Prediagnostic levels of C-peptide, IGF-I, IGFBP -1, -2 and -3 and risk of endometrial cancer


  • The content of this report is solely the responsibility of the authors and does not necessarily represent the official views of the National Cancer Institute. The NSHDS is sponsored by the Swedish Cancer Society and the ORDET cohort is sponsored by the Italian Association of Cancer Research.


Conditions related to chronic hyperinsulinemia, such as obesity, noninsulin dependent diabetes mellitus and polycystic ovary syndrome, are associated with an increased risk of endometrial cancer. Elevated plasma IGF-I and decreased levels of IGF-binding proteins have been shown to be associated with increased risk of several cancer types that are frequent in affluent societies. We investigated for the first time in a prospective study the association of pre-diagnostic blood concentrations of C-peptide (a marker of pancreatic insulin production), IGF-I, IGFBP-1, -2 and -3 with endometrial cancer risk. A case-control study was nested within 3 cohorts in New York (USA), Umeå (Sweden) and Milan (Italy). It included 166 women with primary invasive endometrial cancer and 315 matched controls, of which 44 case and 78 control subjects were premenopausal at recruitment. Endometrial cancer risk increased with increasing levels of C-peptide (ptrend = 0.0002), up to an odds ratio (OR) of 4.76 [95% confidence interval (CI) = 1.91–11.8] for the highest quintile. This association remained after adjustment for BMI and other confounders [OR for the top quintile = 4.40 (1.65–11.7)]. IGFBP-1 levels were inversely related to endometrial cancer [ptrend = 0.002; OR in the upper quintile = 0.30 (0.15–0.62)], but the association was weakened and lost statistical significance after adjustment for confounders [ptrend = 0.06; OR in the upper quintile = 0.49 (0.22–1.07)]. Risk was unrelated to levels of IGF-I, IGFBP-2 and IGFBP-3. Chronic hyperinsulinemia, as reflected by increased circulating C-peptide, is associated with increased endometrial cancer risk. Decrease in the prevalence of chronic hyperinsulinemia, through changes in lifestyle or medication, is expected to prevent endometrial cancer. © 2003 Wiley-Liss, Inc.

Endometrial cancer incidence rates are up to 10 times higher in economically developed, affluent countries than, for example, in Africa or rural areas of Asia.1 Within affluent countries, epidemiological studies have shown that obesity is a strong risk factor for endometrial cancer in both pre- and postmenopausal women.2 Overweight (a body mass index [BMI] > 25 kg/m2) and obesity (BMI > 30 kg/m2) together have been estimated to account for about 40% of endometrial cancer incidence in Western Europe.3

Excess weight is associated with a number of metabolic consequences, the major ones being the development of insulin resistance and hyperinsulinemia.2 Other conditions characterized by insulin resistance and increased blood insulin concentrations, such as noninsulin-dependent diabetes mellitus and polycystic ovary syndrome (PCOS), have also been related to increased endometrial cancer risk.4, 5, 6

Insulin has been shown to promote the growth of cancer cell lines in vitro, including endometrial cancer cells,7 and direct evidence for its role in the pathogenesis of endometrial cancer comes from several case-control studies. In one large study,8 elevated levels of C-peptide (a marker of pancreatic insulin secretion) were related to an increase in endometrial cancer risk, although this association did not persist after adjustment for BMI. Four other very small studies showed higher fasting, post-glucose challenge insulin9, 10, 11 or glycosylated hemoglobin levels12 in endometrial cancer patients than in control women.

Insulin-like growth factor (IGF)-I has well-documented mitogenic and anti-apoptotic properties in endometrial and other tissues,13, 14 and has been implicated in the development of several common cancers.15 Estrogens, the major established risk factor for endometrial cancer, increase IGF-I and IGF-I receptor expression in the uterus, and IGF-I is proposed to mediate their mitogenic effect on the endometrium (“estromedin hypothesis”).16, 17 The biological activity of IGF-I is modulated by IGF-binding proteins (BP) that control the availability of IGF-I to its cellular receptors.18 IGFBP-1 is the most extensively studied IGF-binding protein in endometrial tissue and its synthesis within the uterus is stimulated by progesterone.19 Progesterone is known to counteract estrogen-induced cell proliferation in the endometrium20 and this effect of progesterone could be largely due to an increase in IGFBP-1 synthesis.17, 21 Insulin, in contrast, reduces the synthesis of IGFBP-1 in the endometrium and other tissues.22

In our report, we describe the findings of a case-control study nested within 3 collaborative cohort studies in New York (USA), Umeå (Sweden) and Milan (Italy). This is the first prospective study to report on the association of prediagnostic serum levels of C-peptide, IGF-I, IGFBP-1, -2 and -3 with endometrial cancer risk.


Study cohorts

The collaborating cohorts have previously been described in detail23, 24, 25 and include the New York University Women's Health Study (NYUWHS), the Northern Sweden Health and Disease Study (NSHDS), and the Study of Hormones and Diet in the Etiology of Breast Cancer (ORDET). The main characteristics of these cohorts are presented in Table I.

Table I. Characteristics of the Cohort Studies Included in the Pooled Analysis of C-PEPTIDE, IGF-I, IGFBP-1, -2, -3 and Endometrial Cancer
CohortStudy settingRecruitment periodCohort sizeAge range at enrolmentLast date of complete follow-upNumber of casesNumber of controlsMedian age (range) of the study subjects at recruitment
NYUWHS New York USABreast cancer screening centre1985–199114,27534–65February 19989117357.1 (35.6–66.9)
NSHDS Umeå SwedenGeneral population1986–200143,26830–70December 20006011458.3 (49.1–69.6)
ORDET Milan ItalyHealthy volunteers & women attending breast cancer prevention unit1987–199210,78835–70January 1997152854.6 (36.3–66.6)

In all 3 cohorts, demographic and anthropometric data were collected at enrolment, through self-administered questionnaires in the NYUWHS and the NSHDS, and questionnaires administered by trained nurses (who also performed the anthropometric measurements) in the ORDET. Collection of data on reproductive history, use of oral contraceptives (OC), hormone replacement therapy (HRT) and smoking varied according to study cohort. In the NYUWHS, data on reproductive history were collected at enrolment, whereas data on smoking, OC and HRT use up to the index date (date of diagnosis of the case) were collected through telephone interviews of cases (88%) and matched controls (83%) or using the most recent follow-up questionnaire for participants who did not complete the interview (resulting in data available on 93% of the cases and 96% of the controls). In the NSHDS, smoking information was collected at enrolment. A questionnaire on reproductive life and sex hormone use was administered prospectively to 47% of the subjects and a similar questionnaire was sent out retrospectively to all cases and matched controls to complete and update the collected information at baseline (response rate: 95%). In ORDET, these data were collected at enrolment. Information about ever having a diagnosis of diabetes mellitus or intake of drugs for treatment of diabetes was collected at baseline in the 3 cohorts and from follow-up questionnaires (NYUWHS).

In all 3 cohorts, participants were asked to donate a venous blood sample. Subjects that reported hormone use at baseline were not recruited in the NYUWHS and ORDET cohorts, and potential case and control subjects from the NSHDS were excluded if there was an indication that they were using exogenous hormones at the time of blood donation. In the ORDET cohort, blood samples were collected after an overnight fast and the majority of NSHDS participants fasted for at least 4 hr before donating a blood sample (65% of the subjects had fasted for at least 4 hr and 32% for 8 or more hr). In NYUWHS, nonfasting samples were collected, but the times of last food intake and blood donation were recorded (27% of the subjects had fasted for at least 4 hr and 14% for 8 or more hr). Control subjects were not matched to their index case for fasting status at blood donation.

The NSHDS and ORDET components of the study included only Caucasian subjects. In the NYUWHS, self-reported information about ethnic origin was available for 86% of the subjects included in the present study; of these, 86% indicated that they were nonHispanic Whites, 6% Black, 6% Hispanic and 2% other ethnicity.

Identification of endometrial cancer cases and selection of control subjects

Case subjects were cohort members with primary invasive endometrial cancer diagnosed 6 or more months after the initial blood donation, without preceding cancer diagnosis, and who were identified within the parent cohort by the date of the last complete follow-up. A total of 166 endometrial cancer cases were included for the study from the 3 cohorts (Table I).

Of the 137 malignancies with data about histology available from the parent cohorts, 91 (66%) were endometrioid, 10 (7%) serous, 3 (2%) mucinous and 1 (1%) clear-cell types; 2 (1%) were of mixed type, and 30 (22%) were carcinomas or adenocarcinomas not otherwise specified.

For each case subject, 2 control subjects were selected at random from appropriate risk sets. The risk set for a given case included all cohort subjects alive, free of cancer, who had not had a hysterectomy at index date, and who matched the case on cohort, date (± 3 months), age (± 6 months) and menopausal status at enrolment. The matching for menopausal status was confirmed by follicle-stimulating hormone (FSH) measurement in women younger than age 60 when questionnaire data were equivocal or missing, classifying women as postmenopausal if their FSH measurement was > 12.75 IU/L. The selection of the FSH cut-off point was based on data provided by the kit manufacturer and an analysis of the distribution of FSH levels according to menopausal status as assessed by questionnaire, as well as according to age (< 42 and > 55 years), using data from more than 300 women from the 3 cohorts. Thirty-two control subjects were replaced because they did not meet this criterion; however for 4 matched case-control sets, control replacement was not feasible. A total of 315 control subjects were identified and included in our study (Table I).

The Ethical Review Boards of New York University School of Medicine, the University of Umeå, Istituto Nazionale Tumori in Milan and the International Agency for Research on Cancer, in Lyon, France, periodically reviewed and approved the present study.

Laboratory analyses

The hormone analyses were performed on serum samples obtained from the NYUWHS subjects and heparinized plasma samples obtained from the NSHDS and ORDET subjects.

The laboratory analyses were performed by the laboratory of the Hormones and Cancer Group at the International Agency for Research on Cancer, Lyon, France. The samples pertaining to each matched case-control set were always analyzed with the same assay kit, in the same laboratory batch and on the same day. Laboratory personnel were unable to distinguish between case and control samples. To control the quality of the hormone measurements, samples from a pool of quality control plasma and 3 standard sera were inserted randomly in each batch. C-peptide and IGFBP-2 concentrations were measured by radioimmunoassays and IGF-I, IGFBP-1 and IGFBP-3 by immunoradiometric assays, all reagents from Diagnostic System Laboratories (Webster, TX). The IGF-I assay included an acid-ethanol precipitation of IGFBPs.

The mean intra-batch coefficients of variation were 6.5% for a C-peptide concentration of 2 ng/ml, 1.8% for an IGF-I concentration of 150 ng/ml, 2.5% for an IGFBP-1 concentration of 15 ng/ml, 4.5% for an IGFBP-2 concentration of 400 ng/ml, 1.7% for an IGFBP-3 concentration of 4,000 ng/ml and 4.2% for an FSH concentration of 10 IU/L.

Statistical analysis

Peptide and BP data were log-transformed to reduce departures from the normal distribution. A paired t-test (case value vs. mean of matched control subjects) was used to compare differences in continuous variables of interest between case and control subjects.26 Generalized linear models were used to calculate means and 95% CI and to compare mean hormone levels in subgroups of interest adjusting for potential confounders. Pearson partial correlations were calculated on the basis of sums of squares error estimates. Intra-class correlations between repeated steroid hormone measurements, available for a subset of NYUWHS subjects only, were calculated from variance components estimated by the SAS “MIXED” procedure,27 and their 95% confidence intervals (CI) were calculated as previously described.28

Odds ratios (OR) for disease by quintile levels of the hormone variables were estimated by conditional logistic regression models using the SAS “PHREG” procedure. The cut-points for quintiles of hormone concentrations were based on the distribution of the cases and controls combined. Likelihood ratio tests were used to assess linear trends in ORs over the quintiles, giving quantitative scores of 1, 2, 3, 4 and 5 to the 5 levels. All statistical tests and corresponding p-values were 2-tailed, and p-values <0.05 were considered statistically significant.


The time from recruitment and blood donation to cancer diagnosis for the case subjects ranged from 6.2 months to 14.4 years, with a median of 4.7 years. Mean age at cancer diagnosis was 61 (± 7.8) years. Ninety-six percent of the case subjects were diagnosed at least a year after cohort recruitment. At recruitment, 25% of the subjects (44 cases and 78 controls) were premenopausal. Case subjects were substantially heavier and had a higher BMI than controls and tended to report less use of OCs and more nulliparity (Table II). There were no significant differences between case and control subjects in the mean ages at menarche and menopause, use of HRT, ever smoking cigarettes or diabetes diagnosis.

Table II. Distribution of Risk Factors for Endometrial Cancer in Cases and Controls
 Cases (n = 166)Controls (n = 315)p for case-control difference2
  • 1

    Mean (95% confidence intervals) adjusted for age and study cohort.

  • 2

    Conditional logistic regression (on continuous variables or never vs. ever for the categorical variables).

  • 3

    ORDET subjects and those with missing values from the NYUWHS and NSHDS cohorts were not included.

Height (cm)1161.1 (160.1–162.2)160.8 (160.0–161.6)0.71
Weight (kg)170.8 (68.7–73.0)65.4 (63.7–67.1)0.0001
BMI kg/m2127.3 (26.5–28.0)25.3 (24.7–25.9)0.0001
Age at menarche112.9 (12.7–13.2)13.1 (12.9–13.3)0.29
Full-term pregnancy   
 No39 (23%)54 (17%)0.06
 Yes123 (74%)251 (80%) 
 Missing4 (2%)10 (3%) 
OC use   
 No117 (70%)204 (65%)0.06
 Yes44 (27%)106 (34%) 
 Missing5 (3%)5 (2%) 
Age at menopause150.9 (50.0–51.7)50.1 (49.4–50.7)0.24
HRT use   
 No107 (64%)227 (72%)0.22
 Yes53 (32%)83 (26%) 
 Missing6 (4%)5 (2%) 
ERT use3   
 No113 (80%)229 (85%)0.16
 Yes28 (20%)39 (15%) 
 No74 (45%)133 (42%)0.99
 Yes25 (15%)47 (15%) 
 Ex-smoker36 (22%)69 (22%) 
 Missing values31 (19%)66 (21%) 
 No147 (89%)291 (92%)0.22
 Yes12 (7%)13 (4%) 
 Missing7 (4%)11 (3%) 

A subset of the NYUWHS subjects included in the current study (n = 76) have donated a second blood sample at a return visit to the recruiting center. The lag-time between the baseline and second visit ranged from 11 to 65 months (median duration between visits 14 months). Intra-class correlations between repeated peptide and BP measurements in these samples were 0.66 (0.51–0.77) for IGF-I, 0.67 (0.53–0.78) for IGFBP-1, 0.30 (0.08–0.49) for IGFBP-2 (n = 68), 0.86 (0.79–0.91) for IGFBP-3 and 0.58 (0.41–0.71) for C-peptide.

Pearson correlations adjusted for cohort study, age at blood donation and menopausal status between BMI and levels of C-peptide were: r = 0.15, p = 0.01 among all controls, r = 0.24, p = 0.003 among controls who fasted for at least 4 hr and 0.55, p < 0.0001 among controls who fasted for at least 8 hr before blood donation. BMI correlated inversely with levels of IGFBP-1 (r = −0.30, p < 0.0001) and IGFBP-2 (r = −0.30, p < 0.0001) but exhibited no linear correlation with IGF-I and IGFBP-3. C-peptide correlated inversely with levels of IGFBP-1 (r = −0.40, p < 0.0001) and IGFBP-2 (r = −0.25, p < 0.0001), directly with levels of IGFBP-3 (r = 0.23, p < 0.0001) and did not correlate with levels of IGF-I. There was a direct correlation between levels of IGF-I and IGFBP-3 (r = 0.45, p < 0.0001) and between levels of IGFBP-1 and -2 (r = 0.41, p < 0.0001).

Mean peptide and BP concentrations exhibited no significant differences between ever and never users of OC or HRT, smoking status at baseline, or ever having had a full-term pregnancy. Women who reported a diagnosis of diabetes had higher blood concentrations of C-peptide than women without such diagnosis (4.36 vs. 3.03 ng/ml, p = 0.001).

Mean hormone levels in case and control subjects are presented in Table III. Case subjects had higher mean C-peptide and lower IGFBP-1 levels than control subjects, but similar mean IGF-I and IGFBP-2 and -3 levels.

Table III. Mean (95% CI) C-Peptide, IGF-I, IGFBP-1, -2 and -3 Levels in Endometrial Cancer Cases and Controls, Adjusted for Cohort Study, Age at Sampling, Menopausal and Fasting (<4 Hr, 4–8 Hr, >8 Hr) Status at Blood Donation
HormoneCasesControlsCase-control difference: p-value*
Number of observationsMean (95% C.I.)Number of observationsMean (95% C.I.)
  • *

    Paired t-test (case value vs. the mean of the 2 controls).

C-peptide (ng/ml)1663.38 (2.99–3.77)3132.92 (2.59–3.25)0.01
IGF-I (ng/ml)166169.0 (154.5–183.5)314176.6 (164.4–188.8)0.27
IGFBP-1 (ng/ml)16629.2 (25.5–32.8)31432.9 (29.8–36.0)0.002
IGFBP-2 (ng/ml)164416 (332–500)313474 (403–545)0.10
IGFBP-3 (ng/ml)1654,005 (3,869–4,141)3123,937 (3,824–4,051)0.55

There was a strong direct dose-response relationship between C-peptide concentration and endometrial cancer risk, reaching more than 4-fold increase in risk in the top quintile of C-peptide levels (Table IV). This increase in risk was attenuated after adjustment for BMI categories, but the dose-response pattern and the statistical significance of the test for trend and for odds ratio in the top quintile of C-peptide remained (Table IV). Similarly, adjustment for levels of IGFBP-1 reduced the point estimates, but the association of C-peptide with endometrial cancer risk remained significant (data not shown).

Table IV. Odds Ratios for Endometrial Cancer by Quintiles of C-Peptide, IGF-I, IGFBP-1, -2 and -3
 Quintilep for trend*
  • 1

    Conditional logistic regression on case-control pairs matched for study cohort, age and date at recruitment into the study, menopausal status and day of menstrual cycle for premenopausal women; all C-peptide, IGFBP-1 and IGFBP-2 models were adjusted also for fasting (<4 hr, 4–8 hr, ≥8 hr)

  • 2

    Adjusted for BMI (in categories ≤23, 23–25, 25–30 and ≥30, missing), parity, OC and HRT use

  • *

    Linear trends in ORs over the quintiles by assigning quantitative scores (1, 2, 3, 4 and 5)

C-peptide (ng/ml)      
Crude model11.001.14 (0.56–2.31)1.00 (0.45–2.22)2.52 (1.13–5.36)4.76 (1.91–11.8)0.0002
Adjusted model21.001.12 (0.53–2.37)0.99 (0.43–2.29)1.98 (0.84–4.68)4.40 (1.65–11.7)0.003
Number of cases/controls29/6728/6726/7239/5644/51 
Median quintile values0.711.622.293.465.90 
IGF-I (ng/ml)      
Crude model11.000.98 (0.55–1.75)0.69 (0.37–1.30)0.55 (0.28–1.07)0.72 (0.38–1.37)0.12
Adjusted model21.000.87 (0.46–1.66)0.79 (0.40–1.53)0.61 (0.30–1.26)0.90 (0.44–1.82)0.54
Number of cases/controls37/5938/5831/6527/6933/63 
Median quintile values92.6134.7170.9218.6309.6 
IGFBP-1 (ng/ml)      
Crude model11.000.67 (0.37–1.21)0.33 (0.17–0.65)0.56 (0.30–1.07)0.30 (0.15–0.62)0.002
Adjusted model21.000.89 (0.46–1.74)0.39 (0.19–0.79)0.71 (0.35–1.43)0.49 (0.22–1.07)0.06
Number of cases/controls41/5237/5925/7135/6125/71 
Median quintile values4.3811.720.133.354.7 
IGFBP-2 (ng/ml)      
Crude model11.000.50 (0.26–0.96)0.59 (0.32–1.09)0.61 (0.32–1.18)0.54 (0.27–1.06)0.20
Adjusted model21.000.50 (0.25–0.99)0.67 (0.34–1.30)0.73 (0.36–1.50)0.81 (0.38–1.74)0.98
Number of cases/controls41/5429/6631/6332/6331/63 
Median quintile values25.0199332490834 
IGFBP-3 (ng/ml)      
Crude model11.001.44 (0.78–2.66)1.19 (0.63–2.26)0.96 (0.49–1.90)1.96 (0.94–4.10)0.35
Adjusted model21.001.66 (0.85–3.22)1.45 (0.72–2.94)1.11 (0.53–2.34)2.41 (1.07–5.45)0.20
Number of cases/controls29/6637/5832/6326/6941/55 
Median quintile values3,0013,5373,9154,3985,145 

The case and control subjects were not matched for fasting status, and to explore the effect of fasting, we ran unconditional logistic regression models adjusting for all matching variables. The results were similar to the conditional regression models [ORs for the 2nd to 5th quintile: 1.14 (0.56–2.32), 1.03 (0.48–2.23), 2.31 (1.08–4.95) and 3.75 (1.63–8.65)]. Then we split the study population according to the fasting status of the women at blood donation. The ORs for the top C-peptide tertile for the fasting subjects [2.38 (1.00–5.68)] was similar to that of the nonfasting subjects [2.16 (0.88–5.28)] and to that of the whole study population [1.95 (1.06–3.58)]. The association in postmenopausal women appeared to be stronger than in premenopausal women, but a formal test for interaction was not significant.

BMI categories (≤ 22.5, 22.5–25, 25–30, <30) were directly related to endometrial cancer risk with ORs of 0.73 (0.40–1.34), 1.49 (0.84–2.63) and 2.57 (1.37–4.80) for the 3 highest BMI categories, ptrend < 0.0003. Adjustment for C-peptide only slightly reduced the odds ratio estimates for BMI categories (data not shown).

Endometrial cancer risk was not significantly related to blood levels of IGF-I, IGFBP-2 and IGFBP-3, neither before nor after adjustment for potential confounding variables (Table IV). The molar ratio of IGF-I/IGFBP-3 was inversely related to risk of endometrial cancer [OR for the top quintile 0.51 (0.24–1.09), ptrend 0.03]; however, the association lost statistical significance after adjustment for confounders [OR for the top quintile 0.62 (0.27–1.43), ptrend = 0.23].

Levels of IGFBP-1 were inversely related to endometrial cancer risk, but the association was weakened and lost statistical significance after adjustment for BMI, ever having had a full-term pregnancy, OC and HRT use. However, in postmenopausal women at recruitment, the association was stronger and remained significant after adjustments for BMI and other confounders [0.36 (0.13–0.95) for the top quintile, ptrend = 0.04]. Adjustment of the crude conditional logistic regression IGFBP-1 models for C-peptide did not abolish the significant inverse association of this peptide with endometrial cancer risk in the whole study population [OR of 0.36 (0.17–0.77) for the top quintile of IGFBP-1, ptrend < 0.02] or in analyses restricted to postmenopausal women at blood donation [OR of 0.29 (0.12–0.71), ptrend < 0.01].

Restricting the analyses to women diagnosed 2 or more years after blood donation or to those without history of diabetes did not influence substantially the direction and strength of any of the associations between levels of C-peptide, IGF-I, IGFBP-1, -2 and -3 and endometrial cancer risk. Very similar results were obtained when categories of hormone exposure, used for the conditional logistic regression models, were defined according to the hormone distribution in each study cohort separately or when an FSH value of 30 IU/L (a more conservative estimate) was used as a cut-off point to classify women as postmenopausal.


In this first prospective study, we found a strong, direct, dose-response association of endometrial cancer risk with prediagnostic levels of C-peptide. Prediagnostic levels of IGFBP-1 were inversely associated with risk, but after adjustment for confounders, the association remained statistically significant only in the group of postmenopausal women. Levels of IGF-I, IGFBPs -2 or –3 were not related to risk.

The leading and widely accepted hypothesis for the pathogenesis of endometrial cancer is that of “unopposed estrogens.”20, 29 This hypothesis states that exposure to estrogens insufficiently counterbalanced by progesterone increases the mitotic activity of the endometrial cells and the risk of developing endometrial carcinoma. There is evidence to suggest that before menopause endometrial neoplasia is related especially to progesterone deficiency (as observed in women with chronic anovulatory states), while, after menopause, when ovarian progesterone production ceases altogether, cancer risk is directly related to estrogen levels4, 20 (Fig. 1). It has been proposed that the proliferative effects of estrogens on the endometrial tissue are mediated by an increase in the local production of IGF-I,16 whereas the anti-proliferative effects of progesterone are proposed to be largely due to progesterone-induced increases in the local production of IGFBP-1.17

Figure 1.

Endogenous hormones and endometrial cancer development.

In our study, prediagnostic blood concentrations of IGF-I were not related to risk. The lack of association is unlikely to be due to low measurement reproducibility of blood IGF-I concentrations as several studies, including ours, have shown a good to high reproducibility of blood IGF-I measurements over time.30, 31 Several studies have shown that women with BMI ≥ 25 have lower total IGF-I levels in comparison with women with BMI between 20 and 25 kg/m,2,15, 32 while levels of IGBP-3 are relatively stable across BMI categories or may slightly increase with BMI.33, 34, 35 The prevalence of overweight and obesity in our study was substantially higher in case than in control subjects (58% vs. 39%, respectively) and it is plausible to assume that the observed tendency for inverse associations of IGF-I levels and of IGF-I/IGFBP-3 molar ration and of the direct association of IGFBP-3 with endometrial cancer in the crude models were a reflection of the confounding effect of obesity. Finally, circulating IGF-I levels may not be a good marker for the local expression and action of IGF-I in the endometrium. Unlike most tissues where the key stimulus for the synthesis of IGF-I and IGFBP-3 is growth hormone,18, 22 the major determinant of IGF-I levels in endometrium is estradiol.16, 36, 37

Before menopause, endometrial synthesis of IGFBP-1 is the result of the opposing effects of insulin (which inhibits synthesis) and progesterone (which stimulates synthesis) [Fig. 1]. The endometrial IGFBP-1 mRNA expression follows the pattern of progesterone synthesis during the menstrual cycle; it is minimal during the proliferative phase but reaches peak levels during late luteal phase. In contrast, after cessation of the cyclic ovulatory activity, it is likely that insulin remains the main determinant of endometrial and other tissue IGFBP-1 synthesis.17, 38 Hence, the observed stronger inverse association between IGFBP-1 levels and endometrial cancer, which persisted after adjustments for BMI and other confounders or levels of C-peptide, in the group of women who were postmenopausal at recruitment may be due to the better correlation between circulating and endometrial tissue IGFBP-1 levels in these women.

The major finding of our study is the strong direct association of prediagnostic C-peptide concentrations, as markers of pancreatic insulin secretion, with endometrial cancer risk. Insulin may influence endometrial cancer risk by a number of mechanisms (Fig. 1). First, as a growth factor, it can stimulate cell proliferation and inhibit apoptosis directly through insulin receptors.7, 39 Second, insulin may increase IGF-I bioactivity in many tissues, including the endometrium, by down-regulating the synthesis of IGFBP-1.16, 17 Third, among postmenopausal women in whom the negative feed-back mechanisms regulating estrogen synthesis are no longer operating, insulin induced inhibition of hepatic synthesis of sex hormone-binding globulin, results in an increases in the free estradiol levels.2 Finally, chronically elevated insulin concentrations are a major contributing cause of ovarian androgen excess, which in premenopausal women may cause chronic anovulation and progesterone deficiency.4 Indeed, chronic hyperinsulinemia is a key feature of women who have PCOS, a hyperandrogenic syndrome that affects up to 8% of premenopausal women and that has been related to increased risk of endometrial cancer.4, 6, 40 In women who have PCOS, plasma insulin concentrations correlate with levels of androstenedione and testosterone,2 and more severe insulin resistance and hyperinsulinemia are related to more frequent anovulatory menstrual cycles.4 The central position of insulin as a cause of androgen excess is illustrated by experiments in vitro showing that insulin stimulates androgen production by ovarian stromal tissue and that by observations that in women with PCOS insulin-lowering drugs can cause reductions in androgen levels and restore ovulatory menstrual cycles.2, 4

One of the major causes of insulin resistance and hyperinsulinemia is excess weight and high BMI has been strongly associated with risk of developing endometrial cancer.2 In our analyses, the strength of the association between C-peptide and endometrial cancer risk was lowered by adjustment for BMI, but C-peptide concentrations remained strongly associated with cancer risk, indicating that hyperinsulinemia may be a risk factor for endometrial cancer that is independent of obesity. Conversely, BMI remained directly associated with endometrial cancer risk after adjustment for levels of C-peptide, probably due to the direct effect of obesity on sex-steroid hormone concentrations (e.g., in postmenopausal women adipose tissue is the major site where estrogens are synthesized).

Our results differ somewhat from those from a large case-control study (165 cases and 180 controls) by Troisi et al.,8 who also observed an increased risk among women with elevated serum C-peptide, but the association was abolished after adjustment for BMI. The major difference between the 2 studies is that we measured C-peptide concentrations in prospectively collected samples (at least 6 months before cancer diagnosis), while for the study of Troisi et al., serum samples, obtained after endometrial cancer diagnosis of the cases were available. The prospective design of our study minimizes “inverse causation” bias, which may occur if the metabolic factors studied as potential determinants of cancer risk are in fact themselves subject to changes induced by the presence of a tumor, or by the consequences of the tumor diagnosis (e.g., stress, anticancer treatment). Women with previous diabetes diagnosis were not included in the study of Troisi et al., while we did not exclude women with diabetes from our study. Restriction of our analyses to women without previous diabetes diagnosis slightly weakened the association between C-peptide levels and endometrial cancer risk but did not abolish it. However, only self-reported information about diagnosis of diabetes was available and it is possible that the real prevalence of diabetes in our population was higher than estimated by the questionnaire data. Another difference was that for the study Troisi et al. samples from fasting subjects were available, while in our study only half of the participants have fasted for 4 or more hr. Fasting conditions are known to influence the circulating concentrations of C-peptide and IGFBP-1, as also observed in our data. However, unconditional logistic regression models (adjusted for all matching variables) showed very similar ORs in the groups of fasting and nonfasting women at blood donation.

In conclusion, our data support an etiological role for elevated insulin (measured as C-peptide concentrations) and decreased IGFBP-1 levels in the pathogenesis of endometrial cancer that is independent of obesity. Circulating concentrations of IGF-I, IGFBP-2 and IGFBP-3 do not appear to be related to endometrial cancer risk.


We thank Ms. Y. Afanasyeva, Ms. L. Quinones and Ms. D. Masciangelo who provided technical assistance in the NYU Women's Health Study; Ms. Å. Ågren, Mr. H. Sjodin and Mrs. L Marklund who helped with the management of the Swedish biobank database; Ms. D. Del Sette. Dr. E. Meneghini and Dr. E. Mugno for helping with the ORDET database; Mr. D. Achaintre and Ms. J. Bouzac who contributed to the laboratory analyses and for the help of Mrs. J. Dehedin and Mrs. S. Somerville in the article preparation.