Abstract
Background
High-glycemic-load diets may increase colorectal cancer risk through hyperinsulinemic effects.
Objective
We analyzed data for 191 004 participants in the Multiethnic Cohort Study to determine the risk of colorectal cancer associated with glycemic load (GL), carbohydrate, and sucrose and to ascertain whether this risk was modified by sex and ethnicity.
Design
During 8 y of follow-up, 2379 incident cases of colorectal adenocarcinoma occurred. We used baseline quantitative food-frequency questionnaire data to assess usual dietary intake over the preceding year. Using Cox regression, we calculated adjusted relative risks (RRs) and 95% CIs for colorectal cancer associated with quintiles of GL, carbohydrate, and sucrose.
Results
For both men and women in this cohort, white rice was the major contributor to GL. In multivariate models, RRs for colorectal cancer decreased significantly with increasing GL in women (RR for the highest quintile versus the lowest: 0.75; 95% CI: 0.57, 0.97; P for trend = 0.02) but not in men (RR: 1.15; 95% CI: 0.89, 1.48; P for trend = 0.19). Results for carbohydrate and sucrose were similar. The inverse association with GL was found in women of all ethnic groups (P for interaction = 0.58). In men, an interaction was found between ethnicity and GL (P < 0.01): white men had a positive association with increasing GL (RR: 1.69; 95% CI: 0.98, 2.92; P for trend < 0.01), but men of other ethnic groups did not.
Conclusion
GL and carbohydrate intake appear to protect against colorectal cancer in women in the Multiethnic Cohort, perhaps because a major source of GL is white rice.
Introduction
A Western diet, characterized by higher intakes of meats, fat, and refined carbohydrates and lower intakes of fruit, vegetables, whole grains, and legumes, has been associated with a greater risk of colorectal cancer (1, 2). This dietary pattern may induce high glucose concentrations and an associated elevated insulin response, which could ultimately lead to insulin resistance. Other factors predisposing to insulin resistance are obesity, excess energy intake, and a sedentary lifestyle, all of which also increase the risk of colorectal cancer (3, 4). Thus, there is growing recognition that hyperinsulinemia and insulin resistance may promote colorectal cancer.
Insulin stimulates metabolic pathways that lead to an increase in insulin-like growth factor-I (IGF-I). Both insulin and IGF-I promote cell division and inhibit apoptosis in healthy and cancerous colon epithelial cells (5, 6). Rapidly digested carbohydrates cause spikes in blood glucose that are followed by a heightened insulin response (7). Glycemic index (GI) is a measure of this response that is used to rank carbohydrate-rich foods relative to either white bread or glucose (8). Glycemic load (GL) is the product of the GI and the amount of carbohydrate (in g) in a serving of a food (9).
Epidemiologic evidence for associations of GI, GL, and sucrose and carbohydrate intakes with colorectal cancer has been mixed. Three case-control studies found higher risks from sucrose intake (10), sucrose and GI (11), or GL (12), and one study (13) found sugar to be protective. Cohort studies have found either no association of carbohydrate, GL, or GI (or all 3) with colorectal cancer (14, 15) or a greater risk due to GI and GL in obese women (16), GL in women overall (17), and GL, carbohydrate, and sugar in men (18). Two cohort studies of colorectal adenomas found no association of GI, GL, or carbohydrate intake with distal colorectal adenomas in women (19) and a protective effect of carbohydrates and GL on colorectal adenomas in men (20).
The preponderance of evidence from epidemiologic studies assessing associations of carbohydrate intake and GL with colorectal cancer supports either no association or a greater risk, rather than a protective effect. GL and sucrose and carbohydrate intakes are purportedly linked to colorectal cancer because insulin resistance and associated complications (elevated fasting glucose, insulin, IGF-I, and free fatty acid concentrations) are implicated in colorectal carcinogenesis (21). We hypothesized that GL and carbohydrate and sucrose intakes would be risk factors for colorectal cancer in a large multiethnic cohort and that the extent of risk may vary by ethnic group.
Subjects and Methods
Study population
The Multiethnic Cohort (MEC) study includes >215 000 participants who were 45–75 old in 1993, residing in Hawaii or Southern California (generally, Los Angeles County), predominantly of 5 ethnic groups: African American, white, Latino, Native Hawaiian, or Japanese American. The cohort study was designed to examine the association of diet, lifestyle, and genetics with the incidence of various types of cancer and other chronic diseases (22) and to compare effects across ethnic groups. Between 1993 and 1996, participants completed a 26-page mailed self-administered survey instrument that included a comprehensive quantitative food frequency questionnaire (QFFQ) and that assessed demographics, medical history, use of medications and vitamin supplements, family history of common cancers, lifestyle factors such as physical activity and smoking status, and self-reported height and weight. Women were also asked about reproductive history and use of hormone replacement therapy.
These analyses excluded cohort members outside the 5 major ethnic groups (n = 13 992) and persons whose diets, based on energy and macronutrient intakes from the baseline QFFQ, were deemed implausible (n = 8264). We considered a diet implausible if a person's energy intake or his or her fat, protein, or carbohydrate intake was >3.5 modified SDs from the mean (23, 24). Finally, subjects with a diagnosis of colorectal cancer before the study, identified by self-report or by registry linkages (n = 2560), were excluded. The final analysis included 191 004 cohort members—85 898 men and 105 106 women.
From its inception in 1993 the MEC has been followed for cancer incidence and mortality with the use of computerized linkage to the cancer registries and death certificate files in Hawaii and California and the National Death Index. Incident colorectal cancer cases were identified by record linkage to the Hawaii Tumor Registry, the Cancer Surveillance Program for Los Angeles County, and the California State Cancer Registry. All 3 registries are members of the Surveillance, Epidemiology and End Results Program (known as the SEER Program) of the National Cancer Institute. Case ascertainment was complete through December 31, 2002. Information was available on the anatomical location and histologic type of the tumor and the stage of the cancer. The identification of cases in the present study was limited to patients diagnosed with invasive adenocarcinoma of the large bowel (2379 cases). Colorectal cancer patients who had other invasive tumors of the large bowel or who were diagnosed with carcinoma in situ were not included as cases. In all there were 1782 colon and 578 rectal cancer cases and 19 cases with tumors at both sites.
All subjects gave written informed consent. This investigation was approved by the institutional review boards of the University of Hawaii and the University of Southern California.
dietary data
The QFFQ was developed specifically for the study population and was based on 3-d measured food records kept by ≈60 men and ≈60 women aged 45–75 y from each ethnic group. The records were used to identify a minimum list of foods that contributed ≥85% of the intake of fat, dietary fiber, vitamin A, carotenoids, and vitamin C in each of the 5 ethnic groups. In addition to these foods, traditional foods of each ethnic group were included in the QFFQ, irrespective of their contribution to nutrient intake. A calibration substudy was conducted, and it showed acceptable correspondence between the QFFQ and multiple 24-h recalls for the sex and ethnicity groups being studied; the agreement between the 2 instruments improved after adjustment for energy (25).
The QFFQ asked about consumption of >180 food items during the previous year. There were 8 frequency categories for foods and 9 for beverages, 3 choices of portion size for foods (many represented by photographs), and 4 choices for beverages. Daily consumption (in g) was computed for each food item on the basis of frequency and portion size choices; these values were converted to nutrients by application of a study-specific food composition table (FCT) developed and maintained at the Cancer Research Center of Hawai' i. Because QFFQ food items included multiple foods, FCT values were created for the item as an average of the FCT values for its constituent foods, weighted by frequency of consumption in the calibration study. GI values were assigned to each constituent food by using published values (9, 26–28) and a scale in which glucose = 100. When a direct match could not be found, a GI value was imputed from similar foods. The GL for each food was calculated by using available carbohydrate as GL = GI × (total carbohydrate – fiber)/100. Data on the sucrose content of most foods were taken from a report by the US Department of Agriculture (29). Some of the sucrose values were estimated from British or Asian databases or were calculated from the ingredients in recipes (30).
Statistical analyses
To calculate relative risks (RRs), we used Cox proportional hazards models with age as the time metric. Person-times were calculated from the date of entry into the cohort to either the date of colorectal cancer diagnosis, death, or December 31, 2002, whichever came first; subjects with no colorectal cancer event were considered to be censored. For the 1113 subjects who were slightly <45 y old when they completed the baseline questionnaire, we calculated entry date from the day they turned 45 y old. We found no evidence for any violation of the proportional hazards assumption, according to tests of the Schoenfeld residuals (31).
GL, carbohydrate, and sucrose exposures were adjusted for energy intake by using the residual method (32), because we (25) and others (33) found that energy-adjusted values were better reported; however, the analysis was repeated for absolute intakes, and the results were similar. Four dummy variables were created, separately for each sex, to represent quintiles of consumption for each of the 3 dietary variables. Trend variables assigned sex-specific median values within the quintiles were used to test for dose-response relations. All models included the adjustment variables ethnicity, age at cohort entry, and time since cohort entry (≤2 y, 2–5 y, or >5 y). Additional adjustment variables in multivariate models were family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, body mass index (BMI; in kg/m2), physical activity (strenuous sports, manual labor, or other vigorous physical activity, measured in h/wk), nonsteroidal anti-inflammatory drug use (aspirin or other nonsteroidal anti-inflammatory drug use ≥2 times/wk for ≥ 1 mo; yes or no), multivitamin use (≥ 1 time/wk; yes or no), hormone replacement therapy (women only; any use or never), and energy intake (logarithmically transformed). A second multivariate model further adjusted for intakes of red meat, dietary fiber, folate (food and supplement sources), vitamin D (food), calcium (food and supplement sources), and alcohol. All dietary variables were adjusted for energy except calcium and folate from both food and supplement sources. Differences across ethnic groups in the effects of GL, carbohydrate, and sucrose were assessed by likelihood ratio tests comparing models with interactive terms between ethnicity and trend variables, and models with main effects only. Comparison across ethnic groups was an a priori aim of the MEC study.
Means were adjusted for age and ethnicity by using analysis of covariance, and Pearson's correlation coefficient (r) was used to assess association between dietary factors. All statistical analyses were performed with SAS software (version 9.1; SAS Institute, Cary, NC). Comparisons with P < 0.05 were considered statistically significant. With the use of the Bonferroni adjustment, results were considered significant if P values were ≤0.0167, which yields an overall significance level of 0.05 across the 3 dietary components.
Results
Baseline characteristics for the cohort population are presented separately for men and women by quintile of GL in Table 1. In both sexes, participants with a higher GL tended to be older, of slightly lower BMI, somewhat more active, and less likely to smoke or consume alcoholic beverages. Both men and women in the highest quintile of GL compared with the lowest quintile consumed less meat, potatoes, and vegetables but more energy, fiber, fruit, grains, and sucrose. These variables are potential confounders of an association of GL and colorectal cancer risk.
Table 1. Dietary and nondietary characteristics of participants in the Multiethnic Cohort by quintile (Q) of glycemic load1.
| Adjusted glycemic load (per 1000 kcal) | |||||
|---|---|---|---|---|---|
|
|
|||||
| Characteristics | Q1 | Q2 | Q3 | Q4 | Q5 |
| Men (n) | 15 402 | 16 136 | 15 947 | 15 687 | 14 798 |
| Nondietary characteristics | |||||
| Median age (y) | 59 (52–66)2 | 60 (52–67) | 60 (52–67) | 61 (52–68) | 62 (53–68) |
| Colorectal cancer in parent or sibling (%) | 7.7 | 7.7 | 7.1 | 7.2 | 7.2 |
| History of colorectal polyp (%) | 6.7 | 7.3 | 7.0 | 6.9 | 6.5 |
| Current or past smoker (%) | 78.9 | 72.1 | 67.9 | 65.2 | 62.8 |
| Pack-years of cigarette smoking (n) | 18.9 ± 0.143 | 14.4 ± 0.13 | 13.6 ± 0.12 | 12.5 ± 0.12 | 12.2 ± 0.12 |
| BMI (in kg/m2) | 26.4 ± 0.03 | 26.5 ± 0.03 | 26.2 ± 0.03 | 25.9 ± 0.03 | 25.5 ± 0.03 |
| Vigorous activity (h/wk) | 4.0 ± 0.05 | 4.1 ± 0.05 | 4.0 ± 0.06 | 4.1 ± 0.06 | 4.2 ± 0.06 |
| NSAID user (%) | 49.3 | 50.9 | 50.6 | 51.2 | 50.0 |
| Multivitamin user (%) | 44.5 | 47.1 | 49.2 | 48.9 | 48.0 |
| Median food intakes | |||||
| Glycemic index | 0.59 (0.54–0.62) | 0.62 (0.59–0.64) | 0.63 (0.60–0.64) | 0.63 (0.61–0.65) | 0.64 (0.62–0.65) |
| Glycemic load (g/d) | 96 (71–128) | 127 (98–166) | 148 (114–192) | 169 (132–219) | 209 (161–269) |
| Carbohydrate (% of energy) | 40 (37–43) | 46 (44–49) | 51 (49–53) | 56 (53–58) | 63 (59–67) |
| Sucrose (g/d) | 26 (18–37) | 36 (25–50) | 41 (29–58) | 46 (32–64) | 50 (34–74) |
| Energy (kcal) | 2034 (1475–2769) | 2085 (1560–2790) | 2163 (1625–2873) | 2213 (1700–2950) | 2334 (1780–3050) |
| Fiber (g · 1000 kcal−1 · d−1) | 8.9 (7.0–11.1) | 10.3 (8.2–12.6) | 10.7 (8.4–13.4) | 11.0 (8.3–14.3) | 10.8 (7.6–14.7) |
| Red meat (g · 1000 kcal−1 · d−1) | 72 (44–113) | 69 (41–108) | 62 (36–98) | 54 (30–85) | 40 (21–69) |
| Vegetables (g · 1000 kcal−1 · d−1) | 103 (73–140) | 110 (81–148) | 110 (80–148) | 107 (77–150) | 96 (66–140) |
| Fruit (g · 1000 kcal−1 · d−1) | 52 (25–89) | 77 (42–126) | 93 (51–150) | 107 (55–177) | 117 (55–206) |
| Grains (g · 1000 kcal−1 · d−1) | 280 (228–328) | 344 (297–392) | 378 (326–430) | 412 (351–473) | 466 (378–547) |
| Whole grains (g · 1000 kcal−1 · d−1) | 48 (25–80) | 69 (38–107) | 78 (42-124) | 86 (45–141) | 89 (40–152) |
| Processed grains (g · 1000 kcal−1 · d−1) | 217 (168–266) | 261 (212–312) | 283 (226–344) | 304 (237–376) | 346 (254–446) |
| Potatoes (g · 1000 kcal−1 · d−1) | 13.4 (7.9–20.8) | 13.5 (8.4–20.6) | 12.5 (7.7–19.3) | 10.9 (6.5–17.2) | 8.2 (4.4–13.9) |
| Mean alcohol (% of energy) | 11.2 ± 0.10 | 4.0 ± 0.04 | 2.5 ± 0.03 | 1.7 ± 0.02 | 0.9 ± 0.02 |
| Nutrient intakes | |||||
| Vitamin D intake (IU · 1000 kcal−1 · d−1) | 57.3 ± 0.34 | 64.9 ± 0.32 | 66.1 ± 0.31 | 65.5 ± 0.32 | 59.3 ± 0.32 |
| Calcium intake (mg) | 876 ± 4.04 | 979 ± 4.31 | 1010 ± 4.51 | 1003 ± 4.61 | 953 ± 4.90 |
| Folate intake (μg) | 467 ± 2.52 | 520 ± 2.67 | 561 ± 2.85 | 589 ± 3.05 | 613 ± 3.51 |
| Women (n) | 18 111 | 18 623 | 18 644 | 18 324 | 16 778 |
| Nondietary characteristics | |||||
| Median age (y) | 57 (50–65) | 58 (51–66) | 59 (51–67) | 60 (52–67) | 61 (53–68) |
| Colorectal cancer in parent or sibling (%) | 8.7 | 9.1 | 8.6 | 8.6 | 8.6 |
| History of colorectal polyp (%) | 4.4 | 4.4 | 4.6 | 4.6 | 4.2 |
| Current or past smoker (%) | 53.8 | 45.4 | 41.7 | 38.7 | 35.6 |
| Pack-years of cigarette smoking (n) | 9.5 ± 0.10 | 7.0 ± 0.08 | 6.0 ± 0.08 | 5.4 ± 0.08 | 5.4 ± 0.08 |
| BMI (in kg/m2) | 26.8 ± 0.05 | 26.5 ± 0.04 | 26.0 ± 0.04 | 25.3 ± 0.04 | 24.9 ± 0.04 |
| Vigorous activity (h/wk) | 1.4 ± 0.03 | 1.5 ± 0.03 | 1.5 ± 0.03 | 1.5 ± 0.03 | 1.6 ± 0.03 |
| NSAID user (%) | 54.4 | 53.1 | 52.9 | 51.4 | 50.4 |
| Multivitamin user (%) | 51.1 | 53.0 | 55.1 | 55.7 | 54.5 |
| Replacement hormone user (%) | 44.6 | 46.5 | 47.0 | 47.4 | 47.8 |
| Median food intakes | |||||
| Glycemic index | 0.61 (0.58–0.63) | 0.62 (0.59–0.64) | 0.62 (0.60–0.64) | 0.63 (0.61–0.65) | 0.63 (0.61–0.65) |
| Glycemic load (g/d) | 82 (61–111) | 109 (84–144) | 125 (97–164) | 141 (110–184) | 171 (132–228) |
| Carbohydrate (% energy) | 43 (39–46) | 49 (47–52) | 54 (51–56) | 58 (56–61) | 65 (61–68) |
| Sucrose (g/d) | 24 (17–35) | 33 (24–46) | 38 (27–52) | 42 (29–58) | 47 (33–68) |
| Energy (kcal) | 1565 (1150–2120) | 1698 (1280–2290) | 1764 (1330–2340) | 1804 (1380–2350) | 1920 (1460–2580) |
| Fiber (g · 1000 kcal−1 · d−1) | 10.4 (8.4–12.7) | 11.7 (9.5–14.3) | 12.3 (9.8–15.2) | 13.0 (10.0–16.3) | 13.5 (9.2–17.3) |
| Red meat (g · 1000 kcal−1 · d−1) | 49 (28–80) | 45 (26–74) | 40 (22–65) | 34 (18–56) | 25 (13–48) |
| Vegetables (g · 1000 kcal−1 · d−1) | 134 (97–187) | 137 (99–187) | 135 (98–187) | 134 (95–188) | 120 (81–174) |
| Fruit (g · 1000 kcal−1 · d−1) | 78 (41–129) | 110 (63–172) | 132 (75–205) | 158 (90–243) | 186 (95–297) |
| Grains (g · 1000 kcal−1 · d−1) | 302 (248–353) | 355 (303–408) | 383 (325–441) | 409 (343–475) | 444 (353–529) |
| Whole grains (g · 1000 kcal−1 · d−1) | 63 (37–98) | 83 (50–127) | 95 (56–144) | 105 (60–161) | 109 (57–177) |
| Processed grains (g · 1000 kcal−1 · d−1) | 223 (132–224) | 256 (203–313) | 270 (212–333) | 283 (118–354) | 301 (216–396) |
| Potatoes (g · 1000 kcal−1 · d−1) | 13.3 (8.0–20.8) | 12.6 (7.5–19.5) | 11.1 (6.5–17.5) | 9.5 (5.4–15.7) | 7.2 (3.4–12.9) |
| Mean alcohol (% of energy) | 4.9 ± 0.06 | 1.4 ± 0.02 | 0.9 ± 0.02 | 0.5 ± 0.01 | 0.3 ± 0.01 |
| Nutrient intakes | |||||
| Vitamin D intake (IU · 1000 kcal−1 · d−1) | 57.3 ± 0.34 | 72.0 ± 0.33 | 72.3 ± 0.32 | 70.0 ± 0.32 | 63.1 ± 0.32 |
| Calcium intake (mg) | 938 ± 4.69 | 1149 ± 4.94 | 1099 ± 5.15 | 1127 ± 5.40 | 1124 ± 5.83 |
| Folate intake (μg) | 431 ± 2.27 | 493 ± 2.38 | 537 ± 2.55 | 573 ± 2.70 | 616 ± 3.15 |
NSAID, nonsteroidal anti-inflammatory drug. Q1 through Q5 adjusted glycemic load values were <130.5, 130.5 to <150.5, 150.5 to <167.9, 167.9 to < 188.5, and ≥188.5, respectively, in men and < 113.9, 113.9 to < 128.7, 128.7 to < 141.6, 141.6 to < 156.9, and ≥156.9, respectively, in women. Values were adjusted for age and ethnicity by analysis of covariance, except for median values.
Median; interquartile range in parentheses (all such values).
x̄ ± SE (all such values).
A higher GL reflects either greater carbohydrate consumption, intake of foods with a higher GI, or both. In the population in the present study, GI varied little between the highest and lowest quintiles of GL (range of medians: 0.59–0.64 in men, 0.61–0.63 in women) and was not strongly correlated with either GL (r = 0.43 in men and 0.27 in women) or carbohydrate intake (r = 0.19 in men and −0.03 in women). However, GL and carbohydrate were strongly correlated as absolute intakes (r = 0.97 in men and 0.98 in women) or as nutrient densities (r = 0.94 in men and 0.92 in women), and these correlations were consistent across ethnic groups. Sucrose intake was less strongly correlated with either GL (r = 0.43 in both sexes) or the percentage of energy from carbohydrate (r = 0.45 in men and 0.49 in women). Dietary fiber density was more strongly correlated with the percentage of energy from carbohydrate (r = 0.35 in men and 0.47 in women) than with GL (r = 0.17 in men and 0.24 in women). Meat intake/1000 kcal was negatively correlated with the percentage of energy from carbohydrate (r = −0.34 and −0.36 in men and women, respectively).
The 10 foods contributing the highest percentages of GL in each of the ethnicity and sex groups are shown in Table 2. For both men and women, rice was the single largest contributor to GL. Across ethnic groups, rice, bread, and sugared sodas were the highest contributors; total potato intake contributed only 1–3% of GL in this population.
Table 2. The top 10 contributors to glycemic load (GL) by sex and ethnicity group.
| All | African American | Native Hawaiian | Japanese American | Latino | White | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
|
|
|
|
|
|
|||||||
| Rank | Daily percentage of GL | Rank | Daily percentage of GL | Rank | Daily percentage of GL | Rank | Daily percentage of GL | Rank | Daily percentage of GL | Rank | Daily percentage of GL | |
| % | % | % | % | % | % | |||||||
| Men | ||||||||||||
| White rice | 1 | 17.0 | 3 | 5.3 | 1 | 26.2 | 1 | 33.9 | 5 | 4.0 | 1 | 8.3 |
| Regular soda | 2 | 6.4 | 1 | 8.7 | 2 | 7.3 | 2 | 4.9 | 1 | 7.4 | 2 | 6.1 |
| White bread | 3 | 4.1 | 4 | 4.6 | 3 | 3.9 | 3 | 3.4 | 3 | 4.7 | 3 | 4.4 |
| Bananas | 4 | 3.5 | 5 | 3.7 | 6 | 2.5 | 4 | 3.0 | 6 | 3.9 | 5 | 3.6 |
| Whole-wheat bread | 5 | 3.4 | 2 | 5.6 | 4 | 3.2 | 5 | 2.9 | 7 | 2.8 | 4 | 4.1 |
| Other dry cereals | 6 | 2.4 | 6 | 3.0 | 8 | 1.9 | 8 | 1.8 | 9 | 2.3 | 6 | 3.4 |
| Rolls, buns, etc | 7 | 2.3 | 4 | 4.6 | ||||||||
| Orange or grapefruit juice | 8 | 2.1 | 7 | 2.7 | 10 | 1.6 | 9 | 1.8 | 10 | 2.2 | 8 | 2.5 |
| Other fruit juice | 9 | 2.1 | 5 | 2.9 | 7 | 2.0 | ||||||
| Other fruit | 10 | 1.8 | ||||||||||
| Bran cereals | 11 | 1.8 | 10 | 2.2 | ||||||||
| Pancakes | 12 | 1.8 | 9 | 1.8 | 10 | 1.6 | 7 | 2.6 | ||||
| Cookies | 13 | 1.7 | 10 | 2.4 | ||||||||
| Popcorn | 16 | 1.5 | 9 | 2.4 | 9 | 2.4 | ||||||
| Brown or wild rice | 19 | 1.2 | 6 | 2.1 | ||||||||
| Corn bread or tortillas | 21 | 1.1 | 8 | 2.4 | 2 | 4.8 | ||||||
| Mexican or Spanish rice | 34 | 0.7 | 8 | 2.5 | ||||||||
| Poi | 72 | 0.3 | 7 | 2.3 | ||||||||
| Women | ||||||||||||
| White rice | 1 | 12.9 | 3 | 4.3 | 1 | 18.4 | 1 | 26.2 | 4 | 4.3 | 1 | 5.8 |
| Regular soda | 2 | 4.6 | 1 | 7.4 | 2 | 6.2 | 5 | 3.7 | 1 | 5.1 | 5 | 3.7 |
| Bananas | 3 | 4.1 | 4 | 4.3 | 6 | 3.2 | 2 | 3.9 | 5 | 4.2 | 3 | 4.3 |
| Whole-wheat bread | 4 | 3.9 | 2 | 5.4 | 4 | 3.7 | 3 | 3.5 | 7 | 3.1 | 2 | 4.5 |
| White bread | 5 | 3.7 | 6 | 3.4 | 3 | 3.7 | 4 | 3.2 | 6 | 4.2 | 4 | 4.1 |
| Other fruit | 6 | 2.9 | 7 | 3.0 | 8 | 2.3 | 6 | 2.6 | 8 | 3.1 | 6 | 3.2 |
| Other fruit juice | 7 | 2.5 | 9 | 2.7 | 5 | 3.4 | 7 | 2.5 | ||||
| Rolls, buns, etc | 8 | 2.5 | 3 | 4.3 | 8 | 3.0 | ||||||
| Orange or grapefruit juice | 9 | 2.4 | 8 | 2.8 | 10 | 2.0 | 9 | 2.3 | 9 | 2.4 | 10 | 2.6 |
| Other dry cereals | 10 | 2.3 | 9 | 2.0 | 10 | 1.8 | 7 | 3.0 | ||||
| Popcorn | 11 | 2.0 | 5 | 3.7 | 9 | 2.7 | ||||||
| Bran cereals | 14 | 1.9 | ||||||||||
| Brown or wild rice | 21 | 1.2 | 8 | 2.4 | ||||||||
| Corn bread or tortillas | 27 | 1.0 | 10 | 2.7 | 2 | 4.5 | ||||||
| Mexican or Spanish rice | 37 | 0.7 | 10 | 2.4 | ||||||||
| Poi | 67 | 0.3 | 7 | 2.7 | ||||||||
During a mean follow-up period of 8.2 y (> 1.5 million person-years of follow-up), 1293 men and 1086 women were diagnosed with colorectal cancer. The mean age at diagnosis was 69.1 y for men and 68.6 y for women.
The RR of colorectal cancer in men and women by quintile of GL is shown in Table 3. On the basis of the 1293 colorectal cancer cases among the men, there was a significant inverse trend (P = 0.026) with adjustment for age, ethnicity, and time on study only. The RR for the highest quintile of intake compared with the lowest quintile was 0.80 (95% CI: 0.67, 0.96). In this basic model, a stronger inverse association was seen with rectal cancer (RR for the highest quintile = 0.65; 95% CI: 0.46, 0.91; P for trend = 0.014) than with colon cancer. This analysis was performed (basic model 1); its scope was limited to the 1166 male colorectal cancer cases with no missing values for energy intake, multivitamin use, or the nondietary variables included in Table 1, so as to include the same subjects as in the subsequent multivariate model. The results were somewhat attenuated, and they were significant only for rectal cancer. After multivariate adjustment (multivariate model 1, which included age, ethnicity, time in the study, energy intake, and nondietary variables, and multivariate model 2, which further adjusted for dietary variables), there were no longer meaningful associations between GL and colorectal or colon cancer in men. There was a significant inverse association with rectal cancer in the highest versus the lowest quintile in multivariate model 1; this association was diminished in multivariate model 2.
Table 3. Relative risks (and 95% CIs) of colorectal cancer by quintile (Q) of glycemic load1.
| Cancer site | Cases | Q2 | Q3 | Q4 | Q5 | P for trend |
|---|---|---|---|---|---|---|
| n | ||||||
| Men | ||||||
| Colorectal cancer | ||||||
| Basic model2 | 1293 | 0.83 (0.70, 0.99) | 0.91 (0.77, 1.08) | 0.83 (0.69, 0.98) | 0.80 (0.67, 0.96) | 0.026 |
| Basic model 13 | 1166 | 0.83 (0.69, 1.00) | 0.90 (0.75, 1.08) | 0.86 (0.71, 1.03) | 0.83 (0.69, 1.01) | 0.113 |
| Multivariate model 14 | 1166 | 0.86 (0.71, 1.03) | 0.95 (0.79, 1.14) | 0.91 (0.75, 1.10) | 0.89 (0.73, 1.08) | 0.377 |
| Multivariate model 25 | 1166 | 1.00 (0.82, 1.24) | 1.17 (0.94, 1.45) | 1.16 (0.92, 1.46) | 1.15 (0.89, 1.48) | 0.193 |
| Colon cancer | ||||||
| Basic model | 925 | 0.87 (0.71-1.07) | 0.90 (0.73, 1.10) | 0.93 (0.76, 1.15) | 0.87 (0.71, 1.08) | 0.376 |
| Basic model 1 | 835 | 0.85 (0.68, 1.06) | 0.88 (0.70, 1.09) | 0.98 (0.79, 1.21) | 0.91 (0.73, 1.14) | 0.769 |
| Multivariate model 1 | 835 | 0.87 (0.70, 1.09) | 0.91 (0.73, 1.14) | 1.03 (0.83, 1.28) | 0.97 (0.77, 1.21) | 0.791 |
| Multivariate model 2 | 835 | 1.02 (0.80, 1.31) | 1.12 (0.87, 1.46) | 1.30 (0.99, 1.71) | 1.22 (0.90, 1.65) | 0.082 |
| Rectal cancer | ||||||
| Basic model | 354 | 0.75 (0.59, 1.04) | 0.92 (0.67, 1.26) | 0.59 (0.42, 0.84) | 0.65 (0.46, 0.91) | 0.007 |
| Basic model 1 | 318 | 0.79 (0.55, 1.12) | 0.94 (0.67, 1.31) | 0.60 (0.41, 0.87) | 0.68 (0.47, 0.92) | 0.016 |
| Multivariate model 1 | 318 | 0.83 (0.58, 1.18) | 1.00 (0.72, 1.40) | 0.65 (0.44, 0.94) | 0.73 (0.50, 1.05) | 0.046 |
| Multivariate model 2 | 318 | 0.97 (0.66, 1.43) | 1.24 (0.83, 1.84) | 0.84 (0.53, 1.32) | 0.97 (0.60, 1.56) | 0.689 |
| Women | ||||||
| Colorectal cancer | ||||||
| Basic model | 1086 | 0.90 (0.75, 1.08) | 0.92 (0.77, 1.11) | 0.80 (0.66, 0.96) | 0.66 (0.54, 0.81) | <0.001 |
| Basic model 1 | 920 | 0.91 (0.74, 1.11) | 0.95 (0.78, 1.16) | 0.75 (0.61, 0.93) | 0.66 (0.53, 0.83) | <0.001 |
| Multivariate model 1 | 920 | 0.94 (0.77, 1.15) | 1.00 (0.82, 1.23) | 0.81 (0.65, 1.00) | 0.72 (0.57, 0.91) | 0.003 |
| Multivariate model 2 | 920 | 0.98 (0.79, 1.21) | 1.05 (0.85, 1.31) | 0.85 (0.67, 1.08) | 0.75 (0.57, 0.97) | 0.017 |
| Colon cancer | ||||||
| Basic model | 857 | 0.92 (0.75, 1.14) | 0.94 (0.76, 1.16) | 0.80 (0.64, 0.99) | 0.69 (0.54, 0.87) | 0.001 |
| Basic model 1 | 717 | 0.94 (0.75, 1.18) | 0.96 (0.77, 1.21) | 0.73 (0.57, 0.93) | 0.68 (0.52, 0.88) | <0.001 |
| Multivariate model 1 | 717 | 0.96 (0.77, 1.21) | 1.01 (0.80, 1.26) | 0.78 (0.61, 1.00) | 0.73 (0.56, 0.95) | 0.006 |
| Multivariate model 2 | 717 | 1.00 (0.79, 1.28) | 1.06 (0.83, 1.36) | 0.83 (0.63, 1.09) | 0.77 (0.57, 1.04) | 0.038 |
| Rectal cancer | ||||||
| Basic model | 224 | 0.81 (0.52, 1.22) | 0.86 (0.57, 1.28) | 0.79 (0.52, 1.19 | 0.60 (0.38, 0.94) | 0.037 |
| Basic model 1 | 198 | 0.80 (0.51, 1.25) | 0.90 (0.58, 1.38) | 0.83 (0.53, 1.29) | 0.63 (0.39, 1.03) | 0.103 |
| Multivariate model 1 | 198 | 0.85 (0.54, 1.33) | 0.98 (0.63, 1.51) | 0.93 (0.59, 1.45) | 0.72 (0.44, 1.19) | 0.305 |
| Multivariate model 2 | 198 | 0.88 (0.55, 1.41) | 1.01 (0.63, 1.62) | 0.93 (0.56, 1.55) | 0.70 (0.39, 1.25) | 0.297 |
Energy-adjusted glycemic load was determined by the residual method. Q2 through Q5 adjusted glycemic load values were 130.5 to <150.5, 150.5 to <167.9, 167.9 to <188.5, and ≥188.5, respectively, in men and 113.9 to <128.7, 128.7 to <141.6, 141.6 to <156.9, and ≥156.9, respectively, in women.
Cox regression models adjusted for age, ethnicity, and time since cohort entry (all such).
Cox regression models adjusted for age, ethnicity, and time since cohort entry; restricted to subjects with no missing values for family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, BMI, hours of vigorous activity, nonsteroidal anti-inflammatory drug use, multivitamin use, and replacement hormone use(women only) (all such).
Cox regression models adjusted for age, ethnicity, time since cohort entry, family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, BMI, hours of vigorous activity, nonsteroidal anti-inflammatory drug use, multivitamin use, energy intake (logarithmically transformed), and replacement hormone use (women only) (all such).
Multivariate model 1 plus adjustment for alcohol, red meat, folate, vitamin D, calcium, and dietary fiber intakes (all such).
GL seems to be more protective against colorectal cancer in women than in men. With adjustment only for age, ethnicity, and time in the study, in 1086 colorectal cancer cases, there was an inverse trend (P for trend < 0.001). When the same analysis was limited to the 920 female cases with no missing values for energy intake, multivitamin use, and the nondietary variables (basic model 1), the decrease in risk was similar (RR = 0.66; 95% CI: 0.53, 0.83; P for trend < 0.001). Multivariate models 1 and 2 slightly weakened the associations for the same 920 cases, but they still resulted in significant trends and RRs in the highest quintile groups for both colorectal and colon cancer. In the fully adjusted model, the RR of colorectal cancer for the highest quintile of intake compared with the lowest quintile was 0.75 (95% CI: 0.57, 0.97, P for trend = 0.017). A similar inverse association in the highest quintile was present for colon cancer (RR = 0.77; 95% CI: 0.57, 1.04, P for trend = 0.038).
To determine whether carbohydrate intake was similarly protective for colorectal cancer, we ran the same models with carbohydrate substituted for GL. In men, there were significant protective effects in multivariate model 1 for both colorectal and rectal cancer, but not when the model was adjusted for other dietary variables (Table 4). In women, the protective effect of carbohydrate intake was very similar to that for GL in the fully adjusted model; the RR for the highest quintile of intake compared with the lowest was 0.71 (95% CI: 0.53, 0.95; P for trend = 0.025) for colorectal cancer and 0.69 (95% CI: 0.50, 0.96: P for trend = 0.038) for colon cancer (Table 4).
Table 4. Relative risks (and 95% CIs) of colorectal cancer by quintile (Q) of carbohydrate intake1.
| Cancer site | Cases | Q2 | Q3 | Q4 | Q5 | P for trend |
|---|---|---|---|---|---|---|
| n | ||||||
| Men | ||||||
| Colorectal cancer | ||||||
| Basic model2 | 1293 | 0.89 (0.75, 1.06) | 0.97 (0.82, 1.15) | 0.84 (0.70, 1.00) | 0.72 (0.60, 0.86) | <0.001 |
| Basic model 13 | 1166 | 0.91 (0.76, 1.10) | 0.94 (0.78, 1.12) | 0.84 (0.69, 1.01) | 0.75 (0.62, 0.91) | 0.003 |
| Multivariate model 14 | 1166 | 0.94 (0.79, 1.14) | 0.99 (0.82, 1.19) | 0.89 (0.74, 1.08) | 0.82 (0.68, 1.00) | 0.044 |
| Multivariate model 25 | 1166 | 1.08 (0.89, 1.32) | 1.19 (0.97, 1.47) | 1.12 (0.89, 1.41) | 1.09 (0.84, 1.40) | 0.603 |
| Colon cancer | ||||||
| Basic model | 925 | 0.86 (0.70, 1.06) | 0.99 (0.81, 1.21) | 0.88 (0.72, 1.09) | 0.76 (0.61, 0.94) | 0.027 |
| Basic model 1 | 835 | 0.88 (0.70, 1.09) | 0.97 (0.78, 1.20) | 0.90 (0.72, 1.11) | 0.80 (0.64, 1.00) | 0.083 |
| Multivariate model 1 | 835 | 0.90 (0.72, 1.13) | 1.01 (0.82, 1.26) | 0.96 (0.77, 1.19) | 0.87 (0.69, 1.10) | 0.365 |
| Multivariate model 2 | 835 | 1.02 (0.81, 1.30) | 1.20 (0.93, 1.53) | 1.16 (0.89, 1.52) | 1.10 (0.81, 1.49) | 0.452 |
| Rectal cancer | ||||||
| Basic model | 354 | 0.92 (0.66, 1.27) | 0.91 (0.66, 1.26) | 0.68 (0.48, 0.95) | 0.60 (0.42, 0.85) | 0.001 |
| Basic model 1 | 318 | 0.94 (0.67, 1.32) | 0.87 (0.61, 1.22) | 0.67 (0.46, 0.96) | 0.62 (0.43, 0.91) | 0.002 |
| Multivariate model 1 | 318 | 0.99 (0.70, 1.39) | 0.92 (0.65, 1.31) | 0.72 (0.50, 1.05) | 0.68 (0.47, 1.00) | 0.014 |
| Multivariate model 2 | 318 | 1.15 (0.80, 167) | 1.15 (0.78, 1.71) | 0.96 (0.62, 1.49) | 0.98 (0.60, 1.59) | 0.642 |
| Women | ||||||
| Colorectal cancer | ||||||
| Basic model | 1086 | 0.86 (0.71, 1.03) | 0.88 (0.73, 1.06) | 0.75 (0.62, 0.90) | 0.61 (0.50, 0.75) | <0.001 |
| Basic model 1 | 920 | 0.85 (0.69, 1.04) | 0.86 (0.71, 1.05) | 0.76 (0.62, 0.94) | 0.62 (0.50, 0.78) | <0.001 |
| Multivariate model 1 | 920 | 0.88 (0.72, 1.08) | 0.92 (0.75, 1.13) | 0.83 (0.67, 1.03) | 0.70 (0.55, 0.87) | 0.002 |
| Multivariate model 2 | 920 | 0.91 (0.73, 1.13) | 0.95 (0.76, 1.19) | 0.86 (0.67, 1.10) | 0.71 (0.53, 0.95) | 0.025 |
| Colon cancer | ||||||
| Basic model | 857 | 0.85 (0.69, 1.04) | 0.90 (0.73, 1.10) | 0.76 (0.61, 0.94) | 0.60 (0.48, 0.76) | <0.001 |
| Basic model 1 | 717 | 0.86 (0.68, 1.08) | 0.86 (0.68, 1.07) | 0.78 (0.61, 0.98) | 0.60 (0.47, 0.76) | <0.001 |
| Multivariate model 1 | 717 | 0.88 (0.70, 1.11 | 0.91 (0.72, 1.14) | 0.83 (0.66, 1.06) | 0.66 (0.51, 0.86) | 0.003 |
| Multivariate model 2 | 717 | 0.92 (0.72, 1.16) | 0.94 (0.73, 1.22) | 0.87 (0.66, 1.15) | 0.69 (0.50, 0.96) | 0.038 |
| Rectal cancer | ||||||
| Basic model | 224 | 0.92 (0.62, 1.37) | 0.83 (0.56, 1.25) | 0.66 (0.43, 1.02) | 0.67 (0.43, 1.04) | 0.029 |
| Basic model 1 | 198 | 0.83 (0.54, 1.29) | 0.91 (0.59, 1.39) | 0.67 (0.42, 1.06) | 0.71 (0.45, 1.13) | 0.101 |
| Multivariate model 1 | 198 | 0.89 (0.57, 1.38 | 1.01 (0.66, 1.57) | 0.76 (0.48, 1.22) | 0.85 (0.53, 1.36) | 0.395 |
| Multivariate model 2 | 198 | 0.91 (0.57, 1.45) | 1.01 (0.62, 1.64) | 0.74 (0.43, 1.27) | 0.78 (0.42, 1.44) | 0.337 |
Energy-adjusted carbohydrate intake was determined by the residual method. Q2 through Q5 adjusted carbohydrate intake values were 243.9 to <273.4, 273.4 to <299.2, 299.2 to <331.2, and ≥331.2 g/d, respectively, in men and 210.7 to <234.5, 234.5 to <255.7, 255.7 to <281.1, and ≥281.1 g/d, respectively, in women.
Cox regression models adjusted for age, ethnicity, and time since cohort entry (all such).
Cox regression models adjusted for age, ethnicity, and time since cohort entry; restricted to subjects with no missing values for family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, BMI, hours of vigorous activity, nonsteroidal anti-inflammatory drug use, multivitamin use, and replacement hormone use (women only) (all such).
Cox regression models adjusted for age, ethnicity, time since cohort entry, family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, BMI, hours of vigorous activity, nonsteroidal anti-inflammatory drug use, multivitamin use, energy intake (logarithmically transformed), and replacement hormone use (women only) (all such).
Multivariate model 1 plus adjustment for alcohol, red meat, folate, vitamin D, calcium, and dietary fiber intakes (all such).
We also ran the same models substituting sucrose for GL (data not shown). In men, sucrose had associations with colorectal and colon cancer almost identical to those of carbohydrate, but there were no significant associations with rectal cancer in any of the models. No associations were seen with sucrose for men in the fully adjusted models. In women, the associations of sucrose with colorectal and colon cancer were very similar to those for carbohydrate, but, in the fully adjusted model, the trend for sucrose was not statistically significant for colorectal cancer (RR for the highest quintile compared with the lowest = 0.88; 95% CI: 0.70, 1.11; P for trend = 0.158) or for colon cancer (RR for the highest quintile compared with the lowest = 0.85; 95% CI: 0.66, 1.11; P for trend = 0.155). Finally, we reevaluated the models for GL and carbohydrate intake unadjusted for energy intake, but the RR of colorectal cancer did not change measurably.
Ethnicity-specific analyses are shown in Table 5 for all groups except Native Hawaiians, for whom the small number of cases precluded meaningful analysis. A significant interaction was found between GL and ethnic group in men (P for interaction = 0.009). In the fully adjusted model, we found no significant associations with colorectal cancer in African American, Japanese American, or Latino men but a clear positive risk in white men (RR for the highest versus lowest quintile = 1.69; 95% CI: 0.98, 2.92; P for trend = 0.006). In women, GL was equally protective in all ethnic groups (P for interaction = 0.579). The trend was marginally significant only in Japanese Americans, because of their large sample size (RR for the highest versus lowest quintile = 0.76; 95% CI: 0.42, 1.37; P for trend = 0.050). The consistency across the 4 ethnic groups means that the finding in women is robust. Results were similar for carbohydrate (Table 6), with a significant interaction in men (P = 0.013) and consistent effects across ethnic groups in women (P for interaction = 0.725). The risk for white men was somewhat reduced (RR for the highest versus the lowest quintile = 1.38; 95% CI: 0.77, 2.48; P for trend = 0.089) and no longer significant, and none of the ethnic groups of women showed significant associations. No significant interactions were found between sucrose and ethnic group within either sex.
Table 5. Relative risks (and 95% CIs) of colorectal cancer by quintile (Q) of glycemic load for each ethnic group (multivariate model 2 only)1.
| Cases | Q2 | Q3 | Q4 | Q5 | P for trend | |
|---|---|---|---|---|---|---|
| n | ||||||
| Men2 | ||||||
| African American | 166 | 1.06 (0.65, 1.75) | 1.31 (0.78, 2.21) | 1.18 (0.66, 2.09) | 1.29 (0.68, 2.44) | 0.404 |
| Japanese American | 491 | 1.11 (0.73, 1.69) | 1.13 (0.74, 1.73) | 0.92 (0.59, 1.44) | 0.95 (0.59, 1.53) | 0.399 |
| Latino | 172 | 0.83 (0.55, 1.24) | 1.00 (0.65, 1.54) | 1.06 (0.66, 1.71) | 1.17 (0.67, 2.03) | 0.456 |
| White | 259 | 0.91 (0.60, 1.38) | 1.02 (0.65, 1.62) | 1.77 (1.11, 2.80) | 1.69 (0.98, 2.92) | 0.006 |
| Women3 | ||||||
| African American | 300 | 1.02 (0.71, 1.47) | 1.05 (0.71, 1.56) | 0.93 (0.59, 1.46) | 0.74 (0.43, 1.29) | 0.507 |
| Japanese American | 335 | 1.00 (0.56, 1.79) | 1.25 (0.72, 2.18) | 0.82 (0.47, 1.46) | 0.76 (0.42, 1.37) | 0.050 |
| Latina | 168 | 0.96 (0.62, 1.50) | 0.84 (0.51, 138) | 0.48 (0.25, 0.93) | 0.75 (0.38, 1.46) | 0.107 |
| White | 216 | 1.00 (0.64, 1.56) | 1.05 (0.65, 1.68) | 1.18 (0.71, 1.98) | 0.68 (0.35, 1.33) | 0.594 |
Energy-adjusted glycemic load was determined by the residual method. Q2 through Q5 adjusted glycemic load values were 59.5 to <68.5, 68.5 to <76.3, 76.3 to <85.5, and ≥85.5, respectively, in men and 64.0 to <72.3, 72.3 to <79.5, 79.5 to <88.0, ≥88.0, respectively, in women. In the multivariate model 2, Cox regression models were adjusted for age, ethnicity, time since cohort entry, family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, BMI, hours of vigorous activity, nonsteroidal anti-inflammatory drug use, multivitamin use, energy intake (logarithmically transformed), replacement hormone use (women only), and alcohol, red meat, folate, vitamin D, calcium, and dietary fiber intakes.
P for interaction = 0.009.
P for interaction = 0.579.
Table 6. Relative risks (and 95% CIs) of colorectal cancer by quintile (Q) of carbohydrate for each ethnic group (multivariate model 2 only)1.
| Cases | Q2 | Q3 | Q4 | Q5 | P for trend | |
|---|---|---|---|---|---|---|
| n | ||||||
| Men2 | ||||||
| African American | 166 | 1.10 (0.69, 1.76) | 1.31 (0.80, 2.15) | 1.08 (0.60, 1.95) | 1.31 (0.69, 2.50) | 0.452 |
| Japanese American | 491 | 1.15 (0.78, 1.69) | 1.13 (0.76, 1.68) | 0.96 (0.63, 1.47) | 0.97 (0.61, 1.53) | 0.460 |
| Latino | 172 | 0.82 (0.55, 1.24) | 0.89 (0.57, 1.39) | 0.91 (0.55, 1.49) | 1.06 (0.60, 1.88) | 0.787 |
| White | 259 | 1.10 (0.73, 1.65) | 1.40 (0.91, 2.17) | 1.66 (1.04, 2.68) | 1.38 (0.77, 2.48) | 0.089 |
| Women3 | ||||||
| African American | 300 | 1.10 (0.76, 1.59) | 1.04 (0.69, 1.58) | 1.02 (0.64, 1.63) | 0.66 (0.36, 1.22) | 0.339 |
| Japanese American | 335 | 0.69 (0.42, 1.13) | 0.74 (0.46, 1.21) | 0.68 (0.41, 1.11) | 0.61 (0.35, 1.09) | 0.139 |
| Latina | 168 | 0.87 (0.53, 1.41) | 0.84 (0.49, 1.43) | 0.88 (0.49, 1.58) | 0.53 (0.25, 1.15) | 0.182 |
| White | 216 | 0.90 (0.57, 1.42) | 1.15 (0.71, 1.84) | 0.79 (0.46, 1.40) | 0.65 (0.33, 1.31) | 0.275 |
Energy-adjusted carbohydrate intakes were determined by the residual method. Q2 through Q5 adjusted carbohydrate intake were 43.7 to <48.9, 48.9 to <53.4, 53.4 to <59.1, and ≥59.1 g · 1000 kcal−1 · d−1, respectively, in men and 45.1 to <50.4, 50.4 to <55.0, 55.0 to <60.5, and ≥60.5 g · 1000 kcal−1 · d−1, respectively, in women. In the multivariate model 2, Cox regression models adjusted for age, ethnicity, time since cohort entry, family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, BMI, hours of vigorous activity, nonsteroidal anti-inflammatory drug use, multivitamin use, energy intake (logarithmically transformed), replacement hormone use (women only), and alcohol, red meat, folate, vitamin D, calcium and dietary fiber intakes.
P for interaction = 0.013.
P for interaction = 0.725.
Discussion
In this large, older, multiethnic population, we found both higher GL and higher carbohydrate intake to be protective against colorectal cancer in women but not men. The protection was seen consistently in women in each of the 4 ethnic groups. The interactions with ethnic group and GL and carbohydrate were significant in men, and the association of both dietary components with colorectal cancer was positive in white men. Thus, GL and carbohydrate intake both showed inverse associations with colorectal cancer in women, although each is a significant direct risk factor in white men.
The results for women in the present study seem to contradict the existing hypothesis that GL increases hyperinsulinemia, which in turn promotes colorectal neoplasia. Evidence linking insulin resistance with colorectal cancer supports the prevailing hypothesis (5–7, 21, 34), and the few studies that have examined the associations of GL per se with colorectal cancer (11, 12, 15, 17–19) have found either positive associations or no association. Only Flood et al (20) found an inverse association; however, that association was not with colorectal cancer but with colorectal adenomas, and it was found only in men. In contrast, our finding in women appears to be consistent across 4 ethnic groups, which indicates that this is a robust finding, because it was replicated in the ethnic subcohorts. This finding, which was unexpected, given the evidence for the insulin resistance hypothesis and that from some other studies of GL and colorectal cancer, may indicate that GL is not a good measure of insulin response, particularly in women in the present population.
In white men, the association of GL with colorectal cancer was significantly positive, as has been seen in previous studies in primarily white subjects (18, 20), although we saw no effect in the other ethnic groups of men. Rice contributes relatively less to the GL in white men than in some other ethnic groups, and pancakes, cookies, and popcorn are among the top 10 contributors to GL only in white men.
The cause of the inconsistencies in our results is not immediately apparent. In contrast to other studies carried out in predominantly white populations with potato-based diets, rice is an important staple in the present cohort, contributing ≈4–33% of the GL in each ethnic group. Potatoes, a somewhat high-GI food (9), do not make up a large percentage of the GL in the present population. White men and women obtain ≈3% of their GL from total potato intake, whereas the other groups obtain 1–2% of their GL from potato. In the only other study to list the top 10 carbohydrate-contributing foods (20), white bread and potatoes appeared near the top for both men and women, and rice did not appear on either list. We suggest that a rice-based diet may not provide as robust a measure of a physiologic response to GL as do bread and potatoes. Rice tends to vary widely in GI, depending on the variety of rice, the cooking methods, and the cooking times (9), and these variations may have affected our GL calculations. Further research to investigate the GL of rice-based diets may be warranted.
GL is the product of carbohydrate intake (in g) and GI. In the present population, GL appears to be a surrogate for carbohydrate intake. However, in ethnicity- and sex-specific analyses, risks for colorectal cancer in white men are higher and more significant from GL than from carbohydrate.
GI varied little in the population of the present study (median 10th–90th percentile: 0.59–0.64), perhaps as a result of the averaging of all foods over a typical day. Our respondents in the highest quintile of GL also ate ≈57% more fruit and ≈47% less meat than did those in the lowest quintile. In a cluster analysis of dietary patterns, Austin et al (35) found that a high-fruit and low-meat diet is more protective against colorectal adenomas than are patterns of greater vegetable or meat consumption (or both). In the present cohort, vegetable intake did not vary across quintiles of GL. If they are used as replacements for meat in the diet, carbohydrates, especially fruit, may afford protection, which would explain the inverse findings in the women in the present study.
Nomura et al (36) found a protective effect for fiber intake in men but not women in the MEC. We used identical covariates in the regression models and additionally adjusted for fiber; we found protective effects for GL or carbohydrates in women but not in men in the fully adjusted models. In men, the effect of fiber may override the effect of other components of carbohydrates, whereas, in women, there may be another aspect of carbohydrate that is more important. However, removing fiber from the model in men did not change the associations.
A primary strength of the present study is the large representative sample of participants from 5 diverse ethnic groups. At the same time, it is not certain how well the diet and other lifestyle data collected at baseline reflect the entire follow-up period, but the period is relatively short at 8 y, and changes over time are likely to lead to attenuation in the RRs with the use of the baseline data. Participants in the MEC are currently responding to a repeat of the baseline questionnaire, in which body weight and dietary information and waist and hip circumferences are being collected. Thus, analyses based on secular trends will be possible in the future. Another area of uncertainty, however, is that of residual confounding by lifestyle and dietary factors that could not be fully controlled for in our models. Women in the highest quintile of GL in the present study had lower alcohol, red meat, and potato intakes and higher grain and fruit intakes, as well as slightly lower BMIs than did women in the lowest quintile of GL. These characteristics may act in concert to further protect against colorectal cancer. However, confounding is unlikely to have resulted in a spurious protective finding in women in each of the ethnic groups. In addition, because of the small number of rectal cancer cases in the present cohort, we may have lacked the power to see relations, and the associations for colorectal cancer seem to be driven by colon cancer in women.
The FCT of the Cancer Research Center of Hawai'i provides GI and GL values for almost 3000 individual foods and mixtures. However, the present study required the calculation of GI and GL for each participant with the use of QFFQ data. Because GI is a measure of the effect of discrete foods on individual people, computing these values from a QFFQ requires assumptions about food form, cooking methods, and processing. Furthermore, QFFQs assess foods individually and give no data on meal patterns or food combinations in mixed meals, and those traits make it impossible to ascertain interactions among foods with differing GLs that are consumed at the same time. Levitan et al (37), in a study of 141 men that compared FFQ data with diet records, found that both GI and GL could be measured acceptably using a FFQ. However, they also found that intakes of cakes and pastries were underreported on the FFQ, which would tend to skew GL values.
In conclusion, contrary to our hypothesis, GL and carbohydrate intake both appear to be protective against colorectal cancer in women after adjustment for potential confounders. This divergence from previous reports linking GL to colorectal cancer through insulin resistance indicates that carbohydrate foods may not all have the same predictive effect on insulin response and, thus, on disease. Further investigation of the glycemic effects of rice-based diets is needed.
2.
Supported by grant no. R37CA054281 from the National Cancer Institute, US Department of Health and Human Services.
Footnotes
The authors'uo; responsibilities were as follows—NCH, SPM, and LNK: the study concept and design; LNK, LRW, and BEH: data collection; NCH and LRW: statistical analysis; NCH, SPM, and LNK: interpretation of results; NCH: writing the manuscript; and all authors: critical review and approval of the final manuscript. None of the authors had a financial or personal conflict of interest.
References
- 1.Potter JD. Colon cancer: review of the epidemiology. Epidemiol Rev. 1993;15:499–545. doi: 10.1093/oxfordjournals.epirev.a036132. [DOI] [PubMed] [Google Scholar]
- 2.World Cancer Research Fund. Food, nutrition and the prevention of cancer: a global perspective. Washington, DC: American Institute for Cancer Research; 1997. [DOI] [PubMed] [Google Scholar]
- 3.McKeown-Eyssen GE. Epidemiology of colorectal cancer revisited: are serum triglycerides and/or plasma glucose associated with risk? Cancer Epidemiol Biomarkers Prev. 1994;3:687–95. [PubMed] [Google Scholar]
- 4.Giovannucci E. Insulin and colon cancer. Cancer Causes Control. 1995;6:164–79. doi: 10.1007/BF00052777. [DOI] [PubMed] [Google Scholar]
- 5.Giovannucci E. Insulin, insulin-like growth factors and colon cancer: a review of the evidence. J Nutr. 2001;131(suppl):3109S–20S. doi: 10.1093/jn/131.11.3109S. [DOI] [PubMed] [Google Scholar]
- 6.Kaaks R, Toniolo P, Akhmedkhanov A, et al. Serum C-peptide, insulinlike growth factor (IGF)-I, IGF-binding proteins, and colorectal cancer risk in women. J Natl Cancer Inst. 2000;92:1592–600. doi: 10.1093/jnci/92.19.1592. [DOI] [PubMed] [Google Scholar]
- 7.Jenkins DJA, Kendall CWC, Augustin LSA, et al. Glycemic index: overview of implications in health and disease. Am J Clin Nutr. 2002;76(suppl):266S–73S. doi: 10.1093/ajcn/76/1.266S. [DOI] [PubMed] [Google Scholar]
- 8.Jenkins DJA, Wolever TMS, Taylor RH, et al. Glycemic index of foods: a physiological basis for carbohydrate exchange. Am J Clin Nutr. 1981;34:362–6. doi: 10.1093/ajcn/34.3.362. [DOI] [PubMed] [Google Scholar]
- 9.Foster-Powell K, Holt SHA, Brand-Miller JC. International table of glycemic index and glycemic load values: 2002. Am J Clin Nutr. 2002;76:5–56. doi: 10.1093/ajcn/76.1.5. [DOI] [PubMed] [Google Scholar]
- 10.De Stefani E, Mendilaharsu M, Deneo-Pellegrini H. Sucrose as a risk factor for cancer of the colon and rectum: a case-control study in Uruguay. Int J Cancer. 1998;75:40–4. doi: 10.1002/(sici)1097-0215(19980105)75:1<40::aid-ijc7>3.0.co;2-1. [DOI] [PubMed] [Google Scholar]
- 11.Slattery ML, Benson J, Berry TD, et al. Dietary sugar and colon cancer. Cancer Epidemiol Biomarkers Prev. 1997;6:677–85. [PubMed] [Google Scholar]
- 12.Franceschi S, Dal Maso L, Augustin L, et al. Dietary glycemic load and colorectal cancer risk. Ann Oncol. 2001;12:173–8. doi: 10.1023/a:1008304128577. [DOI] [PubMed] [Google Scholar]
- 13.Levi F, Pasche C, Lucchini F, La Vecchia C. Macronutrients and colorectal cancer: a Swiss case-control study. Ann Oncol. 2002;13:369–73. doi: 10.1093/annonc/mdf110. [DOI] [PubMed] [Google Scholar]
- 14.Larsson SC, Giovannucci E, Wolk A. Dietary carbohydrate, glycemic index, and glycemic load in relation to risk of colorectal cancer in women. Am J Epidemiol. 2007;165:256–61. doi: 10.1093/aje/kwk012. [DOI] [PubMed] [Google Scholar]
- 15.Terry PD, Jain M, Miller AB, Howe GR, Rohan TE. Glycemic load, carbohydrate intake, and risk of colorectal cancer in women: a prospective cohort study. J Natl Cancer Inst. 2003;95:914–6. doi: 10.1093/jnci/95.12.914. [DOI] [PubMed] [Google Scholar]
- 16.McCarl M, Harnack L, Limburg PJ, Anderson KE, Folsom AR. Incidence of colorectal cancer in relation to glycemic index and load in a cohort of women. Cancer Epidemiol Biomarkers Prev. 2006;15:892–6. doi: 10.1158/1055-9965.EPI-05-0700. [DOI] [PubMed] [Google Scholar]
- 17.Higginbotham S, Zhang ZF, Lee IM, et al. Dietary glycemic load and risk of colorectal cancer in the Women's Health Study. J Natl Cancer Inst. 2004;96:229–33. doi: 10.1093/jnci/djh020. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Michaud DS, Fuchs CS, Liu S, Willett WC, Colditz GA, Giovannucci E. Dietary glycemic load, carbohydrate, sugar, and colorectal cancer risk in men and women. Cancer Epidemiol Biomarkers Prev. 2005;14:138–43. [PubMed] [Google Scholar]
- 19.Oh K, Willett WC, Fuchs CS, Giovannucci EL. Glycemic index, glycemic load, and carbohydrate intake in relation to risk of distal colorectal adenoma in women. Cancer Epidemiol Biomarkers Prev. 2004;13:1192–8. [PubMed] [Google Scholar]
- 20.Flood A, Peters U, Jenkins DJA, et al. Carbohydrate, glycemic index, and glycemic load and colorectal adenomas in the Prostate, Lung, Colorectal, and Ovarian Screening Study. Am J Clin Nutr. 2006;84:1184–92. doi: 10.1093/ajcn/84.5.1184. [DOI] [PubMed] [Google Scholar]
- 21.Bruce WR, Wolever TMS, Giacca A. Mechanisms linking diet and colorectal cancer: the possible role of insulin resistance. Nutr Cancer. 2000;37:19–26. doi: 10.1207/S15327914NC3701_2. [DOI] [PubMed] [Google Scholar]
- 22.Kolonel LN, Henderson BE, Hankin JH, et al. A multiethnic cohort in Hawaii and Los Angeles: baseline characteristics. Am J Epidemiol. 2000;151:346–57. doi: 10.1093/oxfordjournals.aje.a010213. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Nielsen SJ, Popkin BM. Patterns and trends in food portion sizes, 1977– 1998. JAMA. 2003;289:450–3. doi: 10.1001/jama.289.4.450. [DOI] [PubMed] [Google Scholar]
- 24.Nöthlings U, Wilkens LR, Murphy SP, Hankin JH, Henderson BE, Kolonel LN. Meat and fat intake as risk factors for pancreatic canter: the Multiethnic Cohort Study. J Natl Cancer Inst. 2005;97:1458–65. doi: 10.1093/jnci/dji292. [DOI] [PubMed] [Google Scholar]
- 25.Stram DO, Hankin JH, Wilkens LR, et al. Calibration of the dietary questionnaire for a multiethnic cohort in Hawaii and Los Angeles. Am J Epidemiol. 2000;151:358–76. doi: 10.1093/oxfordjournals.aje.a010214. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Brand Miller J. [accessed 14 September 2006];Glycemic index database. Internet: www.glycemicindex.com.
- 27.Brand-Miller J, Burani J, Foster-Powell K, Holt S. The new glucose revolution: complete guide to glycemic index values. New York, NY: Marlowe and Company; 2003. [Google Scholar]
- 28.Wolever TMS, Jenkins DJA. The use of glycemic index in predicting the blood glucose response to mixed meals. Am J Clin Nutr. 1986;43:167–72. doi: 10.1093/ajcn/43.1.167. [DOI] [PubMed] [Google Scholar]
- 29.Matthews RH, Pehrsson PR, Farhat-Sabet M. Sugar content of selected foods: individual and total sugars. Washington, DC: US Department of Agriculture; 1987. Home Economics Research Report no. 48. [Google Scholar]
- 30.Hankin JH, Murphy SP, Lau C, Umphress S, Kolonel LN. Techniques for combining American and British food composition data on specific carbohydrates. J Food Comp Anal. 2000;13:419–24. [Google Scholar]
- 31.Therneau TM, Grambsch PM. Modeling survival data: extending the Cox model. New York, NY: Springer, Inc; 2001. [Google Scholar]
- 32.Willett W. Nutritional epidemiology. 2nd. New York, NY: Oxford University Press; 1998. [Google Scholar]
- 33.Kipnis V, Subar AF, Midthune D, et al. Structure of dietary measurement error: results of the OPEN biomarker study. Am J Epidemiol. 2003;158:14–21. doi: 10.1093/aje/kwg091. [DOI] [PubMed] [Google Scholar]
- 34.Sandhu MS, Dunger DB, Giovannucci EL. Insulin, insulin-like growth factor-I (IGF-I), IGF binding proteins, their biologic interactions, and colorectal cancer. J Natl Cancer Inst. 2002;94:972–80. doi: 10.1093/jnci/94.13.972. [DOI] [PubMed] [Google Scholar]
- 35.Austin GL, Adair LS, Galanko JA, Martin CF, Satia JA, Sandler RS. A diet high in fruits and low in meats reduces the risk of colorectal adenomas. J Nutr. 2007;137:999–1004. doi: 10.1093/jn/137.4.999. [DOI] [PubMed] [Google Scholar]
- 36.Nomura AMY, Hankin JH, Henderson BE, et al. Dietary fiber and colorectal cancer risk: The Multiethnic Cohort Study. Cancer Causes Control. 2007;7:753–64. doi: 10.1007/s10552-007-9018-4. [DOI] [PubMed] [Google Scholar]
- 37.Levitan EB, Westgren CW, Liu S, Wolk A. Reproducibility and validity of dietary glycemic index, dietary glycemic load, and total carbohydrate intake in 141 Swedish men. Am J Clin Nutr. 2007;85:548–53. doi: 10.1093/ajcn/85.2.548. [DOI] [PubMed] [Google Scholar]