Dietary glycemic load, carbohydrate, sugar, and colorectal cancer risk in men and women.

Abstract

Hyperinsulinemia may explain excess colorectal cancer among individuals who are overweight or inactive. Recent studies have observed elevated colorectal cancer risk among individuals with elevated insulin levels 2 hours after oral glucose challenge or with elevated plasma C-peptide levels. The effect of consuming a high glycemic diet on colorectal risk, however, remains uncertain. Two prospective cohort studies, the Nurses' Health Study and the Health Professionals Follow-up Study, contributed up to 20 years of follow-up. After exclusions, 1,809 incident colorectal cancers were available for analyses. Dietary glycemic load (GL) was calculated as a function of glycemic index (postprandial blood glucose response as compared with a reference food), carbohydrate content, and frequency of intake of individual foods reported on food frequency questionnaires. Multivariable Cox proportional hazards models were used to adjust for potential confounders. Intakes of dietary carbohydrate, GL, overall glycemic index, sucrose, and fructose were not associated with colorectal cancer risk in women. A small increase in risk was observed in men with high dietary GL (multivariate relative risk, 1.32; 95% confidence interval, 0.98-1.79; highest versus lowest quintile), sucrose or fructose (multivariate relative risk, 1.37; 95% confidence interval, 1.05-1.78; highest versus lowest quintile of fructose, P = 0.008). Associations were slightly stronger among men with elevated body mass index (> or =25 kg/m(2)). Results among women were similar after stratifying by body mass index or physical activity. High intakes of GL, fructose, and sucrose were related to an elevated colorectal cancer risk among men. For women, however, these factors did not seem to increase the risk of colorectal cancer.

Full text

Page 1

Dietary Glycemic Load, Carbohydrate, Sugar, and Colorectal
Cancer Risk in Men and Women
Dominique S. Michaud,1Charles S. Fuchs,3,5Simin Liu,1,4Walter C. Willett,1,2,3
Graham A. Colditz,1,3and Edward Giovannucci1,2,3
Departments of
and
and
1Epidemiology and
2Nutrition, Harvard School of Public Health;
3Channing Laboratory, Department of Medicine
4Division of Preventive Medicine, Harvard Medical School and Brigham and Women’s Hospital;
5Department of Adult Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts
Abstract
Hyperinsulinemia may explain excess colorectal cancer
among individuals who are overweight or inactive. Recent
studies have observed elevated colorectal cancer risk among
individuals with elevated insulin levels 2 hours after oral
glucose challenge or with elevated plasma C-peptide levels.
The effect of consuming a high glycemic diet on colorectal
risk, however, remains uncertain. Two prospective cohort
studies, the Nurses’ Health Study and the Health Profes-
sionals Follow-up Study, contributed up to 20 years of
follow-up. After exclusions, 1,809 incident colorectal cancers
were available for analyses. Dietary glycemic load (GL) was
calculated as a function of glycemic index (postprandial
blood glucose response as compared with a reference food),
carbohydrate content, and frequency of intake of individual
foods reported on food frequency questionnaires. Multi-
variable Cox proportional hazards models were used to
adjust for potential confounders. Intakes of dietary carbo-
hydrate, GL, overall glycemic index, sucrose, and fructose
were not associated with colorectal cancer risk in women. A
small increase in risk was observed in men with high
dietary GL (multivariate relative risk, 1.32; 95% confidence
interval, 0.98-1.79; highest versus lowest quintile), sucrose
or fructose (multivariate relative risk, 1.37; 95% confidence
interval, 1.05-1.78; highest versus lowest quintile of fruc-
tose, P = 0.008). Associations were slightly stronger among
men with elevated body mass index (z25 kg/m2). Results
among women were similar after stratifying by body mass
index or physical activity. High intakes of GL, fructose, and
sucrose were related to an elevated colorectal cancer risk
among men. For women, however, these factors did not
seem to increase the risk of colorectal cancer.
Epidemiol Biomarkers Prev 2005;14(1):138–43)
(Cancer
Introduction
Substantial evidence indicates that hyperinsulinemia may play
an important role in colorectal cancer (1, 2). Many of the
established risk factors of colorectal cancer, including obesity
and physical inactivity, directly influence insulin levels.
Several studies have observed elevated risks of colon cancer
among those with a history of type 2 diabetes mellitus (3-8).
Studies examining insulin levels in blood (using 2-hour insulin
test) or plasma C-peptide levels (a marker of insulin secretion)
have reported 2- to 3-fold elevated risk of colorectal cancer for
those in the highest categories (9-11).
Dietary intake can influence insulin levels, especially
among individuals who are insulin resistant due to other
factors, such as obesity. Dietary glycemic load (GL), which is
a quantitative measure of the glycemic effect of food, has
been associated with triglycerides and high-density lipopro-
tein levels (12-14), as well as the risk of diabetes (15, 16) and
heart disease (17). Two case-control studies reported positive
associations between intakes of GL and colorectal cancer risk
(18, 19). In a recent prospective study, a relative risk (RR) of
2.85 [95% confidence interval (95% CI), 1.40-5.80] was
observed for colorectal cancer risk comparing the highest to
the lowest quintile of GL intake (20). In another prospective
study, GL was not related to colorectal cancer risk, although a
slight increase in risk was observed among those with distal
colon cancer (21). Moreover, no association was observed
between GL, glycemic index (GI), or carbohydrate intake and
the risk of distal colorectal adenoma in the Nurses’ Health
Study (NHS; ref. 22).
In a recent cross-sectional analysis in the NHS I and II, we
observed statistically significant higher plasma C-peptide
levels among women with high intakes of fructose and GL
(23). To provide additional data on the role of the quality
and quantity of carbohydrates on colorectal cancer risk, we
examined the association between dietary carbohydrate,
sucrose, fructose, GI, and GL and the risk of colon and
rectal cancers in two large prospective cohorts with repeated
diet measures and up to 20 years of follow-up.
Material and Methods
Study Populations. Two ongoing cohort studies provided
data for our analyses, the Health Professionals Follow-up
Study (HPFS) and the NHS I. The HPFS was initiated in 1986
when 51,529 U.S. men ages 40 to 75 years responded to a
mailed questionnaire with detailed information on individual
characteristics and lifestyle habits. Fifty-eight percent of the
men in the HPFS cohort are dentists, and the other
professionals include optometrists, osteopaths, podiatrists,
pharmacists, and veterinarians. The NHS I was initiated in
1976 when 121,700 female registered nurses ages 30 to 55 years
responded to a mailed questionnaire. Information on individ-
ual characteristics and habits such as age, marital status,
weight, height, medical history, medication use, menopausal
status, physical activity, and vitamin use was obtained from
the baseline or from follow-up questionnaires. In both cohorts,
changes in lifestyle habits and information on disease onset
have been obtained biennially since study onset using mailed
questionnaires.
For each of the follow-up questionnaires, up to six
mailings were sent to nonrespondents. Most of the deaths
Received 5/10/04; revised 8/13/04; accepted 8/20/04.
Grant support: Grants CA87969 and CA55075.
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.
Requests for reprints: Dominique Michaud, Harvard School of Public Health, Kresge 920, 677
Huntington Avenue, Boston, MA 02115. Phone: 617-432-4508; Fax: 617-566-7805. E-mail:
dmichaud@hsph.harvard.edu
Copyright D 2005 American Association for Cancer Research.
138
Cancer Epidemiol Biomarkers Prev 2005;14(1). January 2005
Cancer Epidemiology, Biomarkers & Prevention

Page 2

in this cohort were reported by family members or by the
postal service in response to the follow-up questionnaires. In
addition, the National Death Index was searched for
nonrespondents; this method has been shown to have a
sensitivity of 98% (i.e., the National Death Index did not
identify 2% of deaths; ref. 24).
The 1986 baseline questionnaire mailed to the HPFS cohort
included a food frequency questionnaire (FFQ, which was
completed by all participants). In NHS I, the dietary question-
naire was included in the 1980 mailing, which was completed
by 98,462 participants. We excluded participants with implau-
sibly high or low caloric intake (<500 or >3,500 kcal/d for
women; <800 or >4,200 kcal/d for men). In addition,
individuals who reported a previous cancer diagnosis (other
than nonmelanoma skin cancer), ulcerative colitis, Crohn’s
disease, or familial polyposis syndrome at baseline were also
excluded, which left 83,927 women and 47,422 men eligible for
analysis.
Dietary Assessment. A 61-item FFQ was mailed to all
NHS I participants in 1980, and a 131-item FFQ was mailed
to HPFS participants in 1986. Follow-up FFQs were mailed
to NHS I in 1984 and 1986 and every 4 years thereafter. The
1984 NHS I FFQ was expanded to include a larger number
of food items (116 items) and resembles the HPFS FFQ. The
HPFS follow-up FFQs are mailed every 4 years after 1986. In
the FFQ, participants are asked to report their average
frequency of intake over the previous year for a specified
serving size of each food. Individual nutrient intakes are
calculated by multiplying the frequency of each food
consumed by the nutrient content of the specified portion
size [obtained from the U.S. Department of Agriculture (25)
and supplemented with information from manufactures] and
are summed to include contributions from all foods.
Nutrient values are adjusted for total energy intake. Total
fructose intake was calculated as free fructose plus fructose
from sucrose intake.
The GI is based on the postprandial blood glucose response
compared with a reference food. GI values for foods that
appear in the FFQ were either obtained from published
estimates (26), or from direct testing of food items at the
Nutrition Center of the University of Toronto (D. Jenkins). The
GI value is calculated by the following formula:
GI¼
Pincrementalbloodglucoseunderthecurveof testfood
Pincrementalbloodglucoseareaunderthecurveof referencefood? 100%
The GI value for a meal containing mixed foods can
be predicted as the weighed mean of the GI values for each
of the component foods (27, 28). When GI values vary for a
specific food, we use the average GI values (using the same
standard reference food) that are most representative of our
populations (i.e., values available on North American foods
and from healthy individuals, and when possible, from the
time period closest to the questionnaire).
Using these GI values, we also calculated the average dietary
GL during the past year for each participant by multiplying the
carbohydrate content (grams per serving) for each food by its
GI value; multiplying that product by the frequency of
consumption (servings of that food per day), and summing
values for all food items reported:
Individual dietary GL ¼ sum of ½GI
?ðcarbohydrate content of foodÞ
?ðservings of food=dÞ?
Each unit of GL represents the equivalent of 1 g of car-
bohydrate from white bread (12). In addition, the over-
all dietary GI was calculated by dividing GL by the total
amount of carbohydrate; this value represents the over-
all quality of carbohydrate intake for each participant.
In a validity study of 173 women from the NHS I, a FFQ was
compared with four 1-week diet records. For individual food
items that have high GI values, the correlation coefficients
between theaverage intakeassessedbytwo 1-weekdietrecords
completed 6 months apart and the FFQ were as follows: 0.71 for
white bread, 0.77 for dark bread, 0.66 for potatoes, 0.84 for
orange or grapefruit juice, and 0.56 for noncarbonated fruit
drinks (includes fruit-flavored punch; ref. 29). Correlation
coefficients for total carbohydrate and sucrose were 0.45 and
0.54, respectively, comparing two 1-week diet records and the
FFQ in the same validation study of women (30). Similar
correlations were observed in a validation study of men in the
HPFS using two 1-week dietary records and comparing to a
FFQ (r = 0.62 for energy-adjusted carbohydrate; ref. 31).
Assessment of Nondietary Factors. Height, current weight,
smoking history, aspirin use, and physical activity were
initially reported at baseline or in 1980. Family history of
colorectal cancer was ascertained in 1982, 1988, 1992, 1996, and
2000 in the women and in 1986, 1990, 1992, 1996, and 2000 in
the men. During follow-up, data on current weight and
smoking status were obtained from the biennial mailed
questionnaires. We estimated body mass index (BMI) from
weight and height (kg/height in m2) as a measure of total
adiposity. Participants were asked about history of diabetes at
baseline and in all subsequent questionnaires.
For physical activity in the NHS I, we derived a score based
on questions asked in the 1980 questionnaire. The physical
activity variable from the 1980 questionnaire has been shown
to predict the risk of noninsulin-dependent diabetes mellitus
in this cohort of women (32). In 1986, the questionnaires
mailed to the two cohorts included a section assessing
physical activity in detail. From these questions, we calculated
a weekly physical activity (MET-hour) score by multiplying
the time spent on each activity by the typical energy
expenditure for that activity expressed in metabolic equiv-
alents (MET). The MET is the caloric expenditure per kilogram
of body weight per hour of activity, divided by the equivalent
per hour at rest.
The reliability and validity of the assessment of physical
activity as used in these two cohorts were tested among 147
participants of NHS II, a similar cohort to NHS I but
participants were younger nurses. The correlation between
physical activity reported on the questionnaire and that
recorded in the four 1-week diaries was 0.62 (33). The validity
of the physical activity questionnaire used in the HPFS in 1986
was assessed among 238 randomly selected participants by
comparisons with four 1-week activity diaries, four 1-week
activity recalls, and resting and postexercise pulse rates (34).
The correlation for vigorous physical activity with the activity
diaries was 0.58. Vigorous activity assessed by the question-
naire was correlated with resting pulse (r = ?0.45) and
postexercise pulse (r = ?0.41).
Identification of Colorectal Cancer Cases. Participants
were asked to report specified medical conditions, including
cancers that were diagnosed in the 2-year period between each
follow-up questionnaire. Whenever a participant (or next-of-
kin for decedents) reported a diagnosis of colorectal cancer, we
asked for permission to obtain related medical records or
pathology reports. If permission to obtain records was denied,
we attempted to confirm the self-reported cancer with an
additional letter or phone call to the participant. If the primary
cause (or secondary cause) of death as reported by a death
certificate was a previously unreported colorectal cancer case,
we contacted a family member to obtain permission to retrieve
medical records, or in the least to confirm the cancer diagnosis.
A physician reviewed the medical records to verify informa-
tion on histologic type, anatomic location, and stage of the
cancer. Colorectal cancers with missing anatomic location
Cancer Epidemiology, Biomarkers & Prevention
139
Cancer Epidemiol Biomarkers Prev 2005;14(1). January 2005

Page 3

(<15%) were grouped with the colon cancers. In the NHS I, a
total of 1,113 colorectal cancer cases (870 colon and 243 rectal)
were diagnosed between baseline (1980) and May 31, 2000,
and in the HPFS, 696 colorectal cancer cases (561 colon and 135
rectal) were diagnosed between baseline and January 31, 2000.
Statistical Analysis. We computed person-time of follow-
up for each participant from the return date of the baseline
questionnaire to the date of colorectal cancer diagnosis, death
from any cause, or the end of follow-up (May 31, 2000 in NHS;
January 31, 2000 in HPFS), whichever came first. Incidence
rates of colorectal cancer were calculated by dividing the
number of incident cases by the number of person-years in
each category of dietary exposure. We computed the RR for
each of the upper categories by dividing the rates in these
categories by the rate in the lowest category.
RRs adjusted for potential confounders were estimated
using Cox proportional hazards models stratified on age in
years. In these models, we included covariates that have
been previously associated with colorectal cancer in the
cohorts, using previously defined categories (35); these
include BMI, physical activity, smoking (pack-years of
smoking before age 30), family history of colorectal cancer,
height, alcohol intake, history of colonoscopy or sigmoidos-
copy, and dietary intakes of folate, calcium, processed meats,
and beef, pork or lamb as a main dish. In addition, we
adjusted for aspirin use in men (36) and women (37), and
menopausal status and postmenopausal hormone use in
women. All Ps are based on two-sided tests. We did tests
for trend by assigning the median value to each category
and modeling this variable as a continuous variable.
We did additional analyses using the 1984 dietary
questionnaire as baseline for the NHS I (because the 1984
FFQ had more food items), and separately, using cumulative
updating of the dietary exposures with follow-up data (1984,
1986, 1990, and 1994 in the NHS I and 1990 and 1994 in the
HPFS; ref. 38).
Results
Age, height, BMI, calcium, and total caloric intake did not
vary appreciably across quintiles of dietary GL in either
cohort (Table 1). Men and women with higher GL were less
likely to smoke, consumed less alcohol and total fat, but had
higher intakes of carbohydrate and folate (Table 1). Other
baseline characteristics varied differently by cohort; for
example, exercise level did not vary across GL in the
women but was positively associated with dietary GL in the
men (Table 1).
No associations were observed for dietary carbohydrate, GL,
GI, sucrose, or fructose and the risk of colorectal cancer in the
NHS I (Table 2). In the HPFS, men with higher intakes of GL,
sucrose or fructose had a slightly elevated risk of colorectal
cancer (Table 2). RRs in the age-adjusted models were not
elevated (data not shown), but these estimates were confound-
ed by alcohol intake; adding alcohol intake to the multivariate
model accounted for almost all of the change in the risk
estimate. The multivariate RR for a 10-unit increment in GL
was 1.03 (95% CI, 1.00-1.06) in men and 1.05 (95% CI, 0.98-1.12)
in women.
In men, associations between GL and sucrose intakes were
more apparent for rectal cancer risk (Table 2). For fructose
intake, the increase in risk was stronger for colon cancer (Table
2). When stratifying by colon site (proximal and distal), no
statistically significant associations were observed in either
cohort (Table 3).
Although we did not control for diabetes in the multivar-
iate models in Tables 2 and 3, adding this variable did not
change the estimates. Removing diabetics from the analyses
slightly strengthened the association between fructose intake
and risk of colorectal cancer in men (RRs for the top four
quintile: 1.04, 1.21, 1.29, and 1.40). We observed no associa-
tions for the dietary factors of interest in either cohort when
repeating the analyses using cumulative updating of dietary
Table 1. Baseline characteristics according to quintile of energy-adjusted GL among NHS cohort participants and HPFS
participants
CharacteristicsNHS (quintiles of GL*)HPFS (quintiles of GL*)
135135
No. individuals
Glycemic load (mean)
Age (y)
Height (in.)
BMI (kg/m2)
Exercisec
Current smokers (%)
Pack-years of cigarettesb
Family history (%)
Endoscopy history (%)
Diabetes (%)
Aspirin use (%)
Mean daily intake
Calories (kcal)
Total fat (g)x
Carbohydrate (g)x
Protein (g)x
Alcohol (g)
Total calcium (mg)x
Total folate (Ag)x
Red meat (servings/wk){
Cereal fiber (g)
RGI
16,776 16,782
119
16,735
175
9,476
127
9,499
177
9,444
230 76
47
65
24
47
65
24
3.2
26
6.9
7.8
15
2.5
51
47
64
25
3.0
27
6.8
8.0
15
2.0
51
54
70
26
18
17
12
7.9
27
4.1
30
54
70
26
21
8.5
11
8.3
28
3.0
29
54
70
25
24
5.4
11
8.9
29
2.2
28
3.2
35
7.5
8.0
15
2.6
51
1,5521,5831,5361,9572,0241,930
82
108
84.8
12
696
344
3.8
1.6
68
7057
201
64.7
3.1
674
378
1.5
2.9
78
82
179
99.8
23
868
435
2.5
3.8
72
73
235
93.3
9.3
910
476
1.8
5.8
76
58
156
77.1
5.4
775
372
2.5
2.6
73
292
83.4
4.3
911
539
2.1
8.2
79
NOTE: All variables (except age) are age-standardized; mean values are presented for continuous variables and percentages are shown for categorical variables.
*Quintile cut points NHS: <93, 93-111, 112-127, 128-148, >148; quintile cut points HPFS: <147, 147-167, 168-185, 186-206, <206.
cExercise in NHS is in hours per week and in METs/week for the HFPS.
bPack-years are calculated for current and past smokers before age 30.
xEnergy-adjusted nutrient values.
{Beef, pork, or lamb as main dish.
Glycemic Load, Sugar, and Colorectal Cancer
140
Cancer Epidemiol Biomarkers Prev 2005;14(1). January 2005

Page 4

exposures, or when using 1984 as baseline for the NHS I
cohort (data not shown).
We considered whether menopausal status and hormone
use altered the association between the dietary variables in
Table 2 and colorectal cancer risk in women. Among
premenopausal women, an association between carbohydrate
intake and colorectal cancer risk was apparent (multivariate
RR, 2.03; 95% CI, 1.01-4.06, highest versus lowest quintile),
but no associations were observed for dietary GL, fructose or
sucrose intake. No associations were observed for any of the
dietary variables in postmenopausal women when stratifying
on hormone use (never, past, and current use).
The effect of diet on insulin response may vary across strata
of BMI or physical activity as these two factors can be strong
determinants of insulin resistance, which can magnify the
adverse influence of a high GL. We examined this possibility
by stratifying our analyses into two BMI and physical activity
strata, separately. In men, the multivariate RR for colorectal
cancer for the highest versus lowest quintile of fructose intake
was 1.47 (95% CI, 1.05-2.07) in the elevated BMI strata (BMI
z25 kg/m2). No association was apparent in the low BMI
strata (multivariate RR, 1.09; 95% CI, 0.69-1.71 for extreme
quintile comparison of fructose intake). Stratifying by BMI
resulted in a similar pattern for GL and sucrose intakes in men.
In contrast, no clear pattern emerged across quintiles of GL,
sucrose, or fructose when stratifying on physical activity level.
The BMI and physical activity stratified analyses in the NHS
yielded no associations (data not shown).
In a recent study by Ma et al. (9) an interaction was
reported between plasma C-peptide and alcohol intake in
men (RR, 3.9; 95% CI, 1.3-11.7 for high C-peptide and high
alcohol intake versus low C-peptide and low alcohol intake).
Given these findings, there is a possibility that alcohol
exacerbates the effect of consuming a high GL or sugar diet.
We examined this in men and observed that among those
with elevated alcohol intake (z20 g/d), the risk of colorectal
cancer increased across quintiles of GL, sucrose, fructose and
carbohydrate intakes (Ps for tests of trend = 0.008, 0.02, 0.01,
and 0.04, respectively). In this strata of alcohol drinkers (z20
g/d), the RR for the highest quintile of GL was 2.11 (95% CI,
0.79-5.63), compared with the lowest quintile in that strata. In
contrast, trends were weak between dietary intakes of glucose
load or sugar and risk of colorectal cancer among men with
low alcohol consumption (data not shown). Among women
with a high alcohol intake (z20g/d), elevated GL, sucrose, or
fructose intakes were associated with slightly higher risks of
colorectal cancer, although these were not statistically
significant (highest versus lowest quintile: multivariate RRs,
1.29; 95% CI, 0.58-2.90 for GL; multivariate RR, 1.39; 95% CI,
0.62-3.10 for sucrose intake; multivariate RR, 1.56; 95% CI,
0.79-3.06 for fructose intake).
Discussion
We observed a 27% to 37% increase in the risk of colorectal
cancer with increasing intakes of carbohydrate, GL, sucrose or
fructose in men (Ps for test of trend were significant for all
except carbohydrate intake) but no associations were observed
in women. No statistically significant associations were
observed for distal or proximal cancer in either gender. In
the NHS I, stratifying by BMI, physical activity, or hormone
use did not change the associations. In men, stratifying by BMI
led to slightly stronger associations for those who were
overweight (BMI z 25) and weaker associations for those with
normal weight.
Although a recent cohort study observed no association for
dietary GL and colorectal cancer (21), another cohort study
(20) and two previous case-control studies reported elevated
risks of colorectal cancer with higher intake of GL (18, 19).
RRs for high versus low intake of GL ranged between 1.7 and
2.9 in these studies (18-20). A number of studies have also
reported significantly elevated risks of colorectal cancer for
individuals with a high, compared with low, sucrose intake
(39-44). The associations between dietary sucrose and
Table 2. Multivariate relative risk of colorectal cancer in
relation to intakes of GL, GI, carbohydrate, and fructose in
the NHS (1980-2000) and HPFS (1986-2000)
Quintiles of intake
P
12345 95% CI*
GI
Men
Index/dc
Colorectal
(n = 683)
Colon
(n = 552)
Rectal
(n = 131)
Women
Index/dc
Colorectal
(n = 1,096)
Colon
(n = 858)
Rectal
(n = 238)
GL
Men
Load/d
Colorectal
Colon
Rectal
Women
Load/d
Colorectal
Colon
Rectal
Carbohydrate
Men
g/d
Colorectal
Colon
Rectal
Women
g/d
Colorectal
Colon
Rectal
Sucrose
Men
g/d
Colorectal
Colon
Rectal
Women
g/d
Colorectal
Colon
Rectal
Fructose
Men
g/d
Colorectal
Colon
Rectal
Women
g/d
Colorectal
Colon
Rectal
697476
1.09
78
1.08
82
1.0 1.06 1.14 0.88-1.48 0.33
1.01.111.161.131.13 0.84-1.51 0.40
1.00.910.850.891.21 0.68-2.15 0.65
657174
1.02
77
1.00
81
1.00.84 1.08 0.87-1.34 0.27
1.0 0.821.10 1.041.06 0.83-1.36 0.29
1.00.91 0.75 0.881.14 0.73-1.78 0.70
131 158177195 223
1.0
1.0
1.0
1.13
1.07
1.32
1.36
1.42
1.07
1.36
1.32
1.48
1.32 0.98-1.79 0.04
1.25 0.88-1.25 0.11
1.61 0.82-3.17 0.17
80103119 137 167
1.0
1.0
1.0
1.03
1.06
0.93
1.10
1.08
1.19
0.91
0.79
1.40
0.89 0.71-1.11 0.15
0.89 0.69-1.15 0.11
0.87 0.52-1.44 0.95
182 214234256 288
1.0
1.0
1.0
1.11
1.05
1.31
1.21
1.23
1.09
1.24
1.22
1.29
1.27 0.93-1.72 0.11
1.21 0.85-1.71 0.20
1.45 0.73-2.38 0.34
110137 155174 202
1.0
1.0
1.0
1.07
1.07
1.09
1.11
1.09
1.18
0.94
0.88
1.18
0.87 0.68-1.11 0.15
0.86 0.65-1.13 0.14
0.91 0.53-1.55 0.78
263644
1.19
1.20
1.21
53
1.40
1.36
1.62
67
1.0
1.0
1.0
1.15
1.18
1.07
1.30 0.99-1.69 0.03
1.25 0.93-1.68 0.13
1.47 0.81-2.66 0.11
1726 33
0.98
1.00
0.90
41
0.98
0.98
0.97
55
1.0
1.0
1.0
1.05
1.17
0.69
0.89 0.72-1.11 0.10
0.99 0.78-1.26 0.49
0.62 0.39-0.99 0.17
2940 4856
1.26
1.25
1.28
72
1.0
1.0
1.0
1.02 1.19
1.01 1.25
1.13 0.99
1.37 1.05-1.78 0.008
1.38 1.03-1.86 0.02
1.31 0.72-2.38 0.33
22 33 4150
0.80
0.80
0.81
68
1.0
1.0
1.0
0.94 0.93
0.92 0.88
1.03 1.15
0.87 0.71-1.07 0.20
0.86 0.68-1.09 0.15
0.92 0.59-1.44 0.47
NOTE: Multivariate RRs are relative risks adjusted for age, family history of
colon cancer, prior endoscopy screening, aspirin use, height, BMI, pack-years of
smoking before age 30, physical activity, and intakes of cereal fiber, alcohol,
calcium, folate, processed meat and beef, pork or lamb as main dish.
*95% CI for q5 versus q1.
cIntakes are provided for medians of quintiles.
Cancer Epidemiology, Biomarkers & Prevention
141
Cancer Epidemiol Biomarkers Prev 2005;14(1). January 2005

Page 5

colorectal cancer risk were substantially stronger among
individuals with low physical activity and elevated BMI in
one study (18).
We have previously shown, in a subset of healthy women
from the NHS I, that our variable for GL (estimated from the
FFQs) predicts fasting plasma triacylglycerol and high-density
lipoprotein levels better than total carbohydrate intake does
(12). The association between triacylglycerol levels and GL was
even stronger among women with BMI >25 kg/m2(12),
indicating that overweight women are particularly susceptible
to the quality of the carbohydrates they consume, probably
because of some degree of insulin resistance. Recently in the
NHS I cohort, we observed an elevated risk of pancreatic
cancer with a higher GL, which was stronger among women
with BMI z 25 kg/m2(45). Glycemic load intake has also been
strongly associated with the risk of diabetes and coronary
heart disease in the NHS I (15-17). These findings indicate that
we are able to measure GI and load with the FFQs in these
cohorts, and that our null findings are unlikely the result of
substantial measurement error.
Although mean GL and carbohydrate intakes were lower in
women than in men, the range of intake was similar. In
addition, the mean and range of GL among women in this
study were similar to those reported in the Women’s Health
Study, where a significant positive association was reported
(20). Therefore, it is unlikely that our null findings in women
result from a narrow range of GL intake.
In a recent study in the NHS I, GL and fructose were
positively associated with plasma C-peptide levels (23). In
the present study, fructose intake was most strongly
associated with risk in men, supporting the insulin hypoth-
esis. We did not, however, find any associations in women.
There are several possible reasons for the gender differences.
One possibility is that men may be more susceptible to
insulin, based on stronger associations with obesity, partic-
ularly central obesity, in men than in women. Another
explanation may be that alcohol intake increases suscepti-
bility to dietary factors that influence insulin levels (as
suggested by our findings); in which case men would be at
higher risk than women (in these cohorts) because they
drink more alcohol (Table 1). Finally, the findings in men
may be due to chance.
Overall,ourresultssuggestthat theglycemicresponse todiet
may not play a major role in colorectal cancer. The GI was
developed to rank foods based on their response to blood
glucose concentrations. However, postprandial insulin
responses are not always proportional to blood glucose
concentrations or to a meal’s carbohydrate content (46).
Insulinotropic factors, such as protein- and fat-rich foods, may
induce substantial insulin secretion despite producing relative-
ly small blood glucose responses (47, 48). In a study comparing
glucose to insulin response to foods, the authors reported that
although the two scores were highly correlated, the glycemic
response accounted for only 23% of the variability in insuline-
mia (46). The findings suggest that an ‘‘insulin index of foods’’
might help increase the accuracy of estimating the insulin
response. Given that insulin has been shown to increase tumor
growth in the colonic epithelial cells in vitro studies, as well as
in rats (49), and hyperinsulinemia is a risk factor for colorectal
cancer (9-11), insulin is likely to be an important factor.
The strengths of this study include a prospective design,
detailed information on diet, and data on many potential risk
factors of colorectal cancer. The prospective design precluded
recall bias, which may have limited previous case-control
studies. In addition, by including data from two large cohorts,
we were able to examine results for men and women and had
sufficiently large numbers of cases to examine the data by
cancer site.
In summary, we observed a slight increase in colorectal risk
for dietary GL and sugar intakes in men, but not in women.
The associations were slightly stronger among men with
elevated BMI, but stratifying by BMI, physical activity or
hormone use did not change the associations among women.
Future studies should examine foods that closely predict
insulin secretion rather than glycemic response.
References
1.
Kaaks R, Lukanova A. Energy balance and cancer: the role of insulin and
insulin-like growth factor-I. Proc Nutr Soc 2001;60:91–106.
Giovannucci E. Insulin, insulin-like growth factors and colon cancer: a
review of the evidence. J Nutr 2001;131:3109–20S.
Nilsen TI, Vatten LJ. Prospective study of colorectal cancer risk and physical
activity, diabetes, blood glucose and BMI: exploring the hyperinsulinaemia
hypothesis. Br J Cancer 2001;84:417–22.
Hu FB, Manson JE, Liu S, et al. Prospective study of adult onset diabetes
mellitus (type 2) and risk of colorectal cancer in women. J Natl Cancer Inst
1999;91:542–7.
Will JC, Galuska DA, Vinicor F, Calle EE. Colorectal cancer: another
complication of diabetes mellitus? Am J Epidemiol 1998;147:816–25.
LeMarchandL,WilkensLR,KolonelLN,HankinJH,LyuL-C.Associationsof
sedentarylifestyle,obesity,smoking,alcoholuse,anddiabeteswiththeriskof
colorectal cancer. Cancer Res 1997;57:4787–94.
La Vecchia C, Negri E, Decarli A, Franceschi S. Diabetes mellitus
and colorectal cancer risk. Cancer Epidemiol Biomarkers Prev
1997;6:1007–10.
2.
3.
4.
5.
6.
7.
Table 3. Multivariate RR of colon cancer by site in relation
to intakes of GL, GI, carbohydrate, and fructose in the NHS
(1980-2000) and HPFS (1986-2000)
Quintiles of intake
P
12345 95% CI*
GI
Men
Proximal (n = 227)c
Distal (n = 228)
Women
Proximal (n = 403)c
Distal (n = 326)
GL
Men
Proximal
Distal
Women
Proximal
Distal
Carbohydrate
Men
Proximal
Distal
Women
Proximal
Distal
Sucrose
Men
Proximal
Distal
Women
Proximal
Distal
Fructose
Men
Proximal
Distal
Women
Proximal
Distal
1.0
1.0
0.96
1.30
1.15
1.09
1.18
1.01
0.99
1.06
0.63-1.57
0.67-1.68
0.72
0.91
1.0
1.0
0.85
0.81
1.19
1.02
1.15
0.92
1.11
0.91
0.77-1.58
0.63-1.34
0.27
0.82
1.0
1.0
0.97
1.01
1.35
1.22
1.06
1.15
1.10
0.87
0.64-1.88
0.51-1.49
0.67
0.85
1.0
1.0
1.06
0.90
0.96
1.14
0.58
0.89
0.77
0.90
0.53-1.11
0.60-1.36
0.02
0.59
1.0
1.0
1.17
0.88
1.26
0.99
1.09
1.01
1.11
0.86
0.64-1.93
0.50-1.47
0.81
0.73
1.0
1.0
1.06
1.04
0.95
1.25
0.67
1.10
0.63
0.95
0.42-0.95
0.61-1.50
0.01
0.85
1.0
1.0
1.18
1.12
1.16
1.09
1.49
1.23
1.17
1.14
0.73-1.89
0.72-1.79
0.41
0.54
1.0
1.0
0.93
1.60
0.79
1.31
0.75
1.23
0.73
1.34
0.51-1.03
0.89-2.02
0.04
0.60
1.0
1.0
0.97
1.14
1.25
1.15
1.33
1.07
1.19
1.30
0.74-1.91
0.82-2.04
0.31
0.34
1.0
1.0
1.09
0.70
0.98
0.79
0.70
0.85
0.83
0.79
0.58-1.17
0.55-1.14
0.06
0.48
NOTE: Multivariate RRs are relative risks adjusted for age, family history of
colon cancer, prior endoscopy screening, aspirin use, height, BMI, pack-years of
smoking before age 30, physical activity, and intakes of cereal fiber, alcohol,
calcium, folate, processed meat and beef, pork or lamb as main dish.
*95% CI for q5 versus q1.
cNumbers of proximal plus distal cases do not add up to colon cancer cases in
Table 2 because colon cases include those for which we could not identify
subsite.
Glycemic Load, Sugar, and Colorectal Cancer
142
Cancer Epidemiol Biomarkers Prev 2005;14(1). January 2005