Adolescent and Early Adulthood Dietary Carbohydrate Quantity and Quality in Relation to Breast Cancer Risk

A higher incidence of breast cancer has been reported in individuals with type II diabetes (1). Among several possible underlying mechanisms, high circulating levels of insulin and insulin-like growth factor I (IGF-I) may play important roles in tumor growth and progression and may increase risk of breast cancer (2–5). IGF-I and estrogen may synergistically stimulate estrogen receptors and cellular proliferation (6).

Several dietary factors contribute to variations in levels of circulating insulin and IGF-I (7, 8). The quality and quantity of ingested carbohydrate, expressed as glycemic index (GI) and glycemic load (GL), respectively, are the major determinants of postprandial blood glucose levels and hence circulating insulin levels (9, 10). The GI is a ranking system for the carbohydrate content of foods based on their postprandial glycemic effects and is a measure of carbohydrate quality. The GL combines the total amounts of carbohydrate usually consumed and its GI values and is a combined measure of carbohydrate quality and quantity that most strongly relates to postprandial insulin (10). Given that protein and fat may also stimulate insulin secretion (11), dietary insulin index, and insulin load scores may more directly address the insulin hypothesis by combining postprandial insulin responses for individual food items, including those with low or no carbohydrate content (11). Although the association between quality and quantity of carbohydrate and breast cancers were not significant in most prospective cohort studies (12–19), a recent meta-analysis of 10 cohort studies found that a diet high in GI, but not GL, was positively associated with breast cancer risk (20). Studies regarding the impact of dietary insulin index and insulin load on breast cancer risk, however, are lacking. Although exposures in childhood and early adulthood may be critical in subsequent risk of cancer (21–23), limited attention has been paid to assess adolescent or early adulthood dietary intake in relation to breast cancer and most of the existing literature is based on diet during midlife and later. However, high intake of refined carbohydrate and added sugar with high GI are reported in adolescence and young adults (24–26); their role in incidence of breast cancer is unclear.

In previous analyses of the Nurses' Health Study II (NHSII; refs. 12, 13), dietary carbohydrate, GI and GL were not associated with risk of premenopausal breast cancer. The current analyses included twelve additional years of follow-up and almost four times the number of cases compared with our initial report. Therefore, we were able to examine quantity and quality of carbohydrate intakes as well as insulin load and dietary insulin index scores in adolescence and early adulthood in relation to breast cancers diagnosed before or after menopause. Furthermore, we investigated the associations between these scores and breast cancer by hormone receptor status.

The NHSII is an ongoing cohort study following 116,430 female registered nurses ages 25 to 42 years at enrollment in 1989 from 14 U.S. states. Information on dietary intake was first obtained on 1991 food-frequency questionnaire (FFQ), this served as baseline for starting follow-up. From the 97,813 women who returned the 1991 FFQ, we excluded women who had an implausible total energy intake (<600 or >3,500 kcal/day) or left more than 70 items blank, who were postmenopausal in 1991, had reported a prior diagnosis of cancer (except nonmelanoma skin cancer) before returning the 1991 questionnaire, or had missing information on age. After exclusions, data from 90,534 women were available for the analysis. The follow-up rate was 95% of total potential person-years of follow-up through 2011.

In 1997, participants were asked about their willingness to complete a supplemental food frequency questionnaire about diet during high school (HS-FFQ). From 64,380 women (55% of the entire cohort) who indicated willingness to complete, 47,355 of them returned the HS-FFQ in 1998. There were minimal differences in baseline demographic characteristics and breast cancer rate between participants who completed the HS-FFQ compared with women who did not provide information on high school diet (13). We excluded women who had any cancer except nonmelanoma skin cancer before 1998, or reported implausible daily caloric intake (<600 or ≥5,000 Kcal) or had missing information on age. After exclusion, data from 44,263 women were available for the present analysis.

This study was approved by the Human Subjects Committee at Brigham and Women's Hospital and Harvard T.H Chan School of Public Health (Boston, MA).

Dietary information during adulthood was evaluated via validated semiquantitative FFQ with approximately 130 items about usual dietary intake and alcohol consumption during the past year (27), which was sent to participants in 1991 and every 4 years thereafter. Dietary intakes in adolescence were obtained from a semiquantitative 124-item HS-FFQ that included food items typically consumed between 1960 and 1980 when they were in high school. To examine the reproducibility of the HS-FFQ, we readministered it to a random sample of 333 NHSII participants in January, 2003; the mean intraclass correlation coefficient was 0.65 (range, 0.50–0.77) for nutrient intakes and 0.58 for carbohydrate intake (28). The reproducibility of the HS-FFQ was also examined by comparing responses to HS-FFQ with 3 24-hour recalls with 10-year interval among 80 young women ages 23 to 29 years at the time of collecting second questionnaire; the mean of corrected correlation coefficients for energy-adjusted nutrient intakes was 0.45 (range, 0.16–0.68; ref. 29). For validity, adolescent dietary intakes reported by 272 NHSII participants using the HS-FFQ were compared with intakes of these participants reported by their mothers; the mean of correlations was 0.40 (range, 0.13–0.59) for nutrients, 0.33 for carbohydrate, 0.43 for GI, and 0.38 for GL (28).

Nutrient intakes were computed by multiplying the frequency of consumption of each unit of food or beverage by the nutrient content of the specified portions and then summing the contributions from all items. The U.S. Department of Agriculture, food manufacturers, and independent academic sources were used to calculate the nutrient intakes (30–32). The food composition database was updated every 4 years to account for changes in the food supply. To calculate the percentage of energy contributed by carbohydrates and other macronutrients, we divided energy intake from that nutrient by total energy intake. GI, GL, insulin load, and dietary insulin index scores were energy-adjusted using the residual method from the regression of these intakes as dependent variable on total energy intake as independent variable (33, 34).

Insulin index values for each food were obtained from either published estimates (31 foods; refs. 11, 35) or direct testing of U.S. food items (73 foods) at the University of Sydney (New South Wales, Australia). The method was described in detail elsewhere (11). Briefly, each person consumed a 1,000 kJ of test foods and the reference food (glucose) on separate days and serum insulin was measured every 15 minutes for 2 hours after consumption, then the area under the 120-minute insulin response curve for 1,000-kJ test food was divided by the area under the 120-minute insulin response curve for 1,000-kJ glucose. Dietary insulin load was calculated by multiplying the insulin index value of each food by the energy content of food, then, summing values for all food items reported [Σ(food insulin index × energy content of food (kcal/serving) × frequency of intake (serving/day)]. Each unit of dietary insulin load indicates the equivalent amount of insulin produced by 1 kcal of glucose. The dietary insulin index was calculated by dividing the dietary insulin load by the total energy intake (36).

GI was calculated from a published database (10) or values derived from direct testing of food items by Prof. David J. Jenkins at Nutrition Center of University of Toronto (Toronto, Ontario, Canada). The method was described in detail elsewhere (10). Briefly, dietary GI was measured by dividing the area under the 120-minute incremental blood glucose curve by ingestion of 50 g carbohydrate from test food by the area under the 120-minute incremental blood glucose curve by ingesting the same amount of glucose as a reference food. The average dietary GL was obtained by summing the products of carbohydrate intake for each food by its frequency of intake and dietary GI (37): |${\rm GL}_{{\rm ave}} = \sum\nolimits_{\alpha = 1}^n {GI\alpha \times CHO\alpha \times {\rm frequency}\alpha}$|⁠, where n is the number of foods consumed, |$GI\alpha$| is the glycemic index for food |$\alpha, \,CHO\alpha$| is the carbohydrate content per serving of food α, and |${\rm frequency}\alpha$| is the consumption frequency of one serving of food α during the past 12 months. The average dietary GI was calculated by dividing the average GL by the total amount of carbohydrate intake (38).

Newly diagnosed invasive breast cancers were identified via biennial NHSII questionnaires. We asked the participant (or next of kin for those who had died) who reported breast cancer for confirmation of the diagnosis and for permission to obtain relevant hospital records and pathology reports. Because of 99% of the self-reported diagnosis of breast cancer were confirmed by pathology report, diagnoses confirmed by participants with missing medical record information (n = 344) were included in the analysis. Information on estrogen and progesterone receptor (ER, PR) status of the breast cancer was obtained from pathology reports. Deaths in this cohort were reported through family members and the postal service in response to the follow-up questionnaires or identified through annual review of the National Death Index.

We collected data on potential risk factors for breast cancer from the biennial NHSII questionnaires including age, height, weight, family history of breast cancer, history of benign breast disease, smoking, race, menopausal status, age at menarche, postmenopausal hormone use, and oral contraceptive use. All variables except race, height, and age at menarche were updated to the most recent information, whenever available. Women were considered premenopausal if they still had periods or had hysterectomy with at least one ovary remaining and were younger than 46 years for smokers or younger than 48 years for nonsmokers. Women were considered postmenopausal if they reported natural menopause or had undergone bilateral oophorectomy. We defined women of unknown menopausal status or who had hysterectomy without bilateral oophorectomy as postmenopausal if they were 54 years or older for smokers or 56 years or older for nonsmokers (39).

Body mass index (BMI) at age 18 was obtained from the 1989 questionnaire and was used as a proxy for BMI during high school. Weight change from age 18 was calculated by taking the difference between current weight and recalled weight at age 18. Data on alcohol consumption during adolescence were obtained from the 1989 NHSII questionnaire.

We conducted the analyses in three groups: among all women, premenopausal women, and postmenopausal women. Follow-up time began with return of the baseline questionnaire in 1991 for early adulthood dietary intake and with return of HS-FFQ in 1998 for adolescent dietary intake, until either June 2011, the date of breast cancer or any other cancers diagnosis except nonmelanoma skin cancer, or death, whichever came first. In premenopausal group, only premenopausal women were included in analysis; therefore, we stopped follow-up after reporting postmenopausal or uncertain menopausal status in this group. For the postmenopausal group, women started contributing person-time from the first 2-year cycle in which they reported postmenopausal status. Cox proportional hazards models, stratified by age in months and 2-year follow-up cycle, were used to estimate relative risks (RR) and 95% confidence intervals (95% CI). Multivariable models also simultaneously adjusted for race, family history of breast cancer in mother or sisters, history of benign breast disease, smoking, height, age at menarche, parity and age at first birth, oral contraceptive use, menopausal status, postmenopausal hormone use, BMI at age 18 years, weight change since age 18 years, age at menopause, and early adulthood intakes of alcohol, and energy. For adolescent dietary intake and breast cancer risk, multivariable models were additionally adjusted for adolescent alcohol intake, and adolescent energy intake (instead of early adulthood energy intake). Tests for linear trend were conducted by modeling the median value for each quintile and treating this as a continuous variable in the regression model. We replaced missing covariate data, which comprised 5.5% of total person-years for oral contraceptive use and less than 5% of total person-years for BMI at age 18 years, smoking, height, age at menarche, age at menopause, parity, and age at first birth, with the carry forward method for continuous variables and missing indicator method for categorical variables (40). To evaluate the effect of dietary intake on breast carcinogenesis over an extended period of time, for sensitivity analyses, we also calculated premenopausal cumulative averaged of GL, GI, insulin index, and insulin load using the 1991, 1995, 1999, 2003, and 2007 dietary data, and stopped updating when a woman reached menopause. Furthermore, we calculated mean of adolescent and early adulthood GI, GL, insulin index, and insulin load. To examine differential associations of dietary intake with breast cancer risk by hormone receptor status, we used Cox proportional cause-specific hazards regression model with a duplication method for competing risk data. This method permits estimation of separate associations of GI for tumors that are both estrogen and progesterone receptor positive (ER⁺/PR⁺) and negative (ER⁻/PR⁻), and has been used to assess whether a risk factor has statistically different regression coefficients for different tumor subtypes (41). We examined effect modification of the association between GL, GI, insulin index, and insulin load scores and breast cancer risk by BMI at age 18 years. A cross-product interaction term between BMI at age 18 and scores of GL, GI, insulin index, and insulin load was included in the multivariable model. P values for tests for interactions were derived using a likelihood ratio test. All P and 95% CI values were two-sided and all analyses were performed using SAS version 9.3 (SAS Institute Inc).

During 1,725,295 person-years of follow-up of 90,534 women, 2,833 women were diagnosed with invasive breast carcinoma, (1,659 premenopausal breast cancers, 875 postmenopausal breast cancers, and 299 cases with uncertain menopausal status). Among 44,263 women with data on adolescent carbohydrate intake, 1,118 women were diagnosed with invasive breast cancer (544 premenopausal, 465 postmenopausal, and 109 uncertain menopausal status) from 1998 to 2011. The age range of the participants at baseline in 1991 was 27 to 44 years (mean 36.4 ± 4.6 years). Compared with women who had a lower GI diet, women with a diet higher in GI were more likely to be younger, to have a lower dietary fiber intake as well as less likely to drink alcohol, to be nulliparous, and to have earlier age at menarche (Table 1).

Among all women, higher early adulthood intake of carbohydrate was associated with lower risk of breast cancer (comparing the highest vs. lowest quintile, RR = 0.88; 95% CI, = 0.78–0.99; P_trend = 0.05). This association was not significant after additional adjustment for dietary fiber (RR for highest vs. lowest quintile = 0.92; 95% CI, = 0.81–1.04; P_trend = 0.26) or red meat (RR for highest vs. lowest quintile = 0.92; 95% CI, = 0.80–1.05; P_trend = 0.30). Among all women, higher GI in early adulthood was not significantly associated with risk of breast cancer (comparing the highest vs. lowest quintile, RR = 1.03; 95% CI, = 0.91–1.16; P_trend = 0.66; Table 2). Similar association was observed among either premenopausal or postmenopausal women. Furthermore, GL, dietary insulin index, and insulin load were not significant predictors of either overall breast cancer or breast cancers among premenopausal or postmenopausal women (Table 2). Results did not differ between age-adjusted and multivariable adjusted models. Additional adjustment for red meat, fruit and vegetables, or fiber intake did not materially change the results (data not shown).

To assess the effects of breast carcinogenesis over an extended period of time, we also calculated premenopausal cumulative average. Similar associations were observed. In multivariable-adjusted model, women in the highest quintile of premenopausal cumulative average GI had an RR of 1.08 (95% CI, 0.96–1.22; P_trend = 0.50) compared with women in the lowest quintile. RRs were 0.96 (95% CI, 0.85–1.08; P_trend = 0.46) for premenopausal cumulative average of GL in the highest quintile compared with lowest quintile. Furthermore, premenopausal cumulative average of either dietary insulin index or insulin load was not associated with breast cancer risk (comparing the highest vs. lowest quintile, RR for dietary insulin index = 1.00; 95% CI, = 0.89–1.14; P_trend = 0.82; and RR for insulin load = 1.01; 95% CI, = 0.90–1.15; P_trend = 0.82).

Adolescent carbohydrate, GI, GL, insulin index, and insulin load was only weakly correlated with early adult intake (1991). The intraclass correlation was 0.11 (0.10–0.12) for carbohydrate, 0.19 (0.18–0.20) for GI, and 0.16 (0.15–0.17) for insulin index. The estimated coefficient of within-subject variance was 0.14 for carbohydrate, 0.05 for GI, and 0.08 for insulin index. Associations between adolescent carbohydrates, GL, GI, insulin index, and insulin load and breast cancer risk are shown in Table 3. Adolescent intake of carbohydrate was not associated with lower risk of breast cancer. A diet high in GI in adolescence also was not associated with a higher risk of breast cancer (for highest vs. lowest quintiles, multivariable RR, 1.14; 95% CI, 0.95–1.38, P_trend = 0.58). This association was not significant in either premenopausal or postmenopausal breast cancer (Table 3). Similarly, nonsignificant associations were observed for adolescent GL, insulin index, and insulin load and breast cancer risk. Additional adjustment for adult GI, GL, insulin index, or insulin load did not change the results (data not shown). Further adjustment for red meat, fruit and vegetables, or fiber intake did not materially change the results (data not shown). Among women with both early adulthood and adolescent dietary data (n = 41,092), we calculated the average of indices at both times. No significant association was observed (data not shown).

Table 4 presents the associations between adolescent and early adulthood GI scores and breast cancer according to hormone receptor status; data are presented for tumors with both ER⁺/PR⁺ and ER⁻/PR⁻. We did not observe associations for adolescent and early adulthood GI scores and breast cancer risk by hormone receptor status, and there was no significant heterogeneity. Furthermore, no significant association or significant heterogeneity was observed for GL and breast cancer risk (data not shown).

In our previous evaluation of quality and quantity of carbohydrate intake, the associations differed by body weight (12). Therefore, we also examined whether these dietary associations with breast cancer risk differed by BMI at age 18 (<25/≥25 kg/m²) (Table 5). Although higher GI score during adolescence was associated with higher risk of breast cancer among women with BMI 25 or higher during adolescence, the interaction was not significant. Furthermore, no significant interaction was observed between BMI at age 18 years and GL, insulin index, or insulin load in adolescence or early adulthood (Table 5).

In this large prospective analysis, we observed no overall association between quality and quantity of carbohydrate intake during adolescence or early adulthood and breast cancer risk. Furthermore, we found no evidence that a diet high in insulin load or insulin index is related to breast cancer risk.

Our results are largely consistent with those published earlier for the NHSII (12, 13) and do not support a positive association between dietary GI or GL and breast cancer risk. Previous cohort studies have produced mixed results. In a recent meta-analysis of 10 prospective cohort studies (20), there was no significant association between dietary GL and risk of breast cancer (RR, 1.04; 95% CI, 0.95–1.15). However, higher dietary GI was associated with 8% higher risk of breast cancer (RR, 1.08; 95% CI, 1.02–1.14). The foods with low GI have other properties that may increase or decrease risk of breast cancer. In our study, women with high GI diet were more likely to have higher intake of red meat and lower intake of fiber. Diets high in red meat were associated positively with breast cancer risk in the current study population (42). However, additional adjustment for red meat, fruit and vegetable, or fiber intake did not change the associations. Similarly, diets low in carbohydrate can be high in red meat and low in fiber, which have been shown to increase risk of breast cancer (42, 43) and no association between carbohydrate and breast cancer was observed after additional adjustment for red meat or fiber intake.

Although there was a positive association between hyperinsulinemia and breast cancer in case–control studies nested within the Nurses' Health Study and NHSII cohorts (44), we observed no association between dietary insulin index and insulin load and risk of breast cancer. Similarly, dietary insulin index and insulin load were not associated with risk of other cancers (45–47). On the other hand, in a recent meta-analysis of 6 prospective studies (48), compared with women with lowest insulin levels, those with higher insulin levels were not at higher risk of breast cancer (pooled RR of breast cancer, 1.08; 95% CI, 0.66–1.78).

Potential limitations need to be considered. Because the participants were predominantly white, educated U.S. adults, generalizability to other race or ethnic groups is questionable; however, it is unlikely that the biology underlying this association differs by race or ethnicity. Assessment of dietary intake using FFQ is prone to random measurement error caused by within-person variation. However, we found similar associations using cumulative averages of repeated dietary assessments before menopause. In addition, high dietary GI measured in the same population with the same dietary assessment has been associated with an increased risk of type II diabetes (49). Women recalled their diet during adolescence when they were 33 to 52 years old. Some degree of measurement error is inevitably present. However, the associations were largely independent of adult diet, and evidence of validity came from the comparison of their dietary reports with the information provided 4 years later or from dietary intake reported by their mother (28, 29). Residual confounding is always of concern in any observational studies. Comprehensive adjustment for many potential confounders minimized residual confounding, although we could not rule out the influence of unmeasured or unknown confounders. We could not exclude the possibility of limited power to detect differences in risk in subgroups, particularly for adolescent diet.

Our study has several strengths. To evaluate the importance of timing, we assessed the association between quality and quantity of carbohydrate as well as insulin index and insulin load during specific life periods (adolescence, early adulthood, and cumulative average of premenopausal period). The large sample size and length of follow-up made it possible to evaluate the associations by menopausal and tumor hormone receptor status. Assessing adolescent and early adulthood dietary intake before breast cancer diagnosis minimized recall bias.

In summary, our results suggest that diets high in GI, GL, insulin index, and insulin load during adolescence or early adulthood were not associated with an increased risk of breast cancer in this cohort study. As the data on diet during childhood and later breast cancer risk remain limited, further studies are needed to better clarify the influence of timing of dietary exposures in relation to risk of breast cancer.

No potential conflicts of interest were disclosed.

The study sponsors were not involved in the study design and collection, analysis and interpretation of data, or the writing of the article or the decision to submit it for publication. The authors were independent from study sponsors.

Conception and design: M.S. Farvid, W.C. Willett

Development of methodology: M.S. Farvid, E. Cho, W.C. Willett

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.H. Eliassen, W.C. Willett

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.S. Farvid, A.H. Eliassen, E. Cho, W.C. Willett

Writing, review, and/or revision of the manuscript: M.S. Farvid, A.H. Eliassen, E. Cho, W. Chen, W.C. Willett

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.S. Farvid, W.C. Willett

Study supervision: A.H. Eliassen, W.C. Willett

We would like to thank the participants and staff of the NHS II for their valuable contributions as well as the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, WY.

The study was supported by the NIH grants (R01 CA050385, UM1 CA176726) and a grant from The Breast Cancer Research Foundation. M.S. Farvid was supported by the Japan Pharmaceutical Manufacturers Association.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.