Glycemic Index and Glycemic Load Dietary Patterns and the Associated Risk of Breast Cancer: A Case-control Study

Habitual consumption of a high glycemic index (GI) food causes postprandial increases in the blood glucose level, resulting in high insulin demand (Jenkins et al., 2002). The GI and GL, which considers both the quantity and quality of carbohydrate intake, have been shown to be related to chronic disease and increased risk of cancer (Patel et al., 2007; Jiao et al., 2009; Bao et al., 2010), including breast cancer (Lajous et al., 2005; Gnagnarella et al., 2008; Lajous et al., 2008; Larsson et al., 2009; Belle et al., 2011; Shikany et al., 2011). Because breast


Introduction
Habitual consumption of a high glycemic index (GI) food causes postprandial increases in the blood glucose level, resulting in high insulin demand (Jenkins et al., 2002). The GI and GL, which considers both the quantity and quality of carbohydrate intake, have been shown to be related to chronic disease and increased risk of cancer (Patel et al., 2007;Jiao et al., 2009;Bao et al., 2010), including breast cancer (Lajous et al., 2005;Gnagnarella et al., 2008;Lajous et al., 2008;Larsson et al., 2009;Belle et al., 2011;Shikany et al., 2011). Because breast et al., 2007;Zhang et al., 2009;Butler et al., 2010, Zeng et al., 2013, association studies between breast cancer risk and simple GI and GL values might not explain the effects of overall diet. According to the Korea National Health and Nutrition Examination Survey (KNHANES) (Korea Health Statistics, 2009: http://knhanes.cdc.go.kr/ 2011.09), the average carbohydrate intake of Korean women was 268g, which accounted for 68.3% of the total energy intake. However, Korean diets are rich in vegetables and fruits in addition to carbohydrates. Thus, the GI and GL patterns considering all of the foods consumed may provide a more precise association with breast cancer risk than simple GI and GL values.
Reduced rank regression (RRR) in dietary pattern analysis is useful method to combine the advantages of a posteriori and a priori approaches because the pattern is extracted considering all predictor variables by maximizing the variation of the response variables (Hoffmann et al., 2004;Edefonti et al., 2009). Association studies between breast cancer risk and GI and GL patterns derived by RRR might remove the confounding effects. The effects of the GI and GL values on breast cancer were found to be inconsistent according to menopausal status and hormone receptor status (Sieri et al., 2007;Wen et al., 2009), and there has been only one study that used the GI or GL dietary pattern score with a population comprising individuals with considerably different food cultures (McCann et al., 2007). We investigated the association between breast cancer risk and the GI and GL menopausal status.

Study population
Cases were breast cancer patients admitted for surgery at the Center for Breast Cancer in the National Cancer Center Hospital in Korea between July 2007 and September 2008. Among the 415 women with breast cancer who agreed to participate in the study, patients with a previous history of cancer (n=14), the inability to complete an interview (n=2) and daily energy intakes of <600 or >3500kcal (n=1) were excluded. Controls were recruited among individuals who underwent health screening examinations at the Center for Cancer Prevention and Detection at the same hospital during same period. Among the 713 women who agreed to participate in the study, women with a history of cancer or missing information about dietary intake were excluded. The remaining 653 women were matched to the 398 cases based on 5-year age groups. In total, 357 cases and 357 controls were selected for analysis. All participants were provided an written informed consent form according to the procedures approved by the institutional review board of the National Cancer Center Hospital (IRB protocol number NCCNCS 07-083).

Data collection and dietary assessment
A trained dietitian conducted in-person interviews using a structured questionnaire covering demographics, lifestyle, and medical history. Physical activity (MET-min/ wk) was evaluated using the short form of the International Physical Activity Questionnaire (IPAQ). The validated food frequency questionnaire (FFQ) covering of 103 types of food was used to assess typical dietary intake (Ahn et al., 2007). The de-attenuated and age-, sex-, and energy-12-day dietary records ranged between 0.23 (vitamin A) and 0.64 (carbohydrates), and the median for all nutrients was 0.39. Correlations between the two FFQs were 0.45 for all nutrient intake levels and 0.39 for nutrient densities. All participants were asked about the average frequency previous year. Three portion sizes (small, medium, and large) and 9 categories of frequency (never or rarely, once a month, 2 or 3 times a month, once or twice a week, 3 or 4 times a week, 5 or 6 times a week, once a day, twice a values of each food were obtained from international tables (Foster-Powell et al., 2002). GI values for foods not found in the tables were estimated based on the most similar food according to physical and chemical factors. The GI values were matched to each of the individual foods, multiplied by the quantity of carbohydrates and divided by total amount of carbohydrate consumed per day; the overall GI was calculated by summing up all of these values. The GL represents both the quality and quantity of carbohydrate intake. The overall GL was calculated by summing up all GI values multiplied by the amount of carbohydrate of each food and dividing the values by 100. The estrogen receptor (ER) and progesterone receptor (PR) status evaluations were performed on tissue sections by immunohistochemistry (Ventana Medical Systems, Tucson, AZ). Any focal positivity, including weakly positive expression, was recorded (Regitnig et al., 2002).

Statistical analysis
Statistical analysis was performed using the SAS version 9.1 statistical package (SAS Institute Inc, Cary, NC). Thirty-nine food groups covering 410 kinds of food contained in 103 FFQ items were used to derive the dietary pattern by reduced rank regression (RRR) using PROC PLS with the RRR method option (Hoffmann et al., 2004). The GI and GL, which were adjusted for total energy intake using the nutrient residual model (Willett and Stampfer, 1986), were used as response variables separately, and the mass of the 39 food groups in grams was considered as predictor variables. Each factor score for GI and GL was categorized into tertiles for further analysis. Trends in the characteristics of the study population with respect using a generalized linear model. Odds ratios (ORs) and calculated across the tertiles of GI or GL using logistic regression after controlling for known risk factors such (log-transformed), age (continuous), body mass index menopausal status (premenopausal, postmenopausal), alcohol consumption (never, ever), smoking (never, ever), parity (yes, no), family history of breast cancer (yes, no), (log-transformed) and postmenopausal hormone use for postmenopausal women. To test the tertile trend, a median value was assigned to each tertile of the GI or GL pattern scores as a continuous variable.

Results
Among the 39 food groups, the factor scores of shown in Table 1. One dietary pattern for each response of extracted factors cannot be greater than the number of response variables in the RRR method. The GI and GL patterns explained 7.5% and 9.0% of the variation in the predictor variables, respectively, and 77.4% and 96.1% of the variation in the response variables, respectively. Grain intake explained most of the variance in the factor scores in both the GI and GL patterns, and green/yellow vegetables  for GI and red meat for GL had negative loadings. The characteristics of the study participants according to the tertiles of GI and GL pattern scores are presented in Table  2. Among the breast cancer cases, smoking and higher education declined across the tertiles of both the GI and GL pattern scores (p for trend<0.001). Cases with higher GL pattern scores were older and reported a lower alcohol intake. The ORs and 95% CIs of breast cancer risk were analyzed across the tertiles of GI and GL pattern scores (Table 3). Women in the highest tertiles of GI and GL pattern scores compared to those in the lowest tertiles   (2755) 2885 (3681) 2858 (3099) 0.918 2195 (2137) 2086 (2290) 2242 (3126) (2957) 2846 (3743) 2716 (2836) 0.438 1997 (1856) 2096 (2338) 2281 (3099) (Table 4). The risk of breast cancer was increased with increasing tertile of the GI pattern score only in women who had ER+ or PR+ cancer, while the risk of breast cancer was increased with increasing tertiles of the GL pattern score in all subgroups. Furthermore, the results differed based on hormone receptor status. The GI and GL pattern scores were positively associated with the risk of breast cancer in premenopausal women with ER+ or PR+ cancer, whereas the risk of breast cancer, although was elevated with increasing GI or GL pattern scores in all subgroups of postmenopausal women. Patients with ER+/ PR+ cancer and ER-/PR-cancer showed trends similar to women with ER+ or PR + cancer and ER-or PR-cancer, respectively (data not shown).

Discussion
The effects of the GI and GL dietary patterns determined by RRR on breast cancer risk were investigated in the present study. Breast cancer incidence risk was elevated with increasing factor loadings of the GI and GL patterns.
Breast cancer risk was increased across the tertiles of GI and GL patterns in both premenopausal and postmenopausal women, but the increase was not GL pattern among premenopausal women. Breast cancer risk was highly associated with postmenopausal status and a high GL pattern score. Previous studies using simple GI and GL values have obtained inconsistent results. GI and GL had no association with breast cancer risk among  French postmenopausal women in a prospective cohort analysis (Lajous et al., 2008), but breast cancer risk was positively associated with both GI and GL among premenopausal women in Italian prospective cohort study (Sieri et al., 2007). Wen et al. (2009) reported that premenopausal Chinese women with high GLs had a higher risk of breast cancer. A meta-analysis revealed that only the GL had a positive association with breast cancer risk (Gnagnarella et al., 2008). Another metabased on menopausal status because the heterogeneity, was observed in both premenopausal and postmenopausal women (Mulholland et al., 2008). Korean meals are comprised of steam-cooked rice with various side dishes that are usually composed of large amounts of vegetables. It was hypothesized that the GI and GL pattern scores might have different effects on breast cancer risk because the high intake of grain typically accompanies a high intake of vegetables and other foods that contain preventive nutrients. Thus, the GI and GL patterns were extracted using the RRR method considering all food groups. The majority of the factor score variation in the GI and GL pattern was explained by grain intake in the present study. This result might be related to the high amount of grain intake by Koreans. The average amount of grain intake by Korean women was 260.4g and constituted 68.6% of carbohydrate intake in 2008 according to the KNHANES (Korea Health Statistics, 2009). White rice, the most frequently consumed food in Korea, accounted for 70% of carbohydrate intake from grains. Thus, the GI and GL pattern scores were mostly affected by white rice intake in the present study. Similar to the results for simple GI and GL values, high factor scores for the GI and GL patterns were associated with an increased risk of breast cancer. There was a previous study that compared the GI and GL dietary patterns derived by RRR and the simple GI and GL values in the estimation of breast cancer risk (McCann et al., 2007). The GI and GL patterns were not associated with breast cancer risk in either premenopausal or postmenopausal women, and simple the GI and GL values yielded results similar to those of the RRR-derived patterns, suggesting that the GI and GL patterns derived by RRR do not provide much new information compared to the GI and GL values. Breast cancer risk decreased with higher combined pattern scores of GI and GL among postmenopausal women and increased with high GL pattern scores in premenopausal women in the study. Although the effects of the GI and GL pattern scores derived by RRR on the risk of breast cancer simple dietary GI and GL values, the pattern scores might give more precise results by removing the confounding effects of foods and because the GI and GL pattern scores affect breast cancer risk differently because food items vary in different countries.
Breast cancer risk was increased across the tertiles of the GI and GL pattern in all hormone receptor status subgroups in postmenopausal women, although this while the risk was positively associated with only ER+ and PR+ patients among premenopausal women. The underlying mechanism might be related to insulin-like growth factor-I (IGF-I). A high intake of carbohydrates increases the risk of breast cancer via elevation of the IGF-I level, which promotes cell proliferation and inhibits cell death (Kaaks, 1996), while long-term low carbohydrate intake reduces the postprandial rise in gut hormones and insulin through suppression of free fatty acids (Jenkins et al., 2002). IGF-I can synergistically increase cell growth and proliferation via the activation of ER (Yee and Lee, 2000;Mawson et al., 2005;Lanzino et al., 2008;Richardson et al., 2011, Wang et al., 2012. Thus, large amounts of high GI and GL food intake can increase breast cancer risk, especially for ER+ premenopausal women. However, information was limited in previous studies. The GI and GL were positively associated with ER+/PR-breast cancer risk in a Swedish cohort, though there was no subgroup analysis based on menopausal status (Larsson et al., 2009), and two other studies status using only postmenopausal women. A marginally cancer was found in a Danish cohort (Nielsen et al., 2005), and dietary GL was associated with increased risk of ERbreast cancer in a French cohort (Lajous et al., 2008). An elevated IGF-I concentration was found to be related to increased breast cancer risk in both premenopausal and postmenopausal women, but a positive association was shown only for ER+ breast cancer patients (Key et al., 2010). Thus, it seems that the association between breast cancer and GI or GL differs based on menopausal status and hormone receptor status, but further studies are needed Biases might have been introduced because the present study was conducted with a case-control design. Cases were recruited among patients who were admitted for breast cancer surgery, while the controls were selected from among individuals who underwent health screening examinations. Thus, even though the controls were agematched to the cases, the control population might have had a healthier lifestyle, especially with respect to the selection of food items. Furthermore, the small sample sizes in the subgroup analysis limited the statistical power. Nevertheless, our data are unique because we focused on a population with a high carbohydrate intake and a high intake of various vegetables to analyze the relationship between the breast cancer risk and the GI or GL pattern score.
In summary, the GI and GL patterns derived by RRR were positively associated with breast cancer. The association was shown only in women with ER+ or ER+/PR+ cancer among premenopausal women and in all subgroups of hormone receptor status among postmenopausal women. Postmenopausal women with high GL pattern scores showed a strong association with breast cancer risk.