of women in this cohort consumed as much as 30 g/day, and only 10 percent consumed as little as 10 g/day. Therefore, it may be possible that breast cancer risk would be reduced in women consuming fiber in much higher amounts than the distribution of intake in this study. A potential limitation of any study is that dietary intakes are measured with error. However, high intakes of cereal fiber and a low dietary glycemic load measured in the same population with the same dietary assessment have been associated with a reduced risk of both coronary heart disease (37, 38) and type 2 diabetes (23). We conclude that midlife dietary intake of total fiber and fiber types, as well as carbohydrate quality, are unlikely to substantially affect breast cancer risk. More thorough evaluations of early-life dietary exposures should be undertaken. However, our findings do not exclude the possibility that diets including a very high intake of fiber (>30 g/day) may modestly decrease risk, and diets with a high glycemic index may modestly increase risk for postmenopausal women. These relations merit further exploration. ACKNOWLEDGMENTS The work reported on in this paper was supported by National Institutes of Health grant CA87969. The authors acknowledge the assistance of Dr. Meir Stampfer in reviewing the manuscript.

Dietary Carbohydrates, Fiber, and Breast Cancer Risk 739 REFERENCES 1. Goldin BR, Aldercreutz H, Gorbach SL, et al. Estrogen excretion patterns and plasma levels in vegetarian and omnivorous women. N Engl J Med 1982;307:1542 7. 2. Cohen LA. Dietary fiber and breast cancer. Anticancer Res 1999;19:3685 8. 3. Willett WC, Hunter DJ, Stampfer MJ, et al. Dietary fat and fiber in relation to risk of breast cancer: an 8-year follow-up. JAMA 1992;268:2037 44. 4. Graham S, Zielezny M, Marshall J, et al. Diet in the epidemiology of postmenopausal breast cancer in the New York State cohort. Am J Epidemiol 1992;136:1327 37. 5. Verhoeven DTH, Assen N, Goldbohm RA, et al. Vitamins C and E, retinol, beta-carotene and dietary fiber in relation to breast cancer risk: a prospective cohort study. Br J Cancer 1997;75:149 55. 6. Järvinen R, Knekt P, Seppänen R, et al. Diet and breast cancer risk in a cohort of Finnish women. Cancer Lett 1997;114:251 3. 7. Terry P, Jain M, Miller AB, et al. No association among total dietary fiber, fiber fractions, and risk of breast cancer. Cancer Epidemiol Biomarkers Prev 2002;11:1507 8. 8. Kaaks R, Lukanova A. Energy balance and cancer: the role of insulin and insulin-like growth factor-i. Proc Nutr Soc 2001;60: 91 106. 9. Gupta K, Krishnaswamy G, Karnad A, et al. Insulin: a novel factor in carcinogenesis. Am J Med Sci 2002;323:140 5. 10. Jeppesen J, Schaaf P, Jones G, et al. Effects of low-fat, highcarbohydrate diets on risk factors for ischemic heart disease in postmenopausal women. Am J Clin Nutr 1997;65:1027 33. 11. Liu S, Manson JE, Stampfer MJ, et al. Dietary glycemic load assessed by food-frequency questionnaire in relation to plasma high-density-lipoprotein cholesterol and fasting plasma triacylglycerols in postmenopausal women. Am J Clin Nutr 2001; 73:560 6. 12. Willett W. Nutritional epidemiology. 2nd ed. New York, NY: Oxford University Press, 1998. 13. Willett WC, Sampson L, Stampfer MJ, et al. Reproducibility and validity of a semiquantitative food frequency questionnaire. Am J Epidemiol 1985;122:51 65. 14. Salvini S, Hunter DJ, Sampson L, et al. Food-based validation of a dietary questionnaire: the effects of week-to-week variation in food consumption. Int J Epidemiol 1989;18:858 67. 15. Foster-Powell K, Miller JB. International tables of glycemic index. Am J Clin Nutr 1995;62:871S 93S. 16. Wolever TMS, Jenkins DJ, Jenkins AL, et al. The glycemic index: methodology and clinical implications. Am J Clin Nutr 1991;54:846 54. 17. Rosner B, Colditz GA, Willett WC. Reproductive risk factors in a prospective study of breast cancer: the Nurses Health Study. Am J Epidemiol 1994;139:819 35. 18. Byrne C, Connolly JL, Colditz GA, et al. Biopsy confirmed benign breast disease, postmenopausal use of exogenous female hormones, and breast carcinoma risk. Cancer 2000;89: 2046 52. 19. Chen WY, Colditz GA, Rosner B, et al. Use of postmenopausal hormones, alcohol, and risk for invasive breast cancer. Ann Intern Med 2002;137:798 804. 20. Colditz GA, Hankinson SE, Hunter DJ, et al. The use of estrogens and progestins and the risk of breast cancer in postmenopausal women. N Engl J Med 1995;332:1589 93. 21. Huang Z, Hankinson SE, Colditz GA, et al. Dual effects of weight and weight gain on breast cancer risk. JAMA 1997;278: 1407 11. 22. Salmeron J, Ascherio A, Rimm EB, et al. Dietary fiber, glycemic load, and risk of NIDDM in men. Diabetes Care 1997;20: 545 50. 23. Salmeron J, Manson JE, Stampfer MJ, et al. Dietary fiber, glycemic load, and risk of non-insulin-dependent diabetes mellitus in women. JAMA 1997;277:472 7. 24. Clinical guidelines on the identification, evaluation, and treatment of overweight and obesity in adults: the evidence report. Bethesda, MD: National Institutes of Health, 1998. (NIH publication no. 98-4083). 25. Howe GR, Hirohata T, Hislop TG, et al. Dietary factors and risk of breast cancer: combined analysis of 12 case-control studies. J Natl Cancer Inst 1990;82:561 9. 26. Boyd NF, Greenberg C, Lockwood G, et al. Effects at two years of a low-fat, high-carbohydrate diet on radiologic features of the breast: results from a randomized trial. J Natl Cancer Inst 1997;89:488 96. 27. Boyd NF, Lockwood GA, Greenberg CV, et al. Effects of a low-fat high-carbohydrate diet on plasma sex hormones in premenopausal women: results from a randomized controlled trial. Br J Cancer 1997;76:127 35. 28. Berrino F, Bellati C, Secreto G, et al. Reducing bioavailable sex hormones through a comprehensive change in diet: the diet and androgens (DIANA) randomized trial. Cancer Epidemiol Biomarkers Prev 2001;10:25 33. 29. White E, Shattuck AL, Kristal AR, et al. Maintenance of a lowfat diet: follow-up of the Women s Health Trial. Cancer Epidemiol Biomarkers Prev 1992;1:315 23. 30. Witte JS, Ursin G, Siemiatycki J, et al. Diet and premenopausal bilateral breast cancer: a case-control study. Breast Cancer Res Treat 1997;42:243 51. 31. Augustin LS, Dal Maso L, La Vecchia C, et al. Dietary glycemic index and glycemic load, and breast cancer risk: a casecontrol study. Ann Oncol 2001;12:1533 8. 32. Muti P, Quattrin T, Grant BJ, et al. Fasting glucose is a risk factor for breast cancer: a prospective study. Cancer Epidemiol Biomarkers Prev 2002;11:1361 8. 33. Terry P, Suzuki R, Hu FB, et al. A prospective study of major dietary patterns and the risk of breast cancer. Cancer Epidemiol Biomarkers Prev 2001;10:1281 5. 34. Cho E, Spiegelman D, Hunter D, et al. Premenopausal dietary carbohydrate, glycemic index, glycemic load, and fiber in relation to risk of breast cancer. Toronto, Ontario, Canada: American Association for Cancer Research, 2003:2568. 35. Mattisson I, Wirfalt E, Johansson U, et al. Intakes of plant foods, fibre and fat and risk of breast cancer: a prospective study in the Malmo Diet and Cancer cohort. Br J Cancer 2004; 90:122 7. 36. Hu FB, Stampfer MJ, Rimm E, et al. Dietary fat and coronary heart disease: a comparison of approaches for adjusting for total energy intake and modeling repeated dietary measurements. Am J Epidemiol 1999;149:531 40. 37. Liu S, Willett WC, Stampfer MJ, et al. A prospective study of dietary glycemic load, carbohydrate intake, and risk of coronary heart disease in US women. Am J Clin Nutr 2000;71: 1455 61. 38. Liu S, Buring JE, Sesso HD, et al. A prospective study of dietary fiber intake and risk of cardiovascular disease among women. J Am Coll Cardiol 2002;39:49 56.