Estrogen Replacement Therapy and Ovarian Cancer Mortality in a Large Prospective Study of US Women FREE

Postmenopausal estrogen use is associated with increased risk of hormone-related cancers. Endometrial cancer incidence increases rapidly with use of unopposed estrogen^1- 2; breast cancer incidence, however, increases only after long duration estrogen use.³ Although there is evidence that pituitary and/or sex hormones play an important etiologic role in ovarian cancer,^4- 6 epidemiologic studies of the association between postmenopausal estrogen use and ovarian cancer have had inconsistent results. Earlier case-control studies report decreased risk,^7- 8 no association,^9- 15 or increased risk.^5,16- 19 More recent and larger case-control studies have suggested increased risk,^20- 23 particularly with long duration of estrogen use.^24- 26 However, even the largest previous studies have had limited statistical power to assess the risk associated with long duration of estrogen use.

We documented an increased risk of ovarian cancer mortality among women exposed to postmenopausal estrogens in an earlier report, including 7 years of follow-up data from the Cancer Prevention Study II.²⁷ We now report on the relation between long-term use of estrogen replacement therapy (ERT) and ovarian cancer mortality, including 14 years of follow-up data and 944 ovarian cancer deaths.

Women in this study were selected from the 676 526 female participants in the Cancer Prevention Study II, a prospective mortality study of about 1.2 million women and men begun by the American Cancer Society in 1982. Participants were identified and enrolled by more than 77 000 American Cancer Society volunteers throughout the United States and Puerto Rico.²⁸ Participants completed a confidential, self-administered questionnaire in 1982 that included personal identifiers; demographic characteristics; personal and family history of cancer and other diseases; reproductive history; and various behavioral, environmental, occupational, and dietary exposures.

The vital status of study participants was determined through December 31, 1996, using 2 approaches. Volunteers made personal inquiries in 1984, 1986, and 1988 to determine whether their enrollees were alive or dead and to record the date of all deaths. Subsequent to 1988, automated linkage using the National Death Index was used to extend follow-up through 1996 and to identify deaths among the few participants lost to follow-up between 1982 and 1988. All aspects of the Cancer Prevention Study II cohort are approved by the Emory University School of Medicine Human Investigations Committee. Informed consent to participate in the study was implied by the return of a completed self-administered questionnaire in 1982.

At completion of mortality follow-up in December 1996, a total of 107 810 women (15.9%) had died, 566 417 (83.8%) were still living, and 2079 (0.3%) had follow-up truncated on September 1, 1988, because of insufficient data for National Death Index linkage. Death certificates were obtained for 98.4% of all women known to have died and were coded according to the International Classification of Diseases, Ninth Revision (ICD-9).²⁹ Deaths from ovarian cancer were defined as those women who died with ovarian cancer (ICD-9 codes 183.0-183.9) as the underlying cause of death through December 31, 1996. We made the following exclusions for our analytic cohort (Table 1): women with a history of cancer (except nonmelanoma skin cancer) at baseline; women who were premenopausal; women who had unknown menopausal status or unknown age at menopause; women with incomplete data on estrogen use; women who reported use of estrogen in cream or injection exclusively; or women who used ERT at age younger than 35 years. Women who reported use of estrogen in cream or injection exclusively were excluded because there were too few of them to analyze separately and because this analysis focuses specifically on oral estrogens. We also excluded women who reported having had a hysterectomy, including those who said their menopause was "artificial" as opposed to "natural." Those who reported previous ovarian surgery were also excluded because we could not distinguish partial from total oophorectomy. After these exclusions, a total of 211 581 women and 944 ovarian cancer deaths remained in the analysis.

The baseline questionnaire for women asked, "Have you ever used female hormones (estrogens) other than oral contraceptives?" Supplemental questions asked the reason for use, age at first use, years of use, and method of use (ie, injection, cream, or pill). Several definitions of postmenopausal estrogen use were investigated: ever use (ever/never), baseline and former use, and total years of use. Women with missing years of use who indicated they were still using hormones at the time of enrollment (1.4% of ever users) were assigned the difference between their age at enrollment and the age at first use as their years of use. Baseline users were defined as those women who either said they were still using estrogen or whose total years of use, added to their age at first use, was within 1 year of their age at enrollment. Former users were defined as those women whose total years of use added to their age at first use was less than their age at enrollment. Women who were unclassifiable as to current or former use or duration of use were excluded from the analyses.

We used Cox proportional hazards modeling to examine the association between ERT and fatal ovarian cancer while adjusting for other potential risk factors. All Cox models stratified on exact year of age at enrollment and race (white, black, or other). Potential confounders included in multivariate models were duration of oral contraceptive use (never, <5 years, 5-9 years, ≥10 years), number of live births (0, 1, 2-3, ≥4), age at menopause (<40, 40-44, 45-49, 50-54, ≥55 years), body mass index (<25.0, 25.0-30.0, ≥30.0 kg/m²), age at menarche (<12, 12, 13, >13 years), and tubal ligation (yes or no). Exercise, education, smoking, daily acetaminophen use, and family history of breast and/or ovarian cancer were also examined as potential confounders. However, we did not adjust for these factors in final models because such adjustment had virtually no effect on our results. Daily acetaminophen use was investigated as a potential confounder because it was associated with both risk of ovarian cancer mortality³⁰ and estrogen use in this cohort.

Twenty-two percent (n = 46 260) of women in the study population had used ERT. Among ever users, 24% (n = 11 024) were users at baseline, and 76% (n = 35 236) were former users. Compared with never users, women who ever took estrogens were more likely to be white, have used oral contraceptives, be former smokers, and have a tubal ligation. They were also more educated, had fewer children, and were thinner. (Table 2a). Baseline estrogen users were also considerably younger than never or former users. Mean duration of estrogen use prior to enrollment was 6.4 years among baseline users and 3.8 years among former users.

A positive association was observed between ever use of postmenopausal estrogens and ovarian cancer mortality (fully adjusted rate ratio [RR] = 1.23; 95% confidence interval [CI], 1.06-1.43) (Table 3). The increased risk was mainly observed among women who were estrogen users at baseline (RR = 1.51; 95% CI, 1.16-1.96). Risk was slightly, but not significantly, increased among former users (RR = 1.16; 95% CI, 0.99-1.37).

We examined duration of use separately among baseline and former estrogen users to see whether the positive association could be due to longer duration of use among baseline users than among former users. Both duration of use and recency of estrogen use were predictors of risk (Table 3). Estrogen use of 10 or more years was associated with increased risk among both baseline (RR = 2.20; 95% CI, 1.53-3.17) and former users (RR = 1.59; 95% CI, 1.13-2.25). Short duration of use (< 10 years) was associated with small and statistically nonsignificant increases in risk.

We also calculated age-adjusted ovarian cancer death rates by duration of estrogen use, standardized to the age distribution of the entire female study population. Annual age-adjusted ovarian cancer death rates per 100 000 women were 64.4 for baseline users with 10 or more years of use, 38.3 for former users with 10 or more years of use, and 26.4 for never users.

Among baseline estrogen users, we further examined duration of use by stratifying women with less than 10 years of use into 2 groups: those with less than 5 years of use and those with 6 to 10 years of use. Risk estimates were similar in both groups and were not statistically significant in either group. We also examined risk of ovarian cancer among baseline users with 15 or more years of estrogen use, but found no evidence of a further increase in risk with increasing duration (RR = 2.04; 95% CI, 1.17-3.53), although this analysis was limited by small numbers (13 ovarian cancer deaths).

Among former estrogen users, we examined years since last use (as reported at study entry) by duration of estrogen use to see if recency influenced risk of ovarian cancer mortality (Table 4). Among former users who took estrogen for 10 or more years, the RR estimate decreased with increasing years since last use. Women who had not used estrogens for at least 15 years were not at clearly increased risk (RR = 1.31; 95% CI, 0.79-2.17), whereas risk was increased among those who had stopped estrogen use within 15 years (RR = 2.05; 95% CI, 1.29-3.25) (Table 4).

Because oral contraceptive use and parity strongly decrease risk of ovarian cancer mortality and both are associated with estrogen use, we further examined the risk of long estrogen use by oral contraceptive use and parity. Risk was significantly increased among women who took postmenopausal estrogens for 10 or more years at all levels of oral contraceptive use and parity. No significant increased risk was observed among women with less than 10 years of use at any level of oral contraceptives use or parity (data not shown).

This large prospective study supports the hypothesis that ERT increases the risk of fatal ovarian cancer. The association was related to both duration and recency of hormone use. The risk of death from ovarian cancer was approximately doubled in women who had used estrogens for 10 or more years within the 15 years prior to enrollment. Several aspects of these results suggest that the association may be causal. These include the biological plausibility, the strength of the association, the consistency of a duration effect in both baseline and former users, the persistence of the increased risk after control for other risk factors, and the similarity between these findings and those of previously published case-control studies.^5,18- 19,26

In an earlier report,²⁷ we documented an increased risk of ovarian cancer mortality among women who ever used postmenopausal estrogen (RR = 1.15; 95% CI, 0.94-1.42) and among ever users of more than 10 years (RR = 1.71; 95% CI, 1.06-2.77). In this study, with 7 additional years of follow-up and 113 additional cases in the exposure group, we were able to estimate risk associated with long duration of use separately among baseline and former estrogen users, and to determine that both duration and recency of estrogen use were significant predictors of risk. To our knowledge, no other prospective study has examined the association between duration of estrogen use and ovarian cancer.

Risk of incident ovarian cancer was elevated in 4^5,18- 19,26 of the 5 case-control studies^{5,8,18- 19,26} that measured the risk associated with estrogen use of 5 or more years. In the largest of the case-control studies, Risch²⁶ reported significant increased risk of serous (odds ratio [OR] = 2.03; 95% CI, 1.04-3.97) and endometrioid (OR = 2.81; 95% CI, 1.15-6.89) ovarian cancer among women who used unopposed estrogen for a total of 5 or more years. No association was found with mucinous cancer (slightly less than half of the cases), suggesting that estrogen use may contribute only to the development of nonmucinous types of ovarian cancer. Hempling et al⁸ found no increased risk (OR = 0.6; 95% CI = 0.3-1.4) with 10 or more years of use. In the remaining 3 studies, the ORs for the longest duration category of estrogen use ranged from 1.6 to 2.8,^5,18- 19 although none of these results were statistically significant due to small numbers.

The mechanisms underlying an association between postmenopausal estrogens and ovarian cancer have not been established. The principal endocrine change of menopause is a decrease in estrogen secretion due to depletion of ovarian follicles, with consequent loss of negative feedback inhibition to the pituitary gland and transient increased levels of gonadotropins. Postmenopausal estrogen therapy raises serum estradiol and estrone levels and decreases the secretion of gonadotropins.³¹ In a prospective study of the association of serum hormone levels with the development of ovarian cancer,³² decreasing levels of gonadotropins were associated with significantly higher ovarian cancer risk.

A second possible mechanism is that postmenopausal estrogen use increases ovarian cancer risk through a direct effect of estrogens on ovarian cells. According to the model of ovarian carcinogenesis proposed by Cramer and Welch,⁶ estrogens from extraglandular sources may promote proliferation and malignant transformation of ovarian cells. A role of estrogen in ovarian carcinogenesis is supported by the identification of estrogen receptors in the cytosol of normal human ovaries,³³ the increased proliferation of human ovarian cancer cell lines when exposed to estrogen,³⁴ and the reported beneficial effect of tamoxifen in some women with ovarian cancer.³⁵

Several limitations of our study should be acknowledged. First, data were not available on type of hormone replacement therapy; the majority of baseline users in 1982 were likely taking unopposed estrogens. Until the late 1970s, most hormonal treatment contained only estrogenic compounds³⁶; since then, the prevalence of combined use of estrogen and progesterone has increased greatly for women with an intact uterus.^37- 38 Therefore, our findings may or may not be relevant to hormone replacement therapy today if the addition of progesterone to hormone therapy influences or protects against development of ovarian cancer.³⁹

Second, assessment of exposure was based on information from a single self-administered questionnaire in 1982; thus, some misclassification of estrogen use is expected with increasing follow-up time, potentially attenuating a true association between current hormone use and ovarian cancer risk. However, in a subgroup of 97 788 women who completed another questionnaire in 1992, we found that 69% of baseline users in 1982 remained current users of hormone replacement therapy. In addition, in a reanalysis of the data comparing the first 7 years of follow-up with years 8 to 14, we found little empirical evidence that increasing misclassification of exposure with time substantially biased our results. Among baseline users, the RRs were 1.65 (95% CI, 1.08-2.52) and 1.44 (95% CI, 1.03-2.01) for the first and second follow-up periods, respectively, and among former users, the RRs were 1.24 (95% CI, 0.96-1.61) and 1.13 (95% CI, 0.91-1.40).

Study participants are on average more educated and affluent than the US population as a whole. While these differences may influence comparisons of absolute rates of disease or exposure between this population and that of the United States, they are unlikely to compromise internal validity. Strengths of our study are its size; prospective design; exclusion of women with cancer at baseline, which eliminates the possibility that disease status might influence or bias the reporting of exposures; and the opportunity to adjust for a number of known and hypothesized ovarian cancer risk factors.

Lifetime risk of ovarian cancer is low (1.7%),⁴⁰ and any increase in risk of ovarian cancer mortality due to long-term estrogen use must be considered in the context of the overall balance of potential risks and benefits.^41- 43 The impact of sequential or combined estrogen and progesterone therapy on ovarian cancer risk is unknown; additional large observational studies are needed to confirm our results and to examine whether effects are similar for unopposed estrogen use and estrogen used in combination with progesterone. If our results are confirmed, clinicians will need to consider ovarian cancer among the health risks associated with 10 or more years of estrogen use.

In summary, women who used postmenopausal estrogens for 10 or more years were at increased risk of fatal ovarian cancer. This increase in ovarian cancer mortality was observed for both baseline users and for women who had used estrogens within the 15 years prior to baseline. Estrogen use for less than 10 years was not associated with increased risk.