Postmenopausal Levels of Endogenous Sex Hormones and Risk of Colorectal Cancer

Most observational studies have found that postmenopausal hormone replacement therapy (HRT) is associated with a reduced risk of colorectal cancer. A meta-analysis of 18 epidemiologic studies reported that women who had ever used HRT had a 20% lower risk [relative risk, 0.80; 95% confidence interval (95% CI), 0.74-0.86] compared with women who had never taken such hormones (1). The risk reduction was observed in both case-control and prospective cohort studies. Confirmation of an inverse association between use of hormonal therapy and risk of colorectal cancer was provided by the Women's Health Initiative (WHI) clinical trial in which women randomized to receive estrogen + progestin HRT had a 37% lower risk of colorectal cancer than women receiving placebo [hazard ratio (HR), 0.63; 95% CI, 0.43-0.92; ref. 2]. In women who had had a hysterectomy, who were randomized to receive either estrogen alone (without progestin) or a placebo, a protective effect was not observed (HR, 0.93; 95% CI, 0.75-1.15; ref. 3).

This protective effect of exogenous estrogen + progestin raises the possibility that higher levels of endogenous estrogen, and/or progesterone, may be associated with a reduced risk of colorectal cancer. To date, only one prospective study has reported on the association between endogenous circulating estrogen levels and subsequent colorectal cancer development. A recent study of 438 colorectal cancer cases and a subcohort of 816 women from the WHI Observational Study (WHI-OS), which includes women who were ineligible for, or not interested in, the clinical trial portion of that study, found, contrary to expectation, an increased risk of colorectal cancer in women with higher levels of endogenous estradiol, despite the fact that women using oral HRT had a reduced risk of developing the disease (4).

In this study, we investigated the association between prediagnostic endogenous concentrations of estradiol, estrone, progesterone, and sex hormone–binding globulin (SHBG) in postmenopausal women and risk of subsequent colorectal cancer. We conducted a nested case-control study within the prospective New York University Women's Health Study (NYUWHS) cohort.

The NYUWHS enrolled 14,274 women ages 35 to 65 y who attended a mammography breast screening clinic in New York City between 1985 and 1991. A total of 7,054 women (49.4% of the cohort) were postmenopausal at enrollment. The cohort has been described in detail previously (5, 6). Women who were pregnant or using exogenous hormones (oral contraceptives or HRT) within the 6 mo before enrollment were not eligible for inclusion. At enrollment, participants answered a self-administered questionnaire with questions about their medical and reproductive history, use of medications, and demographic characteristics. Participants donated 30 mL of peripheral venous blood at enrollment, and aliquots of serum were stored at −80°C.

Cases of incident malignant disease are identified through self-administered questionnaires mailed to participants every 2 to 4 y, with telephone follow-up of nonrespondents. Medical records are retrieved for reported malignancies and reviewed to confirm colorectal cancer diagnoses. Active follow-up is supplemented by record linkages to state cancer registries in New York, New Jersey, and Florida. Ascertainment of vital status and cause of death is accomplished through record linkage with the National Death Index. A capture-recapture analysis estimated the cancer ascertainment rate to be 95% in the NYUWHS cohort (7).

Participants in this study of hormones and colorectal cancer were limited to women who were postmenopausal at the time of enrollment. Women were classified as postmenopausal if they reported not having a menstrual period in the 6 mo before blood donation or having had a bilateral oophorectomy. Women who reported having had a hysterectomy without bilateral oophorectomy before natural menstrual period cessation were classified as postmenopausal if they were at least 52 y of age at the time of enrollment.

Incident cases of invasive colorectal cancer occurring after blood donation and before July 1, 2003 (the end of the last complete round of follow-up through questionnaires and state cancer registry record linkages), were eligible for inclusion in this study. Women who had a previous cancer (other than nonmelanoma skin cancer) were excluded. In total, 148 cases met the eligibility criteria. Two control participants per case were randomly selected from among the members of the cohort who were alive and free of cancer as of the date of diagnosis of the case and matched the case on age at enrollment (±6 mo) and date of blood donation (±3 mo). This study was approved by the Institutional Review Board at the New York University School of Medicine.

Laboratory assays for estrone, estradiol, and progesterone were conducted at the Quest Diagnostics Nichols Institute (San Juan Capistrano, CA). Estrone and estradiol were measured by RIA following organic extraction and celite chromatography. The lower limit of quantitation (LLOQ) was 10 pg/mL for estrone and 2 pg/mL for estradiol. Progesterone was measured using liquid chromatography tandem mass spectrometry (LC/MS/MS assay; LLOQ, 100 pg/mL). SHBG was measured in the laboratory of Dr. Mortimer Levitz at the New York University School of Medicine using a two-site immunometric chemiluminescent assay on an IMMULITE 2000 instrument (Diagnostic Products Corp.). The SHBG assay has a LLOQ of 2 nmol/L.

Samples from all members of a case-control matched set were assayed together on the same well-plate. Duplicate samples were included for 10% of the subjects for quality control purposes. Samples were arranged and labeled in such a way that laboratory personnel were unaware of the case-control status of the samples and the identity of the duplicates. The mean intrabatch and interbatch coefficients of variation were, respectively, 14.9% and 16.9% for estrone, 33.5% and 36.3% for estradiol, and 3.5% and 3.5% for SHBG. Because progesterone values were below the LLOQ for more than 93% of our samples, data for this hormone could not be analyzed and the association between this hormone and risk of colorectal cancer could not be evaluated. The midpoint between the LLOQ and zero was used for 35 participants who had values below the LLOQ for estrone and 7 participants for estradiol. Measurements of estradiol were unavailable for 122 subjects (forty 1:2 matched sets plus one 1:1 matched set) because of lack of sufficient serum volume to rerun the assay after a batch failure. Therefore, these 122 subjects, along with 4 subjects in other batches who did not have sufficient sample material for the assay, are not included in the estradiol analyses.

Hormone concentrations were log₂ transformed to reduce departures from the normal distribution and to yield odds ratios (OR) associated with a doubling in hormone levels (8). We used the square root transformation for SHBG, which resulted in a better approximation to the normal distribution than log transformation and has been used by others (9, 10). Conditional logistic regression, which takes into account the matched study design, was used to calculate the P values comparing cases and controls. Spearman correlation coefficients, which do not require normality, were calculated between estrone, estradiol, SHBG, and body mass index (BMI). Free estradiol concentrations were calculated from the total estradiol and SHBG values according to the mass action laws, assuming a constant serum albumin concentration of 43 g/L (11). Free estradiol was very highly correlated with total estradiol (r = 0.96 in controls) and did not add any information to the analysis, so results for free estradiol are not presented in this article.

Estrone and SHBG variables were divided into quartiles, whereas estradiol was divided into tertiles due to the frequency of missing data and the resultant smaller sample size. Quantile cutpoints were based on the distribution of values in the controls. ORs and corresponding 95% CIs for quantiles of estrone, estradiol, and SHBG were estimated using conditional logistic regression. Tests for trend across quantiles were conducted using the median value of the quantile as the category score. The biomarkers were also analyzed as continuous variables. Potential confounding effects of BMI (kg/m²) at enrollment, regular use of aspirin before index date (three or more times per week for 6 mo or longer), use of HRT (ever/never up to index date), number of full-term pregnancies, use of oral contraceptives (ever/never up to index date), smoking at baseline (current, former, never), education (high school, college, more than college), first-degree family history of colorectal cancer (yes, no), alcohol use at baseline (none, less than seven drinks per week, seven or more drinks per week), and physical activity at baseline (tertiles of vigorous activity and of walking/light activity) were considered by including these variables in the conditional logistic regression models. To adequately compare the adjusted and unadjusted models, we restricted the study population to the subjects who did not have missing data for a given covariate and then compared models in this restricted population to observe the effect of the covariate on OR estimates. BMI was the only covariate that appreciably altered the risk estimates and is included in the adjusted models along with the matching factors. The other potential confounders did not alter the ORs of colorectal cancer associated with hormone levels and are not included in the models.

Analyses were repeated within tumor location subgroups (colon, rectum) and after excluding women who were diagnosed within 5 y of blood donation to reduce potential effects of preclinical disease on any observed associations between hormones and colorectal cancer risk. The relationship between hormones and colorectal cancer risk was also evaluated within subgroups of BMI (<25, ≥25 kg/m²) and within never users of HRT by breaking the matching and using unconditional logistic regression models adjusted for the matching factors and BMI. Because women using HRT within 6 mo of cohort enrollment were not eligible for inclusion in the cohort, only 30 cases and 77 controls in this study had used HRT by diagnosis/index date, which limited our ability to meaningfully evaluate the details of HRT use, such as time of use and type of HRT. All analyses were conducted using the Statistical Analysis System software (version 9.1, SAS Institute).

Descriptive characteristics of the cases and controls are shown in Table 1. The age range of both case and control subjects at blood donation was 53 to 64 years. Cases were less likely than controls to use aspirin on a regular basis (17% versus 29%; P = 0.002) and were more likely to drink alcohol (16% versus 8% consumed at least seven drinks per week; P = 0.02). Cases were also more likely to have a first-degree family member with colorectal cancer (30% versus 17%; P = 0.003).

Although cases and controls did not differ significantly with regard to the other potential confounders shown in Table 1, differences were in the expected direction according to previous studies. Cases had higher BMI than controls (25.6 versus 24.7 kg/m², respectively). Additionally, cases were less likely to have children than controls (77% versus 80% ever pregnant) and cases who did have children were likely to have fewer than the controls (62% cases had two or more children versus 68% of controls; P = 0.15). A lower proportion of cases used HRT (22% versus 28% ever users; OR, 0.7; 95% CI, 0.4-1.1). Although average endogenous levels of SHBG and estradiol were not significantly different between cases and controls, estrone values were statistically significantly higher in cases (21 versus 18 pg/mL; P = 0.02) than in controls.

Estimates of Spearman correlation coefficients among hormones and BMI are reported in Table 2 for the control group. Briefly, estrone and estradiol were strongly positively correlated (r = 0.55). BMI was positively correlated with both estrogens (0.34 ≤ r ≤ 0.42); SHBG was negatively correlated with BMI (r = −0.50) and slightly less so with estrone and estradiol (r ≈ −0.26). Correlations in cases were similar in direction and magnitude to correlations in controls (data not shown).

OR estimates from the conditional logistic regression models are shown in Table 3. The ORs for colorectal cancer were elevated in the highest quartiles of estrone (3rd quartile OR, 1.7; 95% CI, 0.96-3.0; 4th quartile OR, 1.8; 95% CI, 1.0-3.3; P for trend = 0.02 on the continuous scale). Adjustment for BMI at enrollment somewhat attenuated these ORs, but a positive association between estrone and colorectal cancer risk was still apparent (P for trend = 0.09 on the continuous scale). Adjusting for aspirin use, family history of colorectal cancer, and alcohol use (each considered individually and together) resulted in a less than 5% change in the ORs, and, therefore, we did not adjust for these covariates in our final models. There were no statistically significant associations between estradiol or SHBG and risk of colorectal cancer; however, the estradiol measurements were missing for 29% of the population.

As shown in Table 4, the positive association between estrone levels and risk of colorectal cancer persisted after the exclusion of women diagnosed within 5 years of blood donation (n = 32 cases), suggesting that the association was not attributable to existing preclinical disease. Estrone analyses were repeated after stratifying on BMI. The positive association between estrone and risk in overweight and obese (BMI ≥ 25 kg/m²) women was somewhat stronger than the relationship for women with lower BMI but the test for interaction was not significant (P for interaction = 0.90). When analyses were restricted to women who had never used HRT, the ORs still showed an elevated risk for women with higher estrone levels (OR_t3-t1, 1.8; 95% CI, 0.9-3.6; Table 4). Stratifying by colon (n = 125 cases) versus rectal (n = 23 cases) cancer did not yield statistically heterogeneous results (P for interaction = 0.83), and for colon cancer, the ORs increased across tertiles of estrone with a 70% higher risk for the third tertile compared with the first tertile (OR, 1.7; 95% CI, 0.9-3.1; P for trend = 0.10).

In the early 1980s, McMichael and Potter (12) proposed the hypothesis that female sex hormones may be protective against colorectal cancer, citing the protective effect of parity, a potential protective effect of high-dose oral contraceptives, and biological evidence that sex hormones modify hepatic cholesterol metabolism and reduce bile acids. Since then, observational studies (1) and randomized controlled trials (2, 13) have found a protective effect of exogenous hormonal therapy on risk of colorectal cancer. Exogenous and endogenous estrogens have been shown to stimulate cell proliferation and tumor development in the breast and uterus. These protumorigenic effects are thought to be mediated through the estrogen receptor subtype ERα, which is predominant in these tissues (14). It has been hypothesized that estrogens may be protective against colorectal cancer because the predominant estrogen receptor subtype in the colon, ERβ, has been shown to inhibit transcription and induce apoptosis on binding estrogen in vitro (14-16). The current study was designed to test the hypothesis that endogenous estrogen levels are associated with a reduced risk of colorectal cancer.

In this nested case-control study within the NYUWHS cohort, we evaluated the association between endogenous estrogens and colorectal cancer risk in postmenopausal women. Contrary to our initial hypothesis, we observed a positive association between levels of estrone and risk of colorectal cancer. These results were consistent in subgroups according to tumor site (colon/rectum), BMI, and lag time between blood donation and diagnosis. We were unable to reproduce the WHI-OS finding of an association between estradiol and risk of colorectal cancer. However, the lack of association in our study may be due to the large number of participants with missing data for estradiol and the substantial laboratory error in the measurements (average intrabatch coefficient of variation = 33.5%). We also found a nonsignificant inverse association between SHBG and risk of colorectal cancer. This result further argues against a protective effect of estrogens: According to the free hormone hypothesis, only non–SHBG-bound estrogens are available to enter target cells; if estrogens were protective, higher levels of SHBG, leading to reduced levels of bioavailable estrogens, would be expected to be positively associated with risk of colorectal cancer.

Our finding of a positive association between endogenous estrone and colorectal cancer was unexpected given the substantial evidence that exogenous estrogens are protective. Despite this apparent contradiction, the ORs for estrone and risk of colorectal cancer in our study were in the same direction and of similar magnitude to the HRs reported in the WHI-OS for risk of colorectal cancer by tertile of estradiol (HR_t2-t1, 1.63; 95% CI, 1.14-2.46 and HR _t3-t1, 1.43; 95% CI, 0.95-2.16; ref. 4). Our observation of the effect of estrone on risk of colorectal cancer is consistent with a linear trend of increasing risk with increasing levels of estrone (P for trend = 0.02); yet, the similarity of the ORs for the two highest quartiles versus the lowest quartile of estrone is also compatible with a plateau effect, similar to what was observed in the WHI-OS, where the risk was elevated in the second tertile but did not increase further in the highest tertile of estradiol.

After controlling for the effects of BMI, the effect of estrone on risk was reduced and only marginally significant (P for trend on the continuous scale = 0.09). We adjusted for BMI because it may influence colorectal cancer risk through biological pathways other than hormonal, including the insulin and insulin-like growth factor-I (IGF-I) pathways. It could be argued, however, that adjusting for BMI results in overadjustment because adipose tissue is the location of aromatization of androgens to estrogens in postmenopausal women and therefore BMI is antecedent to estrone on the causal pathway to colorectal cancer. For this reason (17-19), results from both BMI-adjusted and unadjusted models are presented.

Gunter et al. (WHI-OS) provided two possible explanations for why oral hormonal therapy may be protective for colorectal cancer, whereas endogenous estrogen has an adverse effect: (a) oral administration of hormonal therapy exposes the liver to a large bolus of estrogen, which results in reduced synthesis of potentially harmful hepatic proteins, such as IGF-I, IGF-binding protein-3, insulin, and low-density lipoprotein, thereby reducing the risk of colorectal cancer development, or (b) because the most commonly prescribed HRT preparations contain estrone rather than estradiol, risk reduction by use of hormonal therapy may be due to antiproliferative effects of estrone seen on colonic cells in vitro (4). Endogenous estrone levels were not measured in the WHI-OS. However, our findings support an adverse effect of estrone similar to the adverse effect of estradiol observed by Gunter et al.

The other possible explanation for the discrepant results proposed by Gunter et al. is that oral estrogens affect the hepatic synthesis of various proteins (“first-pass” effect). In particular, the production of IGF-I and insulin, which are associated with an increased risk of colorectal cancer (20), is reduced by oral estrogens. Only a few studies have compared the effects of oral and transdermal (where there is no first-pass effect) hormonal therapy on colorectal cancer risk, and results have been conflicting (21-23). Future comparisons of the effects of oral and transdermal hormone therapies may shed light on the relevance of the first-pass effect to explain the reduced risk of colorectal cancer observed with HRT.

Another potential explanation for the opposite effects of endogenous and exogenous estrogens is that, because the WHI clinical trial only found a protective effect of hormonal therapy composed of estrogen + progestin and found no effect for estrogen-only therapy, progestin is the protective agent (2, 3). In the current study, we planned to evaluate the hypothesis that endogenous progesterone is associated with a reduced risk of colorectal cancer, but 93% of our study subjects had progesterone values below the LLOQ for the LC/MS/MS assay (100 pg/mL). Other investigators have used a RIA that has a lower LLOQ to measure progesterone in postmenopausal women (24). It should be noted, however, that the RIA is less specific than the LC/MS/MS method and therefore that the levels measured by RIA are likely to be overestimated. Although we were not able to examine the association with progesterone levels below 100 pg/mL, our results confirm that progesterone levels in postmenopausal women are very low, well below levels observed in women taking estrogen + progestin preparations. In light of this and findings from in vitro studies that progesterone does not influence growth of colon cells (25), it is unlikely that endogenous levels of progesterone in postmenopausal women contribute strongly to preventing CRC risk. However, we cannot discount a possible role for exogenous progestins, which may interact with other steroid receptors, including the androgen receptor, to influence colorectal cancer risk (26).

There are several biologically plausible reasons why endogenous estrogens may be associated with increased risk of colorectal cancer. In vitro studies, although somewhat inconsistent, have reported mitogenic and tumorigenic actions of estrogens on colorectal cancer cells (25, 27-30). Furthermore, reduced enzyme-mediated inactivation of estradiol has been observed in colorectal cancer tissues compared with normal tissues (31, 32) and may indicate that malignant colorectal cancer cells are exposed to higher levels of endogenous estradiol. Although epidemiologic studies are not entirely consistent, several have shown that obesity, which is associated with higher levels of endogenous estrogen, is associated with moderately elevated colorectal cancer risk (19).

A limitation of this study is that estrogen and SHBG levels were measured in samples collected at a single point in time and may not reflect changes occurring between blood collection and diagnosis. However, in a temporal reproducibility study using repeated blood samples from a subset of the NYUWHS postmenopausal participants, we found the within-subject variation over a 2- to 3-year period to be substantially lower than the between-subject variation (intraclass correlations were 0.64 for estradiol, 0.65 for estrone, and 0.88 for SHBG), indicating that a single measurement is a reasonably reliable representation of an individual's average long-term level, relative to other individuals (33).

The estradiol batch failure was the result of an instrumentation malfunction and unrelated to sample quality or quantity. Matched sets were randomly allocated into batches, therefore decreasing the likelihood that the samples in the failed batch were systematically different from those in successful batches. To evaluate how the exclusion of participants with missing estradiol measurements may have influenced the results, we repeated the main analyses for estrone in the subsample of women for whom estradiol measurements were available and found no differences in the results for estrone in this subgroup versus the entire study sample. This suggests that the exclusion of samples from the failed batch may not have substantially influenced the estradiol results. However, the high degree of laboratory error (coefficient of variation = 33.5%) in the estradiol measurements decreases our confidence in the estradiol results and prevents us from ruling out an association between estradiol and risk of colorectal cancer.

The NYUWHS cohort study was initiated in a breast cancer screening setting. Participants in the cohort are mostly Caucasian, middle class, and more likely to be health conscious than the general population, which should be taken into consideration when generalizing the results of this study. Information collected from the participants at baseline was primarily related to breast cancer, and thus, for some colorectal cancer risk factors (e.g., family history), data were collected only through follow-up questionnaires. Biases that are common in retrospective studies (i.e., those that collect information about participants after disease diagnosis) may have influenced the measurement of these covariates in participants who had already been diagnosed with disease when they completed follow-up questionnaires. However, inclusion of the potential confounders as covariates in the regression models did not appreciably affect risk estimates (except for BMI, which was calculated at enrollment for all participants).

One of the strengths of this study is that all participants included in the NYUWHS were free of cancer and not using any exogenous hormones for at least 6 months before blood donation, ensuring that measurements of endogenous hormones were not influenced by hormonal medications or existing clinical disease. The latter was further confirmed by an analysis limited to cases that were diagnosed at least 5 years after blood donation.

In conclusion, we found some evidence of an association between SHBG levels in postmenopausal women and risk of colorectal cancer; however, this association was attenuated by adjustment for BMI. We did not find an association between estradiol and risk, although the large amount of missing data and measurement error for estradiol prevents us from ruling out a potential association. Our results support a positive association between endogenous estrone and risk of colorectal cancer.

No potential conflicts of interest were disclosed.

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.

We thank Lynne Quinones for assistance with data collection, Yelena Afanasyeva for database management, Noriko Shimizu for administrative support, and Joseph Katz for performing the SHBG assays.