Body mass index (BMI) was calculated as weight in kilograms divided
by the square of height in meters. CI indicates confidence interval.
Sun Ha Jee, Heechoul Ohrr, Jae Woong Sull, Ji Eun Yun, Min Ji, Jonathan M. Samet. Fasting Serum Glucose Level and Cancer Risk in Korean Men and Women. JAMA. 2005;293(2):194–202. doi:10.1001/jama.293.2.194
Author Affiliations: Department of Epidemiology
and Health Promotion, Graduate School of Public Health (Dr Jee and Mss Yun
and Ji), Department of Preventive Medicine and Public Health, College of Medicine
(Dr Ohrr), and Department of Public Health, Graduate School (Dr Sull), Yonsei
University, Seoul, Korea; and Department of Epidemiology, Johns Hopkins Bloomberg
School of Public Health, Baltimore, Md (Drs Jee and Samet).
Context Diabetes is a serious and costly disease that is becoming increasingly
common in many countries. The role of diabetes as a cancer risk factor remains
Objective To examine the relationship between fasting serum glucose and diabetes
and risk of all cancers and specific cancers in men and women in Korea.
Design, Setting, and Participants Ten-year prospective cohort study of 1 298 385 Koreans (829 770
men and 468 615 women) aged 30 to 95 years who received health insurance
from the National Health Insurance Corp and had a biennial medical evaluation
in 1992-1995 (with follow-up for up to 10 years).
Main Outcome Measures Death from cancer and registry-documented incident cancer or hospital
admission for cancer.
Results During the 10 years of follow-up, there were 20 566 cancer deaths
in men and 5907 cancer deaths in women. Using Cox proportional hazards models
and controlling for smoking and alcohol use, the stratum with the highest
fasting serum glucose (≥140 mg/dL [≥7.8 mmol/L]) had higher death rates
from all cancers combined (hazard ratio [HR], 1.29; 95% confidence interval
[CI], 1.22-1.37 in men and HR, 1.23; 95% CI, 1.09-1.39 in women) compared
with the stratum with the lowest level (<90 mg/dL [<5.0 mmol/L]). By
cancer site, the association was strongest for pancreatic cancer, comparing
the highest and lowest strata in men (HR, 1.91; 95% CI, 1.52-2.41) and in
women (HR, 2.05; 95% CI, 1.43-2.93). Significant associations were also found
for cancers of the esophagus, liver, and colon/rectum in men and of the liver
and cervix in women, and there were significant trends with glucose level
for cancers of the esophagus, colon/rectum, liver, pancreas, and bile duct
in men and of the liver and pancreas in women. Of the 26 473 total cancer
deaths in men and women, 848 were estimated as attributable to having a fasting
serum glucose level of less than 90 mg/dL. For cancer incidence, the general
patterns reflected those found for mortality. For persons with a diagnosis
of diabetes or a fasting serum glucose level greater than 125 mg/dL (6.9 mmol/L),
risks for cancer incidence and mortality were generally elevated compared
with those without diabetes.
Conclusion In Korea, elevated fasting serum glucose levels and a diagnosis of diabetes
are independent risk factors for several major cancers, and the risk tends
to increase with an increased level of fasting serum glucose.
Diabetes mellitus is a serious and costly disease that is becoming increasingly
common in many countries, including Korea.1 Recent
data show that approximately 150 million people have diabetes mellitus worldwide,
and this number may double by 2025, especially in developing countries, because
of population growth, aging, unhealthy diets, obesity, and sedentary lifestyles.2 The association between diabetes mellitus and cardiovascular
mortality is well established.3 However, while
the role of diabetes as a risk factor for cancer is still uncertain, having
diabetes or an elevated glucose level is of interest because of the effect
of insulin on cell growth and of the systemic inflammation associated with
diabetes and the metabolic syndrome.4,5
Recent observational studies have provided consistent evidence on associations
of diabetes with increased risk of cancers of the pancreas,5 liver,6 endometrium,7 and colon/rectum.8 Data on cancers of the esophagus, stomach, prostate,
and breast are more limited and have been inconsistent.8,9 This
lack of consistency may be attributable to the limited number of studies and
their small sample sizes.
We conducted a prospective cohort investigation, the Korean Cancer Prevention
Study (KCPS), among more than 1 million Koreans to assess associations of
fasting serum glucose and of a diagnosis of diabetes with cancer risk.10 In addition, we explored modification of this risk
by obesity in a population with a low average body weight compared with those
of Western countries.
The National Health Insurance Corp (NHIC), previously the Korean Medical
Insurance Corp, provides health insurance to government employees, teachers,
and their dependents. Of the Korean population (approximately 43.7 million
in 1992), 10.7% were insured by this organization, including 1 297 833
workers and 3 364 605 dependents. All workers were required to participate
in biennial medical examinations.11 In 1992,
94% of the insured workers completed examinations; a total of 95% completed
biennial examinations in 1994. For dependents, the numbers were 37% in 1993
and 24% in 1995. This examination included a lifestyle and medical questionnaire,
along with measurement of blood chemistries in a fasting blood sample.
The KCPS cohort includes 1 329 525 Koreans (846 907 men
and 482 618 women) aged 30 to 95 years who received health insurance
from NHIC and who had biennial medical evaluations during the period 1992
to 1995. By year, 784 870 (59.0%) were enrolled in 1992, 367 903
(27.7%) in 1993, 98 417 (7.4%) in 1994, and 78 335 (5.9%) in 1995.10 Of the 1 329 525 participants, 27 192
(2.1%) with missing information were excluded. A total of 3719 people reporting
a history of any form of cancer and 229 who died in the interval between questionnaire
completion and start of follow-up on January 1 of the subsequent year were
also excluded, leaving a final sample size of 1 298 385.
The NHIC biennial examinations, conducted by medical staff at local
hospitals, follow a standard procedure. In the 1992, 1993, 1994, and 1995
questionnaires, participants were asked (1) to describe their smoking habits,
along with some other health habits, including alcohol consumption, and (2)
if they were being treated for diabetes, with the date of diagnosis if they
answered yes. Serum glucose measurements were obtained under fasting conditions
for routine clinical purposes. Each hospital had internal and external quality
control procedures directed by the Korean Association of Laboratory Quality
Control. All follow-up time of the participants was assigned to the first
recorded serum glucose level.
The follow-up period was up to 10 years, through December 31, 2002.
The exact dates of completion of the survey form were not recorded. Consequently,
follow-up accrual began on January 1 of the calendar year following the year
in which the survey form was completed. Because the study involved routinely
collected medical data, participant consent was not specifically obtained.
The study was approved by the institutional review boards of Yonsei University
and the Johns Hopkins Bloomberg School of Public Health.
The principal outcome variables were mortality from cancer by site and
cancer incidence, based on national cancer registry data and hospitalization
records. Although Korea has a national cancer registry, reporting was not
complete during the time of follow-up and, consequently, hospital admission
files were used to identify a first admission event for cancer. An incident
cancer case was coded as occurring based on either a positive report from
the national cancer registry or on a hospital admission for a cancer diagnosis.
Outcomes for mortality were ascertained from the causes of death on the death
certificates. A computerized search of death certificate data from the National
Statistical Office in Korea was performed using the unique identification
number assigned at birth. Causes of death are assigned by trained hospital
abstractors; for this analysis, we used underlying cause.
Fasting serum glucose levels were categorized as less than 90, 90 to
109, 110 to 125, 126 to 139, and 140 or more mg/dL (<5.0, 5.0-6.0, 6.1-6.9,
7.0-7.7, and ≥7.8 mmol/L). For some analyses, the upper categories of fasting
glucose levels were combined into a single stratum because of small numbers.
In all primary analysis, the fasting glucose category of less than 90 mg/dL
(<5.0 mmol/L) was the reference group. Additionally, we created a category
for diabetes combining participants with self-reported treatment for diabetes
or with fasting serum glucose levels greater than 125 mg/dL (6.9 mmol/L).12,13
Age-adjusted death and cancer incidence rates were calculated for each
category of fasting serum glucose level and directly standardized to the age
distribution of the 1995 Korean national population. All analyses were stratified
by sex. We computed hazard ratios (HRs; the hazard for mortality in a specific
fasting glucose category divided by the corresponding hazard in the reference
category [<90 mg/dL]) using Cox proportional hazards modeling14 to
adjust for age and other potential confounding factors. In the Cox model,
smoking status was included as a dummy variable for previous smokers and current
smokers, with 3 categories of amount smoked (1-9, 10-19, and ≥20 cigarettes/d),
and alcohol use as 5 categories (none, 1-24, 25-49, 50-99, and ≥100 g/d).
We also calculated the attributable numbers of deaths associated with having
elevated serum glucose levels, using standard methods.15 All
analyses were conducted using SAS statistical software, version 8.0 (SAS Institute
Inc, Cary, NC).
The population was primarily middle-aged, with approximately twice as
many men as women (Table 1). By the
criteria of self-report and fasting serum glucose level, the prevalence rates
of diabetes were about 5% in men and 4.5% in women. The population had a low
body mass index (BMI; calculated as weight in kilograms divided by the square
of height in meters) on average, and only 23.8% and 0.8% of men and 27.0%
and 2.5% of women had BMI values of 25 or above and above 30, respectively.
Both smoking and alcohol use were substantially more common in men.
During the 10 years of follow-up, 54 385 deaths occurred among
men and 20 362 among women. As shown in Figure 1, fasting serum glucose level was positively associated
with all-cause mortality rates. In the adjusted Cox proportional hazards model,
this effect persisted, with persons in the highest stratum of fasting serum
glucose (≥140 mg/dL) having a higher HR for all causes combined (HR, 2.09;
95% confidence interval [CI], 2.03-2.16 in men and HR, 2.35; 95% CI, 2.24-2.48
in women) compared with the lowest stratum (<90 mg/dL) (Table 2 and Table 3).
A total of 20 566 cancer deaths occurred among men and 5907 deaths
occurred among women during the 10 years of follow-up. In general, HR estimates
were above unity for male participants in the higher strata of serum glucose
level and among those with a diagnosis of diabetes (Table 2). In fact, all point estimates were somewhat elevated in
association with a diagnosis of diabetes. We explored the relationship between
duration of diabetes and cancer risk and did not find consistent associations,
with the exception of pancreatic cancer in men. Among men with diabetes, the
HRs for pancreatic cancer death with diabetes durations of less than 4.9 years,
5.0 to 9.9 years, and 10 years or more were 2.0 (95% CI, 1.2-3.3), 2.4 (95%
CI, 1.4-4.3), and 3.0 (95% CI, 1.8-5.0), respectively, compared with those
We observed linear trends in mortality with increasing fasting serum
glucose level for all cancers combined and for cancers of several sites (Table 2). Compared with the reference category
(<90 mg/dL), men with a fasting serum glucose level above 140 mg/dL had
significantly elevated HRs of death from cancers of the esophagus (HR, 1.44;
95% CI, 1.08-1.93), liver (HR, 1.57; 95% CI, 1.40-1.76), pancreas (HR, 1.91;
95% CI, 1.52-2.41), and colon/rectum (HR, 1.31; 95% CI, 1.03-1.67). Significant
associations were also found for bladder cancer and leukemia for those with
a fasting serum glucose level of 126 mg/dL or higher. Men with a fasting serum
glucose level of 110 to 125 mg/dL had significantly elevated HRs of death
from esophageal, stomach, colon/rectal, liver, and pancreatic cancers. Of
the total of 20 566 cancer deaths in men, 802 were estimated as attributable
to having a fasting serum glucose level of less than 90 mg/dL.
For women, the overall pattern of association was similar to that in
men (Table 3), with all point estimates
increased for those with diabetes. Significant positive linear trends in death
rates were observed for pancreatic cancer; HRs ranged from 1.70 (95% CI, 1.17-2.46)
at fasting serum glucose levels of 110 to 125 mg/dL to 2.05 (95% CI, 1.43-2.93)
at fasting serum glucose levels of 126 mg/dL or higher (Table 3). Significant associations with diabetes were also found
for cancers of the liver, lung, breast, and cervix, while associations were
not observed for risk of death from cancers of the stomach or colon/rectum.
Of the total 5907 cancer deaths in women, 46 were estimated as attributable
to having a fasting serum glucose level less than 90 mg/dL.
The numbers of incident cases during the 10 years of follow-up were
37 759 among men and 16 074 among women. Trends were generally similar
for mortality and incidence. We observed positive linear trends in cancer
incidence with increasing fasting serum glucose levels for cancers of the
liver, pancreas, and kidney (Table 4).
However, fasting serum glucose level was inversely associated with prostate
cancer incidence among men with a fasting glucose level of at least 126 mg/dL.
The association of fasting serum glucose level and cancer incidence was similar
whether the analysis was based on the total population or on those who had
at least a 5-year follow-up period (data not shown). Furthermore, the incidence
findings were unchanged with adjustment for BMI. For example, men with diabetes
had similar HRs of death from pancreatic cancer before (HR, 1.71; 95% CI,
1.42-2.06) and after (HR, 1.73; 95% CI, 1.42-2.07) BMI adjustment.
In women, positive linear trends were observed in incidence of pancreatic
cancer with increasing fasting serum glucose level (Table 5). We observed a significant association for liver cancer
for the stratum with fasting serum glucose levels of 110 to 125 mg/dL. The
observed associations in women were unchanged with adjustment for BMI (data
To control for potential confounding by obesity and to evaluate effect
modification, the data were further stratified by BMI (<20, 20 to <23,
and ≥23). Because of limited numbers for some cancer sites, the analyses
were carried out for all sites combined and for cancers of the colon/rectum,
liver, and pancreas. For all cancers, a trend was evident in mortality risk
by fasting serum glucose level in each stratum of BMI (Figure 2). Positive trends in death rates were observed for cancers
of the liver and pancreas in all BMI groups as well. The risk of death due
to liver and pancreatic cancers associated with fasting serum glucose level
was not modified by BMI. For cancer incidence, the association of fasting
serum glucose level with risk for liver cancer remained consistent and lacked
evidence of effect modification by obesity. The association of pancreatic
cancer incidence with fasting serum glucose level was not as consistent as
for mortality, although all strata showed increased risk compared with the
We also assessed potential modification of the effect of serum glucose
level by smoking and alcohol consumption. We did not find significant interactions
for cancer incidence or mortality in either men or women.
This cohort of Koreans, in comparison with Western populations, is notable
for the low frequency of obesity in its participants. The average BMI was
23.2, and only one fourth of participants had a BMI above 25. Based on other
studies in Korea, almost all cases of diabetes could be expected to be type
2.16 Nonetheless, we documented, as in Western
populations, that serum glucose level and presence of diabetes are associated
with cancer incidence and mortality.
Other studies have addressed diabetes or glucose intolerance and risk
for cancer. Although diverse in design and in their measures of glucose intolerance,
the majority have shown that increased cancer risk, either overall or for
particular cancer sites, is associated with glucose intolerance. The magnitude
of the association of glucose intolerance or diabetes with risk for all cancers
was small in the KCPS but was within the range found in some studies.5,6,17 The multicancer site
effect is consistent with postulated mechanisms of systemic consequences of
hyperinsulinemia.4,5 In interpreting
the findings of these studies and of the KCPS, potential confounding by obesity
is of concern. In the KCPS, we found that the increased cancer risk associated
with high serum glucose or diabetes was unchanged when controlling for BMI;
additionally, most KCPS participants were not overweight. The association
of serum glucose level with cancer risk did not vary by BMI (Figure 2).
The study confirmed the excess risk of digestive cancers reported in
several studies,5,6,8 particularly
of cancers of the pancreas, liver, esophagus, and colon among persons with
diabetes. For pancreatic cancer, a 1995 meta-analysis of cohort and case-control
studies estimated a 2-fold increase in risk of pancreatic cancer, comparing
patients with and without diabetes.18 The KCPS
estimates were similar, and the HR increased with increasing fasting serum
glucose level (Table 2 and Table 3), as found in a Chicago cohort study.4 Our prospective results, along with the unchanged
findings with exclusion of the first 5 years of follow-up, weigh against the
possibility that the presence of pancreatic cancer increases blood glucose
levels; ie, reverse causality. In the KCPS, we observed a significant positive
linear trend for risk of pancreatic cancer with fasting glucose level: the
HR was 1.7 (95% CI, 1.4-2.1) in men with diabetes and 1.5 (95% CI, 1.2-1.9)
when data were restricted to men with at least 5 years of follow-up time.
The complementary findings with these exposure measures, serum glucose level,
and report of diabetes, and the presence of a dose-response relationship with
fasting serum glucose level support a causal interpretation of these associations.
The findings for several other cancer sites further support this interpretation.
In men, we found increased risks of cancers of the esophagus, liver, stomach,
colon/rectum, kidney, and bladder and leukemia (Table 2), while in women, risk was increased for cancers of the
liver, lung, breast, and cervix (Table 3).
Other studies have also found increased risk for these cancer sites.19,20 In Korea, hepatitis B infection is
common and is an important cause of liver cancer.21 Serum
glucose level was not associated with hepatitis B surface antigen status in
the subset of participants with assay results, indicating that hepatitis B
does not confound the results for liver cancer. On hospital admission diagnoses,
we found little mention of nonalcoholic steatotic hepatitis, suggesting that
the increased risk of liver cancer associated with higher serum glucose level
reflects a direct pathway rather than an indirect pathway through obesity
and fatty liver damage.
For prostate cancer, for both incidence and mortality, we found no evidence
for an association with either fasting serum glucose level or diabetes
(Table 2 and Table 4),
consistent with other reports. One large, population-based Swedish cohort
study found that men with diabetes had a 10% lower risk of developing prostate
cancer than the general male population.22 A
reduced risk for men with diabetes was found in a case-control study in New
York.23 In contrast, the American Cancer Society’s
Cancer Prevention Study showed no association between diabetes at baseline
and prostate cancer mortality,24 and a study
of incidence in 823 participants in the Baltimore Longitudinal Study of Aging
also found association with fasting insulin and glucose levels.20 The
lack of association for prostate cancer weighs against observation bias as
contributing to the positive associations for other sites; the medical care
and frequent blood chemistries associated with diabetes might be expected
to increase the opportunity for detecting prostate cancer, but the negative
association is inconsistent with such bias.
Hyperinsulinemia has been cited as a possible risk factor for breast
cancer, and supporting laboratory findings have been reported,5 but
results of epidemiological studies have been mixed.25 In
a recent publication based on the Nurses’ Health Study, women with type
2 diabetes had a small increase in risk (HR, 1.17; 95% CI, 1.01-1.35). The
association was apparent among postmenopausal women but not among premenopausal
women.25 We did not find an association, although
95% CIs for our estimates covered the value from the Nurses’ Health
Study. With stratification at age 55 years, we did not find increased risk
in the older stratum, corresponding to an age range when most women would
The potential limitations of our study result primarily from using data
collected for clinical purposes. Serum glucose was measured under fasting
conditions using clinical laboratories operating with standard quality assurance
and control protocols in place. A single measurement of fasting serum glucose
made for clinical purposes is used as a diagnostic standard and matches the
World Health Organization’s recommended approach for epidemiological
studies.26 We further relied on self-report
of a diagnosis of diabetes; the serum glucose level of those with reported
history of diabetes was 66 mg/dL higher than for those not reporting diabetes,
suggesting that the self-reported information was valid. We do not anticipate
that these clinical data would artifactually introduce association of cancer
risk with fasting serum glucose level.
Cancer mortality is subject to misclassification on death certificates,
particularly with regard to attribution to a particular site. The limitations
of death certificate data on cancer have been characterized in some countries,27,28 but we are uncertain as to the applicability
of the findings of these studies in Korea. A small study within KCPS showed
high validity for a death certificate listing of lung cancer.29 Additionally,
we have found that of KCPS participants with incident liver cancer, 73% of
their deaths were attributed to liver cancer over a follow-up interval of
at least 5 years. In Korea, cancer registration is not yet complete nationwide;
it is currently estimated at 90%.30 Consequently,
we used hospital admission for cancer as a further indication of cancer incidence.
We were able to take potential confounding by smoking and alcohol consumption
into account and explored effect modification by obesity. We cannot attribute
the associations of fasting serum glucose level and diabetes to uncontrolled
confounding, particularly given the dose-response relationships observed with
While the generalizability of the findings is uncertain, we have shown
that fasting serum glucose level and diabetes are associated with cancer risk
in a population far leaner than the Western populations in other studies.
These associations do not reflect confounding by obesity, suggesting that
the mechanism of increased cancer risk reflects the consequences of hyperinsulinemia.
Glucose intolerance may be one pathway by which obesity increases cancer risk,
and rising obesity may increase future cancer rates.
Corresponding Author: Sun Ha Jee, PhD, MHS,
Department of Epidemiology and Health Promotion, Graduate School of Public
Health, Yonsei University, Seoul, Korea (firstname.lastname@example.org, email@example.com).
Author Contributions: Dr Jee had full access
to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Jee, Ohrr, Samet.
Acquisition of data: Jee, Yun.
Analysis and interpretation of data: Jee, Sull,
Drafting of the manuscript: Jee, Yun, Ji, Samet.
Critical revision of the manuscript for important
intellectual content: Ohrr, Sull.
Statistical analysis: Jee, Sull, Yun, Ji.
Obtained funding: Samet.
Study supervision: Ohrr, Samet.
Funding/Support: This study was funded by grant
1R03 CA94771-02 from the National Cancer Institute.
Role of the Sponsor: The funding source had
no role in the design and conduct of the study, in the collection, analysis,
and interpretation of the data, or in the preparation, review, or approval
of the manuscript.
Acknowledgment: We thank the staff of the Korean
National Health Insurance Corp. We also thank Walter Willett, MD, DrPH, Frederick
Brancati, MD, MHS, and Elizabeth Platz, ScD, MPH, for their helpful comments
and Charlotte Gerczak, MLA, for editorial assistance.