Ovarian cancer risk and survival according to tumor sex hormone receptor expression: An ovarian Cancer association consortium and ovarian tumor tissue analysis consortium pooled analysis.

OCAC was established in 2005 to research the associations between epidemiologic and genetic factors and EOC risk [ 24 ]. OTTA was formed in 2010 to validate prognostic markers for EOC by histotype [ 25 ]. We included 20,888 controls and 4762 cases from 13 case-control studies with epidemiologic factors in OCAC and hormone receptor expression data by immunohistochemistry ( IHC ) in OTTA to estimate the association between hormonally-linked factors and EOC risk [ 26 – 35 ] ( Supplemental Table 1 ). We included 5737 cases from 14 case-control studies and 5 case-only studies with hormone receptor expression data in OTTA to estimate survival by tumors defined by hormone receptor expression [ 26 – 36 ] ( Supplemental Table 2 ). We estimated the association between hormonally-linked factors and survival of EOC tumors defined by hormone receptor expression by combining the OCAC risk factor and OTTA outcomes datasets. Participating institutions obtained approval from relevant ethics committees and, where applicable, all participants provided informed consent. Data on AR, ER, and PR staining were obtained from OTTA [ 19 , 37 ]. Briefly, OTTA studies procured FFPE tumors from primary cytoreductive surgery and created tissue microarrays. IHC analyses were performed by the Genetic Pathology Evaluation Centre (Vancouver, BC, Canada) for ER and PR and by Ventana Medical Systems Inc. (Tucson, AZ, USA) for AR in Vancouver using the Ventana Discovery Ultra machine, as previously described [ 19 , 37 ]. Two observers scored the staining intensity for each IHC biomarker. In the original dataset, a 5-tiered system (no tumoral tissue, necrosis or hemorrhage, no staining in tumoral cells or just cytoplasmic staining, just stromal cells staining, just tumoral cells staining, and both tumoral and stromal cells staining) was used for AR and a 3-tiered system (50 % of tumor cell nuclei positive) was used for ER and PR. As done previously [ 19 , 37 ], we defined positive receptor expression as stromal or/and tumoral cell staining for AR and nuclear expression in ≥1 % of tumor cells for ER and PR. We identified 11 hormonally-linked harmonized factors available in the OCAC core dataset that have previously been examined as potential EOC risk factors [ 2 – 4 , 38 – 43 ]: physical inactivity (associated with increased estrogen [ 44 ]), recent body mass index ( BMI ) greater than or equal to 25 kg/m 2 (associated with hormonal imbalances, included increased estrogens especially in post-menopausal women [ 45 , 46 ]), smoking status (associated with anti-estrogenic effects [ 47 , 48 ]), history of ever OC use (associated with altered hormonal milieus [ 9 – 11 , 49 , 50 ]), history of pregnancy (associated with excess progesterone [ 12 ]), history of breastfeeding (associated with decreased estrogen [ 13 ]), age at menarche less than 13 years (associated with excess estrogen [ 51 , 52 ]), menopausal status at diagnosis (associated with decreased endogenous estrogen and progesterone [ 53 – 55 ]), history of endometriosis (associated with increased excess estrogen and reduced progesterone [ 14 , 56 ]), history of ovary-sparing hysterectomy (associated with altered hormonal milieus) [ 57 ], and history of HT use (associated with increased estrogen [ 15 , 16 ]). Other variables from OCAC included age at diagnosis (cases) or interview (controls), race, and family history of breast or ovarian cancer. All covariate information was self-reported and represented pre-diagnosis data. We obtained harmonized OTTA data related to survival, including histotype, grade, stage, residual disease, chemotherapy use, vital status, and overall survival time. Each site procured these data from medical record review, patient contact, and/or linkage with centralized cancer registries or death-record databases. To ensure accurate histotype classification, OTTA conducted IHC analyses of samples using histotype-specific markers. Based on marker results and centralized pathology review, histotype classifications were revised and used in this study. The primary outcome was EOC risk defined by the individual and joint expression of AR, ER, and PR. We also evaluated overall survival according to individual and joint receptor expression. Polytomous logistic regression estimated odds ratios ( OR s) and 95 % confidence intervals ( CI s) for hormonally-linked factors and EOC tumor types defined by individual and joint receptor expression, comparing cases to controls and controlling for potential confounders: study site, age at diagnosis (cases) or interview (controls), family history of breast or ovarian cancer, duration of OC use, number of pregnancies, duration of breastfeeding, menopause status, history of ovary-sparing hysterectomy, and HT use. Unknown/missing data were treated as indexes. Inclusion of breastfeeding duration and history of hysterectomy did not alter findings and were omitted from final models. Age at menarche, history of endometriosis, physical inactivity, BMI, and smoking history were excluded from models due to the large number of missing data. Wald tests were used to assess heterogeneity. The impact of missing receptor expression data was assessed by adding an unknown category for cases without receptor status to models. Kaplan-Meier curves were used to visualize survival by individual and joint receptor expression, with log-rank tests assessing differences between survival functions. Cox proportional hazard models using time from diagnosis until death or last follow up as the time-to-event were used to estimate adjusted hazard ratios ( HR s) and 95 % CIs according to receptor expression controlling for clinical characteristics known to be associated with survival: study site, age at diagnosis, histotype, grade, stage, and debulking status. Wald and likelihood ratio tests evaluated interaction effects. For serous-specific analyses, grade 2 or 3 cases were categorized as high-grade serous ( HGSOC ), and grade 1 cases as low-grade serous ( LGSOC ). Statistical analyses were performed in Stata/SE version 16.1 (StataCorp, College Station, TX), and two-sided P values of less than 0.05 were considered significant.

Table 1 summarizes characteristics of controls and cases defined by individual receptor expression included in the risk analyses. There were 1390 cases with AR− tumors, 563 cases with AR+ tumors, and 10,411 controls from 10 case-control studies included in the AR analyses. Compared to women with AR+ tumors, women with AR− tumors were older and less likely to report family history of breast or ovarian cancer and history of hysterectomy. We included 564 cases with ER− tumors, 1282 cases with ER+ tumors, and 16,606 controls from 8 case-control studies in the ER analyses. Women with ER− tumors were more likely to be non-White and physically inactive compared to controls and women with ER+ tumors. Women with ER+ tumors were more likely to report family history of breast or ovarian cancer and history of hysterectomy. There were 1528 cases with PR− tumors, 1125 cases with PR+ tumors, and 19,851 controls from 10 case-control studies in the PR analyses. Women with PR− tumors were older and were more likely to be non-White than either controls or women with PR+ tumors. Women with PR+ tumors were less likely to report HT use compared to either controls or women with PR− tumors. The associations between hormonally-linked factors and EOC risk did not significantly vary by AR expression ( Table 2 ). Overweight/obesity was associated with higher risk of ER− tumors relative to controls (OR 1.53, 95 % CI 1.18–1.98) but not with ER+ tumors (OR 1.13, 95 % CI 0.95–1.35; P for heterogeneity ( P heter ) = 0.05). Relative to controls, hysterectomy prior to diagnosis had a greater association with ER+ cases (OR 4.99, 95 % CI 4.27–5.83) compared to ER− cases (OR 3.67, 95 % CI 2.93–4.60; P heter =0.02). Post-menopausal status was associated with a higher risk of PR− tumors relative to controls (OR 1.52, 95 % CI 1.26–1.83) but not with PR+ tumors (OR 0.98, 95 % CI 0.80–1.20; P heter < 0.001). Increasing gravidity and duration of OC use were associated with dose-response protective effects by individual receptor expression, except for OC duration and AR+ tumors. Increasing breastfeeding duration showed dose-response protective effects based on individual AR and ER expression (all P trend 0.20). Sensitivity analyses comparing cases to controls from all 13 case-control studies with receptor data and using an unknown index for cases with no individual receptor expression outcome data did not alter the observed associations ( Supplemental Table 3 ). Results restricted to HGSOC were similar to overall findings ( Supplemental Table 4 ). The magnitude of the association between hysterectomy and EOC risk differed by the joint expression of AR and ER ( P heter =0.003): hysterectomy was associated with a higher risk of AR−/ER+ tumors relative to controls (OR 9.31, 95 % CI 7.17–12.09), with smaller magnitudes of positive associations also observed for AR−/ER−, AR+/ER+, and AR+/ER− (OR 4.50, 5.97, and 6.71, respectively; Table 3 Panel A ). The magnitude of the association between hysterectomy and EOC risk also differed by the joint expression of AR and PR ( P heter =0.04): hysterectomy was also associated with higher risks of AR+/PR− tumors and AR+/ER+ tumors (ORs 11.90 and 10.62, respectively) compared to similar risks of AR−/PR− tumors and AR+/PR+ tumors (ORs 7.40 and 7.09, respectively; Table 3 Panel B ). Post-menopause was associated with a lower risk of AR−/PR+ tumors (RR 0.65, 95 % CI: 0.45–0.94) and a higher risk of AR−/PR− tumors (OR 1.74, 95 % CI 1.31–2.32), but was not associated with AR+/PR− tumors and AR+/PR + tumors ( P heter < 0.001; Table 3 Panel B ). Post-menopause was also associated with a higher risk of ER+/PR− tumors (OR 2.17, 95 % CI:1.60–2.95) but not with ER−PR− tumors, ER−PR+ tumors, and ER + PR+ tumors ( P heter < 0.001; Table 3 Panel C ). Sensitivity analyses comparing cases to controls from all 13 case-control studies with receptor data did not alter any association (data not shown). Results were similar when restricting cases to HGSOC ( Supplemental Table 5 ). The association of hysterectomy and menopause status with EOC risk by co-expression of two receptors was not modified by stratifying by a third receptor ( Supplemental Table 6 ). The number of women according to AR, ER, and PR expression included in the survival analyses, together with their clinical characteristics and hormonally linked risk factors, are summarized in Table 4 . Most AR+ and ER+ tumors were HGSOC (70 % and 71 %, respectively) compared to AR− and ER− tumors (56 % and 36 % HGSOC, respectively). A greater percentage of PR+ tumors were of endometrioid histotype compared to PR− tumors (29 % vs. 6 %). Clear cell tumors were more likely to be hormone receptor negative. ER+ tumors were more likely to be advanced stage than ER− tumors (60 % vs. 36 %). ER+ tumors were more common among White women than ER− tumors (93 % vs 81 %). Women with ER+ tumors were also more likely to report a family history of breast or ovarian cancer than women with ER− tumors (15 % vs 8 %). More PR− tumors had higher grades than PR+ tumors (91 % vs. 79 %). Women with PR+ tumors were more likely to be diagnosed at a younger age than women with PR− tumors (57 vs 59 years). PR+ tumors were more likely to be diagnosed among White women than PR− tumors (92 % vs. 89 %). Women with ER− tumors had longer survival than women with ER+ tumors (median 7.26 years vs. 4.94 years; Table 4 ). Women with PR+ tumors had longer survival than women with PR− tumors (median 6.93 years vs 4.22 years). Women with AR+ and AR− tumors had similar survival times (median 5.11 years and 4.92 years, respectively). The Kaplan-Meier curves by individual and joint hormone receptor expression are presented in Supplemental Figs. 1 and 2 . Survival did not vary by AR expression but did vary by ER and PR expression. The curves for EOC tumors defined by the joint receptor expression crossed in all analyses, although women with ER+/PR− had worse survival compared to women with all other combinations of ER/PR expression (P from log-rank test comparing ER+/PR− to all other combinations <0.001). There was no clear distinction among groups defined by combinations of all three receptor expressions ( Supplemental Fig. 2 Panel D ). Survival by histotype and by individual receptor expression is presented in Supplemental Fig. 3 . Women with PR+ HGSOC had longer survival than those with PR− HGSOC (P from log-rank test ≤0.003; Panel C1 ). There was no noted survival difference in HGSOC by individual ER or AR expression status ( Panels A1 and B1) , by the joint expression of AR and ER ( Supplemental Fig. 4 Panel A ), or by the joint expression of AR and PR ( Supplemental Fig. 4 Panel B ). Women with ER−PR+ HGSOC had the best survival compared to women with HGSOC defined by the other three ER and PR expression groups ( Supplemental Fig. 4 Panel C ). Women with ER+ or PR+ endometrioid tumors had longer survival compared to women with ER− or PR− tumors (P from log-rank test ≤0.001; Supplemental Fig. 3 Panels B2 and C2 ). These associations were not altered by AR expression ( Supplemental Fig. 5 ). Women with ER+ clear cell tumors had shorter survival than women with ER− clear cell tumors ( Supplemental Fig. 3 Panel B4 ). There were no differences in survival based on AR or PR expression among women with clear cell tumors. There was no difference in survival based on any individual hormone receptor expression in women with mucinous cancers. Because of the limited sample sizes for clear cell and mucinous tumors, we did not perform analyses based on the joint expression of receptors. Similarly, due to low number of cases, we did not perform any receptor-survival analyses for LGSOC. Being obese or overweight was associated with poorer survival for women with AR+ EOC but not for women with AR− EOC ( Table 5 : HR = 1.57 vs 0.95; P for interaction 0.02). This association was not modified by ER or PR expression (data not shown). Physical inactivity was associated with worse survival for women with AR+ tumors but not for women AR− tumors (HR 1.71 vs 1.02; P for interaction 0.005). This relationship was attenuated when controlling for BMI but remained significant (HR 1.48 vs 1.04; P for interaction 0.04). Similarly, physical inactivity was associated with poorer survival in women with ER+ tumors but not for women with ER− tumors (HR 1.47 vs 0.75; P for interaction 0.14). Results were unchanged when adding obesity/overweight to the model (HR 1.45 vs. 0.74; P for interaction 0.14). No other differences in overall EOC survival based on hormonal factors and tumor receptor expression were observed. When restricting cases to HGSOC only, HT use was associated with greater survival regardless of individual receptor expression, except for PR+ cases ( Supplemental Table 7 ). The association of HT use with survival did not vary by tumor type defined by the co-expression of receptors (data not shown). Post-menopausal women with HGSOC appeared to have better survival regardless of individual receptor expression ( Supplemental Table 7 ). We did not evaluate survival associations for non-HGSOC cases due to limited sample sizes.

We pooled data from 19 studies in two international consortia to examine the associations between hormonally-linked factors and EOC risk and survival according to AR/ER/PR tumor expression. We observed significant EOC risk and survival differences based on individual and joint receptor expression. The association between being overweight or obese before diagnosis and EOC risk varied by ER expression but not by AR or PR expression, with overweight/obesity associated with increased risk of ER− tumors. Being postmenopausal at diagnosis and EOC risk varied by PR expression and was not altered by ER expression; post-menopausal women had higher risk of PR− EOC, which varied based on AR expression. We observed dose-response protective effects of OC duration and gravidity with all individual hormone receptor expression groups, except AR+ tumors. In contrast, increasing duration of breastfeeding showed a protective dose-response effect regardless of AR or ER expression; there was no association with PR expression. Finally, we found different risk relationships between pre-diagnosis ovary-sparing hysterectomy and EOC types defined by AR/ER/PR individual and joint expression. However, we are cautious in interpreting these findings because the hysterectomy-EOC relationship is complex and potentially confounded by history of endometriosis and HT use [ 58 ]. Notably, we controlled for HT type in analyses ( Table 2 and Supplemental Tables 3 – 6 ), and inclusion of endometriosis in models did not alter findings. We also confirmed previously reported OCAC data showing significant differences in EOC survival based on tumor ER and PR receptor expression [ 19 ] and expanded those results to show that the association may be modified having history of physical inactivity in general and by HT use and being postmenopausal for HGSOC specifically. We further found that while AR expression was not associated with survival in general, being obese/overweight before diagnosis did differentially impact survival in women with AR+ and AR− tumors. We also confirmed that women with ER−/PR+ HGSOC tumors have longer survival compared to women with HSGOC defined by other ER/PR expression combinations [ 19 ]. Moreover, women with ER−/PR− endometrioid tumors had worse survival compared to women with the other three receptor-defined endometrioid tumor groups. Also, women with ER+ clear cell tumors had worse survival compared to women with ER− clear cell tumors. Collectively, our findings suggest that ER and PR may potentially serve as prognostic biomarkers for HGSOC and endometrioid EOC [ 59 , 60 ] while ER may serve as a prognostic biomarker for clear cell EOC. Our finding that risk varied by PR receptor and postmenopausal status is consistent with results from a pooled study of 197 Nurses’ Health Study (NHS) cases, 42 NHS-II cases and 76 New England Case-Control Study (NECC) cases [ 61 ] and a study of 157 NHS-only cases [ 62 ]. In our study, 58 % of cases were PR− tumors ( N = 1528) and 42 % were PR+ tumors ( N = 1125). The distribution was similar to the NHS/NHS-II/NECC study [ 61 ], but the percentage of PR− tumors was lower than that in the NHS-only study [ 62 ]. Our study and the NHS/NHS-II/NECC study considered PR+ if ≥ 1 % of cells stained positive, while the NHS-only study considered PR+ if >10 % of cells stained positive. Thus, regardless of how PR+ was defined, the association of PR expression with EOC risk appears to be modified by menopausal status at diagnosis. Furthermore, consistent with previous studies [ 20 , 61 – 66 ], we observed that PR− tumors were more likely to be advanced stage, higher grade, suboptimally debulked, and associated with worse overall survival compared to PR+ tumors. Although the exact biological mechanism underlying these findings is unknown, they may reflect that postmenopause is associated with decreased progesterone levels and that progesterone exposure is associated with reduced EOC risk [ 67 ] and improved survival [ 67 , 68 ]. When examining survival by tumors defined by individual receptor expression status, we recapitulated previously reported findings and report novel results. We confirmed and expanded prior findings that women with PR− HGSOC tumors, PR− endometrioid tumors, ER− endometrioid tumors, and ER−/PR− endometrioid, and ER+ clear cell tumors had worse survival and that survival did not vary by receptor expression for women with mucinous tumors [ 19 ]. The heterogeneity of the association between the expression of hormonal receptors and survival across histotypes suggests different AR, ER, PR signaling among EOC histotypes. Our study contrasts with 90 serous cases from the Malmö Diet and Cancer Study (MDCS) and Malmö Preventive Project (MPP) cohorts, which indicated AR expression alone was a favorable prognostic factor for serous histotypes [ 21 ], and the Swedish cohort study, which indicated AR+/PR+ tumors present better survival [ 20 ]. The reasons for conflicting findings might be the definition of AR-positive and the outcomes of interest. In the MDCS and MPP study, serous tumors were not separated into HGSOC and LGSOC, and the cut-off point for AR was 10 % stained tumor cells [ 21 ]. The Swedish cohort study combined serous and endometrioid tumors and used 10 % stained tumor cells as a cutoff for classification [ 20 ]. Our survival analyses did not identify significant interaction for hormonally-linked factors and hormone receptor expression by histotypes. However, these analyses were limited by sample size, especially for LGSOC and mucinous tumors. Moreover, we only had 52 ER−/PR+ HGSOC tumors, and the better survival for women with these tumors compared to women with HGSOC defined by the other ER/PR expression groups needs to be interpreted cautiously. In addition to the reduced sample size for histotype-specific analyses and although OTTA has performed IHC marker analyses and centralized pathology review to validate histotype classification, we cannot exclude potential histotype misclassification, which could mask histotype-specific findings. Also, because studies procured tissue using clinically available samples, we cannot exclude the possibility that included cases are a biased sample. However, when comparing the cases utilized by this study to all cases from the cohorts, we found cases with a tissue specimen similar to all cases ( Supplemental Table 8 ), suggesting that the included cases were representative of all cases. Our study has few non-White women, limiting the generalizability of findings. Similarly, the studies included were exclusively from North America, Europe, and Australia, excluding the generalizability of results to women from Asia, Africa, and South America. Moreover, due to the large number of associations evaluated, we cannot exclude that some results are due to chance. However, the associations we examined involve factors previously shown to be associated with EOC risk and/or survival and that have potential biological relevance to risk or survival. Thus, they are not random. Under this scenario, adjustment for multiple comparisons may be overly conservative and mask a potential association, supporting our approach [ 69 , 70 ]. The consistency of findings under different sensitivity analyses also provides confidence in the robustness of results. In conclusion, we identified hormonally-linked risk factors associated with EOC tumors defined by the expression of AR, ER, and/or PR hormone receptors, individually and jointly. EOCs also presented different outcomes depending on histotypes and hormone receptor expression status, underscoring the need to consider both histotype and hormone receptor expression when evaluating patient prognosis. Our findings further suggest that potential biologic mechanisms underlying the association between hormonally-linked factors and EOC need to be studied by both histotypes and tumor AR, ER, and PR expression to truly illuminate the etiology of and prevention and treatment modalities for this highly fatal group of diseases.

Epithelial ovarian cancer ( EOC ) is the most lethal gynecologic malignancy [ 1 ]. The consistent association of oral contraceptive ( OC ) use [ 2 ], pregnancy [ 3 ], and breastfeeding [ 4 ] with reduced risk strongly suggests a role for sex hormones in EOC etiology. Specifically, the hormone hypothesis postulates that androgen and estrogen stimulation of the ovarian epithelium leads to increased risk, while progesterone stimulation has a protective effect [ 5 ]. Recent in vitro findings suggest that hormones similarly affect the fallopian tube epithelium [ 6 , 7 ], where most EOCs are believed to arise [ 8 ]. A role for hormones in EOC etiology is supported by the protective effects of OCs, associated with reduced androgen and estrogen levels [ 9 – 11 ]; pregnancy, associated with excess progesterone [ 12 ]; and breastfeeding, associated with decreased estrogen [ 13 ]. The increased risk associated with endometriosis, which leads to progesterone resistance [ 14 ], and hormone therapy ( HT ) use, which increases estrogen levels [ 15 , 16 ], lends further support to the hormone-EOC relationship. However, the underlying biological mechanisms whereby hormonally-linked factors influence EOC remain unknown. One possibility is through hormone receptors, which mediate sex hormone effects on EOC cell proliferation, migration, invasion, and apoptosis [ 17 – 19 ]. To date, no studies have evaluated the relationship between hormonally-linked factors and EOC risk defined by the individual and joint expression of the androgen receptor ( AR ), estrogen receptor α ( ER ), and progesterone receptors ( PR ). In addition to a potential link with risk, individual hormone receptor expression has been associated with EOC survival [ 19 – 21 ]. However, to date, no study has investigated EOC survival by the paired expression of AR and ER, nor by the co-expression of all three receptors. Nor has any study evaluated the association of hormonally-linked factors with EOC survival according to tumors defined by hormone receptor expression. Insight into the relationship between hormonally-linked factors and tumor biology could help us understand the exposure-EOC risk and survival relationships. We pooled data from studies participating in the Ovarian Cancer Association Consortium ( OCAC ) and Ovarian Tumor Tissue Analysis ( OTTA ) consortium to evaluate associations between hormonally-linked factors with EOC risk and survival according to tumor types defined by AR, ER, and PR expression. Because EOC is a group of etiologically distinct diseases [ 22 , 23 ], we further investigated the impact of hormone receptor expression on risk and survival according to EOC histotype. We hypothesized that risk and survival would vary by tumors defined by hormone receptor expression, and hormonally-linked factors vary in their association with risk and survival based on tumors defined by hormone receptor expression.