Hormonal and Reproductive Risk Factors for Epithelial Ovarian Cancer by Tumor Aggressiveness

Ovarian cancer is the fifth most common cause of cancer death among women (1). In Surveillance Epidemiology and End Results (SEER) data (1988–2007), 47.1% of patients died within 3 years of diagnosis, but 10-year survival was 34.1% (2). The wide variability in ovarian cancer survival may be due, partly, to its heterogeneity, which can be characterized by differences in histology and molecular alterations (3). Recent data suggest that high-grade serous carcinomas originate in the fallopian tube rather than the ovarian surface epithelium (OSE; ref. 3). On the basis of these data, ovarian tumors have been classified into 3 developmental pathways (3). Type 1/2 tumors include low-grade serous, mucinous, endometrioid, and clear cell carcinomas that arise from the OSE or from endometriotic implants, and are characterized by mutations in KRAS, BRAF, PTEN, or PIK3CA. Type 3 tumors are high-grade serous and endometrioid tumors with mutations in TP53; these tumors arise in the fallopian tube and have the worst prognosis (3, 4).

Ovarian cancer risk factors associations may differ by developmental pathway. However, it is not feasible to determine developmental pathway in large-scale epidemiologic studies, primarily because determining cell of origin requires extensive sectioning of the tubes and ovaries, and tumor molecular profiling has become cost-effective only recently. However, the most aggressive tumors (i.e., type 3) are likely to be quickly fatal, while women with less aggressive tumors (i.e., types 1/2) will have longer survival. Identifying differences in risk factor associations between the most rapidly fatal versus less aggressive cancers could improve our understanding of ovarian carcinogenesis and better target prevention. Therefore, we conducted an analysis comparing risk factor associations between women who died within 3 years of diagnosis to women who survived at least 3 years postdiagnosis in 4 studies.

The Nurses' Health Study (NHS) was established in 1976 among 121,700 U.S. female registered nurses, in the age range 30 to 55 years. Women completed an initial questionnaire about lifestyle, health behaviors, and medical history and, thereafter, have completed biennial questionnaires to update exposures and disease diagnoses (5). Identification and follow-up of ovarian cancer cases has been described previously (5). At baseline, we excluded women who reported previous history of cancer (except nonmelanoma skin cancer), previous bilateral oophorectomy, radiation-induced menopause, or who had an unknown birth date, leaving 110,493 women for analysis. Women were censored at ovarian cancer diagnosis, death, report of any other cancer, bilateral oophorectomy, or radiation-induced menopause. This study was approved by the Institutional Review Board (IRB) of the Brigham and Women's Hospital (BWH).

The AARP was established between 1995 and 1996 by inviting 3.5 million AARP members in the age range 50 to 71 years in 6 states and 2 metropolitan areas to complete a baseline questionnaire on diet, demographics, reproductive, and medical history (6). Identification of ovarian cancer cases has been described previously (6). We excluded study participants who used a proxy respondent, were male, reported previous history of cancer, had bilateral oophorectomy/unknown oophorectomy status, developed borderline or nonepithelial ovarian cancer during follow-up, or had no follow-up information, leaving 153,180 women for analysis. The AARP study was approved by the National Cancer Institute's Special Studies IRB.

The Australian Ovarian Cancer Study (AOCS) is an Australia-wide population-based case–control study of 1,861 ovarian cancer cases and 1,509 controls recruited between 2002 and 2006 (7). Controls were matched to cases on age and state of residence. Participants completed a health and lifestyle questionnaire about exposures 1 year before diagnosis (cases) or in the previous 12 months (controls; ref. 7). This analysis was limited to invasive cases (n = 1,247) with known vital status. Ethics approval was received from the Human Research Ethics Committees at the Queensland Institute of Medical Research, Peter MacCallum Cancer Centre, University of Melbourne, and participating hospitals and cancer registries.

Details regarding enrollment of cases and controls in the New England Case-Control (NECC) study of ovarian cancer have been described previously (8). Briefly, 2,203 cases and 2,100 controls, matched to cases on age and state of residence, were interviewed in-person about exposures to known and suspected ovarian cancer risk factors that occurred at least 1 year before diagnosis (cases) or more than 1 year before the interview date (controls). This analysis was limited to 1,642 invasive cases. The study was approved by the BWH and Dartmouth Medical School IRBs.

Invasive ovarian cancer cases were divided into 2 groups on the basis of time between diagnosis and death. Rapidly fatal cases died due to ovarian cancer within 3 years of diagnosis, except in the NECC, where rapidly fatal cases died due to any cause within 3 years, as cause of death was unavailable. In the NHS, less than 3% of cases died due to other causes within 3 years. Less aggressive cases died from other causes within 3 years of diagnosis, died more than 3 years postdiagnosis, or did not die during follow-up. Cases had to have at least 3 years of follow-up postdiagnosis. Borderline ovarian cancers were excluded from this analysis.

Analytic exposures included age, parity, oral contraceptive use, tubal ligation, family history of breast or ovarian cancer, intrauterine device (IUD) use, hysterectomy, age at menarche and natural menopause, height, body mass index (BMI), menopausal status, age at first birth, and smoking. In the NHS and NECC, we also investigated estimated lifetime number of ovulatory cycles, calculated as current age (if premenopausal) or age at natural menopause minus age at menarche, years of oral contraceptive use, and parity (1 year per pregnancy). In the NHS, we analyzed type and duration of postmenopausal hormone (PMH) use. The AARP did not contribute data on timing of oral contraceptive use, tubal ligation, family history of ovarian cancer, or IUD use. For AARP variables that were collected in categories, the median of the category was assigned to create a “continuous” variable. As a sensitivity analysis, we also meta-analyzed study-specific results excluding the AARP when that study only had data in categories.

For the cohort studies, we used the Cox proportional hazards competing risks analysis (Statistical Analysis Software v9.2) to calculate relative risks (RR) and 95% confidence intervals (CI); for the case–control studies, ORs and 95% CI were calculated using polytomous logistic regression (STATA v9). For AARP, baseline exposures were analyzed; these exposures were not updated until 2004, when most of the cases had already occurred. For NHS, exposures were updated whenever new data were obtained. In the NHS, NECC, and AOCS, models were adjusted for age, parity, duration of oral contraceptive use, tubal ligation, duration of breastfeeding, family history of ovarian cancer, and menopausal status. The case–control studies were additionally adjusted for matching factors. AARP analyses were adjusted for age, parity, duration of oral contraceptive use, and menopausal status.

To combine RRs across studies, we used random-effects meta-analysis. P values for heterogeneity among the studies were calculated using the Q statistic (referred to as P_het). To calculate a P value for differences between rapidly fatal and less aggressive ovarian cancer, we calculated a Z-score by taking a weighted average of the differences in the betas for rapidly fatal and less aggressive disease (referred to as P_diff). Weights were the inverse of the variance of the difference in betas.

Because the selection of a 3-year cut off was based on mortality trends in the NHS (data not shown), we conducted analyses using 5 and 10 years postdiagnosis to distinguish the case groups. In addition, to evaluate confounding by tumor histology or stage, we conducted analyses restricted to serous or stage III tumors. We also adjusted the RRs for stage by fitting a competing risks model with 4 case groups (rapidly fatal/low stage, rapidly fatal/high stage, less aggressive/low stage, and less aggressive/high stage) and calculating a weighted average of the associations across stage (9). Moreover, although we did not have grade, treatment, and debulking information in all studies, the NECC had this information on a subset of cases, in which we adjusted for these factors. We also planned to adjust for chemotherapy, but more than 85% of the cases received platinum-based therapy, indicating that potential confounding by chemotherapy was limited. In addition, because case–control studies may not include the most aggressive cases, we compared NHS cases that were fatal within 1.5 years of diagnosis with cases that died within 1.5 to 3 years postdiagnosis (Supplementary Tables S1–S5).

Compared with less aggressive cases, rapidly fatal cases were older and tended to have shorter duration of oral contraceptive use (Table 1). Rapidly fatal cancers were more likely to be high-stage cancers with serous histology. However, the majority of less aggressive cases were also high stage and had serous histology.

We observed marked differences in the association between the 2 case groups for age, parity, and oral contraceptive use (Table 2). Increasing age was more strongly associated with rapidly fatal disease (RR, 5-year increase: 1.39; 95% CI, 1.29–1.49) than less aggressive disease (RR, 1.09; 95% CI, 1.03–1.16; P_diff < 0.0001). Similarly, the inverse association with oral contraceptive use was stronger for rapidly fatal disease (RR, 5-year increase, 0.69; 95% CI, 0.58–0.82) than less aggressive disease (RR, 0.81; 95% CI, 0.74–0.89; P_diff = 0.002). Oral contraceptive use within the last 20 years was associated with a significantly decreased risk of rapidly fatal (RR, 0.48; 95% CI, 0.38–0.61), but not less aggressive disease (RR, 0.86; 95% CI, 0.72–1.03; P_diff < 0.0001). There were no differences for oral contraceptive use more than 20 years ago. When recency and duration of oral contraceptive use were combined, the strongest difference in association was observed for use for more than 5 years within the last 20 years (P_diff < 0.0001).

In contrast, less aggressive disease was more strongly associated with parity. Among parous women, each birth was associated with a 13% decreased risk (95% CI, 0.81–0.93) for less aggressive disease compared with a 2% decreased risk for rapidly fatal disease (95% CI, 0.94–1.03; P_diff < 0.0001). The reduction in risk for the first birth was similar between the 2 groups (RR, rapidly fatal, 0.82; RR, less aggressive, 0.78; P_diff = 0.07). We also observed that being postmenopausal (vs. premenopausal) was associated with a 53% increased risk of rapidly fatal disease (95% CI, 1.24–1.89), but a nonsignificant decreased risk of less aggressive disease (RR, 0.88; 95% CI, 0.74–1.05; P_diff < 0.0001). We observed no differences in association by tumor aggressiveness for duration of breastfeeding, tubal ligation, family history of breast or ovarian cancers, IUD use, hysterectomy, age at first birth, age at menarche, age at menopause, height, BMI, or smoking status (Tables 2 and 3).

In the NHS and NECC, a 5-year increase in ovulatory years was associated with a greater increased risk of rapidly fatal disease (RR, 1.34; 95% CI, 1.16–1.55) than less aggressive disease (RR, 1.23; 95%CI, 1.17–1.29; P_diff = 0.001). In the NHS, ever use of estrogen plus progesterone was not associated with rapidly fatal disease (RR, 0.90; 95% CI, 0.65–1.26), but was associated with an increased risk of less aggressive disease (RR, 1.39; 95% CI, 1.06–1.82; P_diff = 0.04). Ever use of unopposed estrogen was associated with increased risk of both tumor types.

We observed clear differences in the associations of ovarian cancer risk factors by tumor aggressiveness among 4,342 ovarian cancer cases. Specifically, older women and women who never used oral contraceptives were at a greater increased risk of rapidly fatal disease. By contrast, increasing parity was associated with decreased risk of less aggressive disease. Importantly, although the studies varied in design, geographic location, and timing, we generally observed highly consistent results, suggesting that the observed differences are robust.

No previous study has compared risk factors for ovarian cancer by tumor aggressiveness. In general, our findings of increased risk of fatal ovarian cancer with increasing age is consistent with SEER data showing poorer ovarian cancer survival with increasing age (2). In addition, several groups have examined incidence of fatal ovarian cancer. Tubal ligation was associated with decreased risk of fatal ovarian cancer in the 1 cohort study (10), which is consistent with our findings, although 4 studies (11–14) of fatal disease observed no association. In addition, a higher number of ovulatory years was associated with decreased survival in 1 study (13), but not in another (14). Parity and oral contraceptive use generally have not been associated with ovarian cancer survival (14–17); however, among postmenopausal women, having 8 or more children was associated with decreased ovarian cancer mortality (18). We observed significant differences in ovarian cancer risk by tumor aggressiveness for parity and oral contraceptive use, suggesting that these factors may influence a tumor's developmental pathway.

Several studies have examined differences by histologic subtype. In general, no differences have been observed for oral contraceptive use (19–21), although the Oxford Collaborative Group on Epidemiological studies of Ovarian Cancer reported that oral contraceptive use was not associated with risk of mucinous tumors (20–22). However, although a pooled analysis of 10 case–control studies, including the NECC, observed no differences in association with parity (19), increasing parity was more protective for endometrioid cancers in the NHS (20) and for nonserous tumors in the AARP (21). Previous findings for parity and risk of nonserous tumors are consistent with the current study, as less aggressive tumors are more likely to be nonserous. However, that most studies reported no difference in the association with oral contraceptive use by tumor histology suggests that the observed protection against more aggressive tumors is not driven through histology.

The factors most strongly inversely related to developing rapidly fatal cancer in our study, that is, long duration of oral contraceptive use and fewer lifetime ovulatory cycles, suggest that the effects of ovulation, such as wound healing or chronic inflammation, may drive an ovarian tumor toward an aggressive phenotype. However, it is unclear why increasing parity was associated only with decreased risk of less aggressive cancers. Compared with long-term oral contraceptive use, each birth involves a shorter interruption of ovulation; thus, parity may be acting through another mechanism. A Finnish cohort of grand multiparous women (≥5 births) was at decreased risk of ovarian cancer compared with the general Finnish population, but there was no evidence of decreasing risk with increasing parity (23), suggesting that each pregnancy does not have the same effect on ovarian cancer risk. Furthermore, in a pooled analysis of 8 case–control studies, including the NECC, a history of multiple births (among parous women) was associated with decreased risk of nonmucinous ovarian cancers (24). Given that progesterone levels increase during pregnancy, and are higher in multiple pregnancies (25, 26), these findings suggest a potential role for progesterone in preventing less aggressive ovarian cancers.

A limitation of this study was that we could not account for grade or treatment, except in a subset of women. Treatment, especially cytoreductive surgery, is an important determinant of survival (27). It is possible that our categorization of cases is a proxy for treatment or that our results are confounded by treatment. However, in the NECC, we examined chemotherapy and debulking status among patients with this information (n = 607). We observed no difference in chemotherapy regimens between rapidly fatal and less aggressive tumors (data not shown), suggesting that tumor aggressiveness is not a proxy for chemotherapy. Furthermore, we observed that adjusting for debulking status had little impact on the observed differences by case group (Supplementary Table S5). We also observed no differences in analyses limited to stage III or serous tumors, suggesting that tumor characteristics are not important confounders. Similarly, when we adjusted for tumor grade in the NECC study (Supplementary Table S6), our results were largely unchanged, suggesting that our observations were not driven through differences in tumor grade between the 2 case groups.

In addition, case–control studies may not include the most aggressive cases due to delays in recruitment. To address this, we compared NHS women who died within 1.5 with 1.5 to 3 years postdiagnosis (Supplementary Table S4). We observed no differences between the 2 groups, except for PMH use. Moreover, AARP lacked information on several exposures, including PMH type, limiting our sample size for some analyses. However, when the AARP was excluded, we had 1,168 rapidly fatal and 2,519 less aggressive cases, ensuring adequate power for all analyses. In addition, for some variables, such as parity and duration of oral contraceptive use, the AARP study collected data in categories; we applied the median of each category to make “continuous” variables. To evaluate whether this difference in the way data were collected impacted heterogeneity across studies, we also conducted a sensitivity analysis excluding the AARP for these variables (Supplementary Table S7); in general, the associations were similar when we included or excluded AARP, but heterogeneity across studies was reduced.

The ideal classification of tumors in this study would have been to combine histology, grade, cell of origin (fallopian tube vs. ovary), and mutational profile to determine each tumor's carcinogenic pathway. However, this is not feasible in large-scale epidemiologic studies, in which tumor tissue is limited, grade is not commonly reported, and tumor mutational status is often unknown. In a recent study in 3 Danish cohorts, in which the carcinogenic pathway was determined using histology, grade, and mutational status, type 3 tumors had a median survival of approximately 2.5 years compared with 6 years for type 1/2 (4), suggesting that classifying tumors by time to death may be a useful proxy for carcinogenic pathway. However, even if our system is not a good proxy for carcinogenic pathway, we observed clear differences between tumors classified by time to death, an endpoint that is crucial to ovarian cancer patients. Our observation of differences between the 2 groups shows that prediagnostic exposures can have important impacts on ovarian cancer survival, possibly by driving the tumor's carcinogenic pathway. Although increasing parity and oral contraceptive use cannot be recommended on a population level for ovarian cancer prevention, if we can identify modifiable factors that prevent the most aggressive cases, then we will gain insight into ovarian cancer biology and improve prevention. Thus, this study provides proof of principle that there are differences in risk by ovarian tumor aggressiveness. Future research should apply this classification system to other potential risk factors.

The strengths of this study include its large sample size and detailed information on ovarian cancer risk factors. In addition, our sensitivity analyses show the robustness of our results. Our findings that the associations with age and oral contraceptive use are stronger for rapidly fatal tumors suggest biological differences independent of histologic subtype. This has important implications for ovarian cancer prevention because oral contraceptives are not prescribed in postmenopausal women and the protective effect of oral contraceptive use lessens after cessation (22); thus, there is no known effective preventive strategy among older women, who have the highest incidence of aggressive disease.

In conclusion, we observed clear differences in risk for rapidly fatal versus less aggressive disease by age, parity, oral contraceptive use, and PMH use, even when accounting for tumor characteristics. These findings have important implications for preventing the most aggressive forms of ovarian cancer. Future research should validate our classification system against the carcinogenic pathway. Additional studies should evaluate modifiable risk factors, including diet and physical activity, by tumor aggressiveness; identification of population-level modifiable risk factors for rapidly fatal ovarian cancer could have dramatic effects on the incidence of this fatal disease.

No potential conflicts of interest were disclosed.

Conception and design: S.S. Tworoger

Development of methodology: E.M. Poole, S.S. Tworoger

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.E. Hankinson, Y. Park, P.M. Webb, D.W. Cramer, N. Wentzensen, S.S. Tworoger

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.M. Poole, M.A. Merritt, S.J. Jordan, H.P. Yang, S.E. Hankinson, B.A. Rosner, N. Wentzensen, S.S. Tworoger

Writing, review, and/or revision of the manuscript: E.M. Poole, M.A. Merritt, S.J. Jordan, H.P. Yang, S.E. Hankinson, Y. Park, P.M. Webb, D.W. Cramer, N. Wentzensen, K.L. Terry, S.S. Tworoger

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.J. Jordan, Y. Park

The NECC study thanks Allison Vitonis, Cameron Fraer, and Mary De Pari. The NHS investigators thank the following state cancer registries: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, and WY. The AOCS gratefully thanks the AOCS Group.

NHS: The Nurses' Health Study (NHS) was financially supported by the National Cancer Institute grants P01 CA87969 and R03 CA143928; Dr. E.M. Poole was supported by the National Cancer Institute grant T32 CA009001. In the NIH-AARP study,

AARP: H.P. Yang, Y. Park, and N. Wentzensen were supported by the Intramural Research Program of the National Institutes of Health.

AOCS: The Australian Ovarian Cancer Study (AOCS) was supported financially by the US Army Medical Research and Materiel Command DAMD17-01-1-0729, the Cancer Councils of New South Wales, Queensland, South Australia, Tasmania, and Victoria, the Cancer Foundation of Western Australia, and grants No. 199600 and 400413 from the National Health and Medical Research Council of Australia. S.J. Jordan and P.M. Webb were financially supported by fellowships from the National Health and Medical Research Council of Australia.

NECC: The New England Case-Control (NECC) study received funding from the National Cancer Institute via grants R01 CA54419, P50 CA105009, and the Department of Defense grant W81XWH-10-1-0280; M.A. Merritt was supported by National Cancer Institute grant R25 CA098566.

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.