Tel.: +404-929-6811, Fax: +404-327-6450
Postmenopausal hormone use and incident ovarian cancer: Associations differ by regimen
Article first published online: 17 JUN 2010
Copyright © 2010 UICC
International Journal of Cancer
Volume 127, Issue 12, pages 2928–2935, 15 December 2010
How to Cite
Hildebrand, J. S., Gapstur, S. M., Feigelson, H. S., Teras, L. R., Thun, M. J. and Patel, A. V. (2010), Postmenopausal hormone use and incident ovarian cancer: Associations differ by regimen. Int. J. Cancer, 127: 2928–2935. doi: 10.1002/ijc.25515
- Issue published online: 17 JUN 2010
- Article first published online: 17 JUN 2010
- Manuscript Accepted: 7 JUN 2010
- Manuscript Received: 20 JAN 2010
- ovarian cancer;
- postmenopausal hormone use;
Ovarian cancer has been associated in epidemiologic studies with postmenopausal hormone use. Whether associations differ by hormone regimen, current status or duration of use is unclear. We examined epithelial ovarian cancer incidence in relation to unopposed estrogen (E-only) and estrogen plus progestin (E + P) among 54,436 postmenopausal women of the Cancer Prevention Study II Nutrition Cohort, a US cohort prospectively followed for cancer incidence since 1992. Demographic, medical, reproductive and lifestyle information was collected at enrollment and updated throughout follow-up via self-administered questionnaire. Extended Cox models were used to estimate age- and multivariate-adjusted relative risk (RR) of ovarian cancer according to hormone regimen, current status and duration of use. During 15 years of follow-up, 297 incident cases were identified. Relative to “never” use of hormones, current E-only use was associated with a twofold higher risk [RR 2.07, 95% confidence interval (CI) 1.50–2.85]; each 5-year increment of use was associated with a 25% higher risk (RR 1.25, 95% CI 1.15–1.36); ≥20 years of use was associated with a near threefold higher risk (RR 2.89; 95% CI 1.71–4.87; trend p = 0.01). Past E-only use was not significantly associated with ovarian cancer, although a modest increase in risk per each 5-year increment of use was suggested (RR 1.14, 95% CI 0.92–1.41). Neither current nor former E + P use was associated with ovarian cancer risk (RR 1.08, 95% CI 0.86–1.35; RR 1.08, 95% CI 0.68–1.71, respectively, per 5-year increment). These findings suggest that progestins may mitigate some of the detrimental effects of estrogen on the ovarian epithelium.
Ovarian cancer is the second most commonly diagnosed and most fatal cancer of the female reproductive system with an estimated 5-year survival rate <50%.1 Although the etiologic origins of ovarian cancer are poorly understood, most tumors are thought to arise in the epithelium, possibly influenced by fluctuations in the hormonal environment related to ovulation and pregnancy, as well as exogenous estrogens and progestins. Use of unopposed postmenopausal estrogen has been associated with a 25–50% increase in risk of ovarian cancer.2–13 Parity and use of oral contraceptives (OCs) are associated consistently with reduced risks,8, 14–18 perhaps due in part to suppression of ovulation, hence a reduction in cyclic trauma and repair to the ovarian surface. However, the strong reductions in risk observed with OC use exceed those which would be expected solely on the basis of fewer ovulatory cycles, suggesting a possible protective effect of progestin, a main component of OC formulations.16 Data from in vitro and animal studies support the hypothesis that progesterone or progestins may act in the ovaries to help protect against carcinogenesis.19, 20
It is unclear whether the addition of progestin to postmenopausal hormonal preparations might mitigate the detrimental effects of estrogen in the ovaries. In contrast to unopposed estrogen (E-only), regimens of estrogen plus progestin (E + P) have not been consistently associated with risk of ovarian cancer. Most studies have found either no association3, 4, 7, 8, 10 or a weaker association than that observed with E-only.5 A recent meta-analysis found a weaker association with use of E + P [relative risk (RR) 1.10, 95% confidence interval (CI) 1.04–1.16, per 5 years of use] than with E-only (RR 1.22, 95% CI 1.18–1.27, per 5 years of use).21 However, not all studies support the idea of a protective or opposing role for progestin. Two prospective cohort studies found no statistical differences between E-only and E + P as related to ovarian cancer.2, 6 Moreover, the Women's Health Initiative (WHI) randomized controlled trials found a 58% increase in risk (RR 1.58, 95% CI 0.77–3.24).22
To clarify the relationships between ovarian cancer and the two most commonly prescribed postmenopausal hormone regimens, we examined whether E-only and E + P were differently associated with epithelial ovarian cancer in the Cancer Prevention Study II (CPS-II) Nutrition Cohort and whether the associations also differed by current status or duration of use.
Material and Methods
Study population and data collection
Study subjects were selected from the 97,786 female participants in the CPS-II Nutrition Cohort, a prospective study of cancer incidence, initiated in 1992.23 The approximately 184,000 men and women in the Nutrition Cohort were selected from 1.2 million participants in CPS-II, a prospective mortality study established by the American Cancer Society in 1982.24 Members of the CPS-II mortality cohort, who were aged 50–74 years in 1992 or 1993, were invited to participate in the Nutrition Cohort, as described elsewhere.23 The Study is approved by the Emory University Institutional Review Board.
Consenting participants completed a self-administered questionnaire that included information on demographic, medical, behavioral, environmental, occupational and dietary factors and were enrolled into the study on the date of completion of the questionnaire. Follow-up questionnaires were sent to cohort members in 1997, 1999, 2001, 2003, 2005 and 2007 to update exposure information and to ascertain newly diagnosed cancers. The response rate for each of the follow-up questionnaires was ≥89%. As described in detail elsewhere,23 incident cancers are first self-reported or identified by death certificate. Reported cancers are subsequently verified by medical records or linkage with state cancer registries covering the entire Nutrition Cohort population. Deaths are identified through biennial linkage of the entire cohort with the National Death Index. Mortality linkage for this analysis is complete through December 31, 2006, and registry linkage is complete through June 30, 2007; therefore, follow-up for this analysis ended on June 30, 2007. End of follow-up for participants who developed ovarian cancer during follow-up was the date of diagnosis; for cases identified by death certificate, death date was used. For all other participants, follow-up ended on June 30, 2007 or death date.
Analyses were restricted to postmenopausal women. Women were excluded at baseline if they were premenopausal, or their menopausal status was unknown as of 1999 (N = 2,254). Women who were pre- or peri-menopausal at baseline, but reported being postmenopausal on the 1997 or 1999 follow-up questionnaire, began contributing person-time to the analysis on the date of the questionnaire reporting the change in menopausal status (N = 2,576). Women were excluded if they were lost to follow-up after 1992 (N = 1,083). They were also excluded if any of the following were reported on the 1992 baseline questionnaire: prevalent cancer, other than nonmelanoma skin cancer (N = 12,218); unknown type or duration of hormone use (N = 6,667); history of both E-only and E + P use (N = 4,075); use of only oral progestin or vaginal cream (N = 1,834); reported current use of E-only with an intact uterus or current use of E + P after hysterectomy (N = 1,815); bilateral oophorectomy (N = 13,389); reported ovarian cancer that could not be verified (N = 8); verified nonepithelial ovarian cancer (N = 7). After exclusions, the final analytic cohort consisted of 54,436 postmenopausal women.
Women of the analytic cohort who developed cancer during the follow-up period, other than ovarian or nonmelanoma skin cancer, were censored and thus exited the study on the date of their questionnaire reporting the cancer diagnosis. Those who reported an unverifiable ovarian cancer during follow-up were censored on the date of their most recent cancer-free questionnaire. Women with missing hormone data in 1997, 1999, 2001, 2003 or 2005 were censored on the date of the questionnaire missing the information; those not returning a questionnaire were censored at the end of the previous interval. Those who reported switching types (E + P to E-only or vice versa), contra-indicated hormone use (i.e., E-only users with intact uterus, E + P users after hysterectomy) or bilateral oophorectomy were censored on the date of the questionnaire reporting such information.
Postmenopausal hormone use
Information on hormone use was collected at enrollment and on all follow-up questionnaires. Subjects were asked about the types of hormones used, whether hormone use was previous or current and the duration of use. For this analysis, women were categorized at baseline into mutually exclusive groups of never hormone users (N = 31,715), E-only users (N = 13,446), or E + P users (N = 9,275). Those who were pre/perimenopausal in 1992 were categorized according to their hormone use after they became menopausal and began contributing person-time to the analysis. Hormone status was updated with each follow-up questionnaire beginning in 1997; current users who reported ceasing hormone use, never users initiating hormone use and former users recommencing E + P or E-only were recategorized at the time of the report.
Incident ovarian cancer
We identified 297 incident ovarian cancer cases diagnosed in the analytic cohort between enrollment and June 30, 2007. The majority of cases (N = 186) were first identified by self-report on a follow-up questionnaire, then verified by medical record abstraction or through linkage with state cancer registries when medical records could not be obtained.23 Our respondents have been found to accurately report a cancer diagnosis with high sensitivity (0.93).25 Case verification of other reported cancers through state cancer registries identified an additional 20 ovarian cancer cases. Ninety-one cases were identified by death certificates where ovarian cancer was listed as the underlying cause of death;26 59 of these were subsequently verified through state cancer registries. Ovarian cancer histology was obtained from medical records or registry linkage for 264 ovarian cancer cases, of which 141 were identified as serous tumors (International Classification of Diseases for Oncology, 2nd edition, morphology codes 8441-2,8460-2); 35 endometrioid (8380-1,8560,8950-1,8980), 15 mucinous (8470-3,8480-1), 13 clear cell (8310), 58 other epithelial (8010,8140,8260,8450) and two unspecified (8000); histology codes were unavailable for 33 ovarian cancer cases identified by death certificate.
Incidence rates of ovarian cancer by hormone use type and duration were standardized to the baseline age distribution of CPS-II Nutrition Cohort women. Extended Cox models, measuring time as person-years from date of enrollment, were used to calculate the age-adjusted and multivariate-adjusted hazard rate ratios for approximation of RRs and 95% CIs.27 All models were stratified on age at enrollment. RRs reported in the text are multivariate adjusted. Values for all covariates in the final models were based upon status at enrollment in 1992.
Postmenopausal hormone use was modeled as a time-dependent variable. At the start of each interval, defined by completion of a new questionnaire, the hormone status of each woman was updated (i.e., never user, current or former E-only user, current or former E + P user), and total lifetime duration of use was calculated. For current E-only users, total duration of use was classified as <10, 10 to <20, or ≥20 years; former E-only use was classified as <5 or ≥5 years; current E + P use was classified as <5 or ≥5 years; duration of former E + P use remained a single category due to small numbers of users with ≥5 years use. p-Values for linear trend were generated from models of years of hormone use as a continuous variable.
Covariates were chosen based upon their potential to confound the association between hormone use and ovarian cancer as determined through univariate and stepwise analyses. Variables included in the final models were education (highschool or less, some college or vocational school, college graduate), race (white, nonwhite), age at menarche (<12, ≥12), parity (live births: none, 1–2, ≥3), OC use (never, <5 years, ≥5 years, user of unknown duration), tubal ligation (yes/no), difficulty becoming pregnant (yes/no), body mass index (BMI: 18.5 to <25, 25 to <30, ≥30), sedentary behavior (hours per day sitting: <3, 3 to <5, ≥6), and family history of ovarian or breast cancer (yes/no). Smoking, alcohol consumption, vegetable, fruit, red/processed meat intake, total calcium and energy intake, use of nonsteroidal antiinflammatory drugs, multi-vitamins and family history of colorectal cancer were found to have no appreciable influence (<2% change) on the associations of interest. Because we excluded subjects with reported contra-indicated hormone use, we did not adjust for hysterectomy, although we did conduct a sensitivity analysis of current E-only restricting both the exposure and referent group to women with hysterectomy.
Age at menopause in the Nutrition Cohort is defined as age of the last menstrual period not induced by exogenous hormones; among women with hysterectomy, age at menopause reflects age at hysterectomy, the average of which is considerably younger than the average age of naturally occurring menopause. To examine any potential confounding by the inclusion of women who had hysterectomy at a young age, but might still have had ovarian function, we conducted a sensitivity analysis excluding those women posthysterectomy who were younger at study entry than the median age at natural menopause.
We examined whether associations between current postmenopausal hormone use and ovarian cancer risk varied by BMI (18.5 to <25, ≥25) and evaluated effect modification by modeling multiplicative interaction terms for current use of each hormone type. p-Values for interaction were generated by the likelihood ratio test that compared a multivariate-adjusted model with main effects to a model that also contained interaction terms.
To evaluate whether or not results were influenced by the inclusion of 35 ovarian cancers with unknown or unspecified histology, we conducted a sensitivity analysis excluding those cases. We also repeated our analysis for serous ovarian cancers; case numbers for endometrioid, mucinous and clear cell cancers were too sparse for separate analyses.
At baseline, 58.3% of women had never used postmenopausal hormones, 10.4% were current E-only users, 14.3% were former E-only users, 13.6% were current E + P users and 3.4% were former E + P users (Table 1). Postmenopausal hormone users of any type were more likely to have used OCs as well. E + P users were more likely to have had tubal ligation, have a later age at menarche, be more educated and leaner, compared to never users or E-only users. Current E + P users were the youngest and former E-only users the oldest at study entry in 1992, which is consistent with temporal trends in hormone use patterns.28
Ovarian cancer incidence rates were highest among current E-only users. Current E-only use was strongly and linearly associated with ovarian cancer (RR 2.07, 95% CI 1.50–2.85; p for trend = 0.01; Table 2). RR increased by 25% (RR 1.25, 95% CI 1.15–1.36) with each 5-year increment of current E-only use; ≥20 years use was associated with near threefold higher risk (RR 2.89, 95% CI 1.71–4.87). RR was also elevated 25% among the subset of women currently using E-only for <5 years (RR 1.25, 95% CI 0.54–2.86), based upon six cases that developed among this group.
Despite the relatively higher incidence rates among former users with ≥5 years use, past E-only use was not statistically significantly associated with RR of ovarian cancer; however, a modest elevation in risk by duration of use was suggested (RR 1.14, 95% CI 0.92–1.41 per 5-year increment; Table 2). Incidence rates among current and former E + P users were not significantly elevated in any duration category, and all RR estimates were consistent with a null finding (per 5-year increment: current E + P, RR 1.08, 95% CI 0.86–1.35; former E + P, RR 1.08 95% CI 0.68–1.71). No evidence of effect modification was found when results were stratified on BMI (Table 3).
In analyses restricted to women posthysterectomy, the incidence of ovarian cancer in never users was significantly lower than in the total population of never users (33 versus 52, per 100,000 person-years). As a result, RRs associated with current E-only use increased substantially (RR 3.00, 95% CI 1.57–5.71). A greater than fourfold higher risk was observed for ≥20 years of use (RR 4.43, 95% CI 2.04–9.61).
Only 331 women, who reported artificially induced menopause, were younger than the median age of 51 at study entry, and none of these developed ovarian cancer. Thus, excluding them had no impact on results.
Sensitivity analyses excluding cases with unknown or unspecified tumor histology (N = 35) generated results almost identical to those of the main analysis, the only difference being a somewhat higher RR for ≥20 years of current E-only use (RR 3.38, 95% CI 2.00–5.73). When cases were further restricted to serous tumors only, results remained unchanged.
In this large prospective cohort study, we found a statistically significant, positive dose-response relationship between E-only use and incident ovarian cancer. The risk increased by 25% with each 5-year increment of use and was nearly three times higher in women with 20 or more years of E-only use relative to never users. A modest elevation in risk with duration of past E-only use was suggested, though not statistically significant. No associations were observed for current or past use of E + P, although statistical power to evaluate E + P was limited by the smaller numbers of E + P users with ≥5 years use among our cohort.
Our finding of a strong positive association between ovarian cancer and E-only, but not E + P, is consistent with most published studies on the topic, including two recent meta-analyses.21, 29 Inconsistent results found in some studies2, 6, 22 may be due to chance or possibly methodological issues. A large Danish study,2 which accrued >3,000 total ovarian cancer cases, found positive associations for both E-only and E + P use, but no difference in magnitude of the associations. That study, however, lacked information on hormone use prior to baseline in 1995. In addition, many subjects switched hormone regimens during follow-up (E-only to E + P and vice versa), although the daily updating of exposure and time-dependent analysis should have minimized the potential for misclassification bias. In the NIH-AARP cohort, a large prospective US study,6 E-only and E + P were also found to be similarly associated with ovarian cancer risk. However, the associations found with E + P appeared to be at least partly driven by stronger associations for sequential, as opposed to continuous combined, E + P, a finding which is consistent with other studies that have examined differences between the two regimens.2, 7, 9 Results from those studies suggest that the lower progestin exposure from sequential regimens may be insufficient to counteract the effects of estrogen in the ovaries. The continuous combined pill came to market in 1995, which was after our baseline questionnaire; because we could not delineate subgroups of sequential versus continuous E + P users, we were unable to address this question directly. Finally, the 58% increase in ovarian cancer risk observed with continuous combined E + P in the WHI randomized trial, which accrued 32 cases of ovarian cancer during an average 5.6 years of follow-up,22 cannot be ascribed to either misclassification or differences in progestin dose. However, the effect estimate was not statistically significant and could be due to chance.
Simple hysterectomy has consistently been associated with a lower RR of ovarian cancer,30–34 and the association appears to be independent of parity. A “screening” effect, whereby abnormalities that may predispose the ovaries to malignancy are found and treated or removed during pelvic surgery, could partly account for the reduced risk. However, it is also plausible that hysterectomy, as with tubal ligation, might confer protection by terminating access of endogenous and exogenous carcinogens to the ovarian environment via an otherwise intact uterus and fallopian tubes.35 Regardless of the process, hysterectomy impacts ovarian cancer risk and has the potential to confound or modify associations with hormone use. When our analyses were restricted to women with hysterectomy, the RR of ovarian cancer associated with current E-only use increased substantially, suggesting that the benefits of hysterectomy are outweighed by an escalation in risks incurred with estrogen use. Stronger associations for E-only use among women with hysterectomy were also found in the Million Women Study5 and in the Breast Cancer Detection Demonstration Project;10 this result was not replicated, however, in the NIH-AARP cohort, where the rate ratios of ovarian cancer among the E-only versus never users did not vary substantially by hysterectomy status.6
Stronger associations between postmenopausal hormone use and breast cancer are consistently observed among lean women. This association is usually attributed to lower levels of endogenous estrogens among lean women compared to women with significant peripheral adipose tissue.36 Whether BMI likewise modifies the relationship between hormone use and ovarian cancer is unclear. Of the few studies to have evaluated this potential interaction,3–5, 10 one3 found the association between E-only and ovarian cancer to be more pronounced in lean (BMI <25) women, but the others found no differences by BMI. Consistent with the latter, our findings suggest that the influence of estrogens on ovarian cancer risk is not dependent upon BMI status.
Our results, consistent with the majority of literature on the topic, suggest that the effects of unopposed estrogens may be more detrimental to the ovaries than when combined with progestins. The hypothesis of a protective or opposing effect of progestin in relation to postmenopausal ovarian cancer was originally proposed as a possible explanation for the strong reductions in risk consistently observed with OCs16, 17 and is supported by several lines of evidence. Data from experimental and in vitro studies suggest that while estrogen acts to stimulate proliferation of ovarian epithelial cells,20, 37, 38 progestins appear to promote apoptosis.19, 20, 39
Our study has several notable strengths. It is one of the largest to date in the US, allowing reasonable statistical power for estimation of RRs of ovarian cancer, and its prospective design eliminates bias due to differential recall of hormone use. Because of the availability of detailed, regularly updated information about hormone use, we were able to define mutually exclusive groups of never-users, E-only users and E + P users, and thus reduce the potential for misclassification. It should be noted, however, that a small possibility of nondifferential misclassification of current hormone status may exist due to the longer interval between enrollment in 1992 and the 1997 questionnaire, which marked the first of five exposure updates during follow-up. Additional strengths of our study are the high response rate achieved with each follow-up, and the ability to control for important confounders. A limitation of our study was the inability to assess the differential effects of sequential versus continuous E + P regimens, nor were our data suitable for assessing variation in routes of administration or time since last use of hormones. We also lacked the statistical power to evaluate the effects of hormone use on histologic subtypes other than serous. Hormone use in our study was self-reported, although self-reported hormone use has been shown to be in good agreement with physician records.40 Our study participants were predominantly white, middle aged or elderly, and well educated; therefore, our results may not be generalizable to populations with different characteristics.
In accordance with post-WHI guidelines, hormone therapy for management of menopausal symptoms and chronic disease prevention is no longer routinely prescribed, as the risks are now known to outweigh the benefits. Postmenopausal women today are more likely to be former users or to initiate hormone use only for short-term relief of vasomotor symptoms. Yet, despite sharp declines in hormone use since 2002, data suggest that substantial numbers of postmenopausal women may continue to use hormones.41 Because of the rarity of ovarian cancer, our findings may not figure prominently in decisions regarding hormone use, which tend to be based upon a woman's underlying risk for more prevalent conditions (e.g., cardiovascular disease, breast cancer) as well as the severity of vasomotor symptoms. For women currently using or considering E-only for any duration, however, the generally poor outcome associated with an ovarian cancer diagnosis may also be worth considering: 5-year survival is 45.5% for all stages combined and 30.6% for distant stage diagnosis.1 As ovarian cancer is a highly fatal disease for which no screening program and few early detection tools presently exist, our findings highlight clinical considerations for women currently using or contemplating use of unopposed estrogens for management of menopausal symptoms. Furthermore, given the previously high prevalence of postmenopausal hormone use, the suggestion that risks may remain elevated among former long duration E-only users warrants further investigation, and underscores the importance of continued follow-up of cohorts with detailed information on lifetime hormone use.
- 1American Cancer Society. Cancer facts & figures 2009. Atlanta: American Cancer Society, 2009.
- 27Survival analysis: a self-learning text, 2nd edn. New York: Springer, 1996. 590 p., .