Ovarian Cancer Risk Factor Associations by Primary Anatomic Site: The Ovarian Cancer Cohort Consortium.

This analysis included women from 15 of the prospective cohorts participating in the OC3 ( 13 ) ( Supplemental Table 1 ). OC3 cohorts were required to have: (i) prospective follow-up for incident ovarian cancer diagnoses and deaths and (ii) information on age at recruitment, oral contraceptive (OC) use, and parity. For this study, information on incident peritoneal and fallopian tube cancer cases was also required. All participating studies received institutional approval for cohort data collection and follow-up and the OC3 data coordinating center and analytic approaches were approved by the institutional review board of the Brigham and Women’s Hospital (Boston, MA). Data from baseline questionnaires for 14 full cohorts (n=14) and one case-cohort study with weights were centrally harmonized. Risk factors selected for this study were known and putative ovarian cancer risk factors, with data available and centrally harmonized for the OC3. Exposures in this analysis were: age at menarche (continuous, per 2 years), OC use (never, ever; continuous duration, per 5 years for ever users), parity (nulliparous, one or more pregnancies; continuous for each additional pregnancy), age at first birth (continuous, per 5 years in parous women), age at last birth (continuous, per 5 years in parous women), duration of breastfeeding (continuous, per 6 months in parous women), hysterectomy (never, ever), unilateral oophorectomy (never, ever), tubal ligation (never, ever), menopausal status (premenopausal, postmenopausal), age at menopause (continuous, per 2 years in postmenopausal women), duration of menopausal hormone therapy (MHT) (never, ever; continuous, per 5 years in MHT users), height (continuous, per 5 cm), body mass index (BMI) at ages 18–21 (continuous, per 5 kg/m 2 ), BMI at baseline (continuous, per 5 kg/m 2 ), family history of breast cancer (no, yes), family history of ovarian cancer (no, yes), and smoking status (ever, never). If a study did not collect information on a specific exposure, that study was excluded from the analysis of that factor ( Supplementary Table 2 ). Cases of epithelial ovarian, primary peritoneal, and fallopian tube cancer were confirmed through medical record review or through cancer registries ( 13 ). When using cancer registry data, cases were classified by ICD-O-3 / ICD-10 (ovary: C56.9; fallopian tube: C57.0, C57.4; peritoneum: C48.1, C48.2, C48.8, C57.1) or ICD-O-1 / ICD-9 (ovary: 183.0; fallopian tube: 183.2, 183.8, 183.9; peritoneum: 158.8, 158.9, 183.3) codes. Cases based on medical record review were generally classified based on pathologist expert opinion; during the timeframe of case ascertainment (1980–2015) this was based on the anatomic site of the dominant mass. Women with a personal history of cancer (except non-melanoma skin cancer) at baseline or bilateral oophorectomy were excluded. We used competing-risks Cox proportional hazards regression to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations between the selected exposures and risk of ovarian cancer, peritoneal cancer, and fallopian tube cancers ( 14 ). Participants were censored at the date of ovarian, peritoneal, or fallopian tube cancer diagnosis, death, or end of follow-up, whichever came first. A Kolmogorov-type supremum test indicated no evidence of violation of the proportional hazards assumption, with the exception for the associations between height and all fallopian tube cancers and between BMI and serous ovarian cancer. Data from the 15 cohorts were pooled and all analyses were stratified by cohort and year of birth to allow for baseline hazards to vary by these factors; no statistically significant heterogeneity was observed in random effects meta-analysis. A priori, all models were adjusted for age at study entry, parity, number of pregnancies beyond the first, and duration of OC use; hysterectomy analyses were also adjusted for HT use duration. To assess heterogeneity of associations by tumor site, we used likelihood ratio tests to compare a model that allowed the association for the risk factor of interest to vary by tumor site with one that did not ( 14 ). Missing indicators were included in the model for any missing data in covariates (parity, 2.6% missing; duration OC use, 2.0% missing). The Sister Study was excluded from analyses of family history because all participants had a family history of breast cancer. We examined the associations between the risk factors and invasive ovarian, peritoneal, and fallopian tube cancers overall, and restricted to known serous or high-grade serous tumors. We used SAS 9.4 software (SAS Institute, Cary, NC) to conduct the analyses and p<0.05 was considered statistically significant.

From the 891,731 participants (951,538 with the inclusion of full cohort for the case-cohort study), we identified 3,738 incident invasive cases with the ovary as the primary anatomic site, 337 invasive primary peritoneal cancer cases, and 176 invasive fallopian tube cancer cases ( Table 1 ). For tumors with known histology (n=3,487, 82%) serous was the most common histotype (ovarian=56.9%, fallopian tube=71.8%, primary peritoneal=50.0%). Of the examined reproductive and hormonal risk factors, first pregnancy (p het =0.04) and tubal ligation (p het =0.01) were differentially associated with risk of ovarian and primary peritoneal cancers overall ( Table 2 ; all other p het ≥0.15). First pregnancy and tubal ligation were inversely associated with risk of tumors with the primary anatomic site at the ovary (first pregnancy compared with nulliparous, HR=0.81 [0.73–0.90]; tubal ligation, 0.82 [0.71–0.93]), while no evidence of an inverse association was observed for primary peritoneal cancers (first pregnancy, 1.15 [0.78–1.70]; tubal ligation, 1.31 [0.93–1.84]). Results were similar for serous cases. We observed no statistically significant heterogeneity for any reproductive risk factor comparing ovarian with fallopian cancers, but in analyses restricted to serous cancers, heterogeneity of associations was observed for first pregnancy and hysterectomy (both p=0.02). First pregnancy was associated with higher risk of serous fallopian tube cancer (3.52 [1.02–12.1]), but not serous ovarian cancer (0.90 [0.78–1.04]); in further analyses parity of 1 vs. nulliparity was suggestively associated with serous fallopian tube cancer (1.64 [0.88–3.09]), and significantly inversely associated with ovarian cancer (0.81 [0.68–0.97]). Hysterectomy was inversely associated with serous fallopian tube (0.46 [0.23–0.90], but not ovarian (1.00 [0.89–1.13]), cancers. We observed no heterogeneity in associations comparing fallopian tube with primary peritoneal cancers for all cases (p het ≥0.08) or in analyses restricted to serous cases; though the sample size in these subgroups was limited. Other risk factors were similarly associated with risk of the three endpoints. When restricting to cases with the high-grade serous histotype, the patterns of associations were generally similar to those observed for all histotypes or the serous histotype ( Supplemental Table 3 ). We further observed significant heterogeneity for OC use with risk of ovarian and primary peritoneal cancers (p=0.02) and age at first and last birth (both p=0.02) with risk of high-grade serous fallopian tube and primary peritoneal cancers. Specifically, ever OC use (relative to never) was significantly inversely associated with ovarian (0.82 [0.73–0.93]), but not associated with primary peritoneal cancers (1.48 [0.92–2.38]). Older age at first birth was positively associated with high-grade serous primary peritoneal cancer (per 5 years, HR=1.30 [1.04–1.62]), but not high-grade serous fallopian tube cancer (0.72 [0.49–1.05]); a similar pattern was observed for age at last birth. The association between early adult (ages 18–21) BMI and, for serous tumors, baseline BMI differed by anatomic site ( Table 3 ). Higher early adult BMI was positively associated with risk of peritoneal cancer (per 5 kg/m 2 , HR=1.29 [1.07–1.57]), but not ovarian (p het =0.02; 0.99 [0.92–1.06]) or fallopian tube (p het =0.03; 0.85 [0.64–1.13]) cancers; results were similar for (high-grade) serous tumors. The associations between baseline BMI and serous ovarian and peritoneal cancer were significantly different (p het =0.01), with a suggestively lower risk of serous ovarian and suggestively higher risk of primary peritoneal cancer (per 5 kg/m 2 , ovarian, 0.96 [0.92–1.01]; primary peritoneal, 1.14 [0.99–1.30]). When cases were restricted to high-grade serous tumors, additional heterogeneity in associations between height (p het <0.01) and family history of breast cancer (p het =0.03) and ovarian and primary peritoneal cancers was observed ( Supplemental Table 4 ). Taller height was more strongly associated with high-grade serous primary peritoneal disease (per 5 cm, HR=1.30 [1.15–1.47]) than ovarian (HR=1.04 [1.00–1.09]). Family history of breast cancer was associated with high-grade serous ovarian cancer (HR=1.25 [1.06–1.48]), but not primary peritoneal cancer (HR=0.58 [0.27–1.26]).

In our prospective analysis of >890,000 women, selected reproductive factors, body size, and family history displayed variability in associations for ovarian, peritoneal, and fallopian tube cancers; however, the majority of exposures were similarly associated with risk regardless of anatomic site within the peritoneal cavity. Understanding of the tissue(s) of origin of primary ovarian, fallopian, and peritoneal cancers have evolved rapidly, with hypothesized shared tissue of origin for cancers at these sites, particularly for high-grade serous disease arising from the fallopian tube. Overall, our results in which most risk factors were associated similarly across anatomic sites supports this hypothesis, although select risk factors may have differential influence on the primary tissues on which tumors present. Tubal ligation was inversely associated with ovarian cancer (18% lower risk) but not with fallopian tube and primary peritoneal cancer. Previous studies have shown consistent associations to our study ( 5 – 7 , 13 , 15 ). Studies have evaluated risk factors by tumor dominance as a proxy for ovarian or tubal origin ( 16 , 17 ), including one in the OC3 ( 17 ) (i.e., dominant mass corresponding to tumor of ovarian origin, and non-dominant mass corresponding to tumor of tubal origin ( 18 )), and observed significant inverse associations between tubal ligation and risk only for dominant tumors, in line with our findings. Mechanistically, tubal ligation is thought to prevent primarily endometrioid, clear cell and low-grade serous tumors by blocking the transit of precursor cells from the endometrium. Consistent with this, we only identified a differential association when examining all histotypes or serous only, but not when restricting to high-grade serous disease. Hysterectomy was inversely associated with serous fallopian tube cancer (all cases: 35% lower risk; serous cases: 54% lower risk), but not serous ovarian or peritoneal cancer. Case-control studies have reported positive associations with risk of ovarian cancer ( 5 , 6 ), but not primary peritoneal ( 5 ) or fallopian tube ( 6 , 7 ) cancers for women reporting hysterectomy, while a case-only study reported no differential associations by site ( 3 ). Fallopian tube cancers are thought to arise predominantly from serous tubal intraepithelial carcinomas (STICs), and a proportion of women undergoing hysterectomy would have had concurrent salpingectomy. Our findings may, at least in part, be an artifact of concurrent unilateral or bilateral salpingectomy. In general, the first pregnancy was more inversely associated with risk of ovarian cancer than the other types. In fact, we observed a positive association for serous fallopian tube cancer, which is inconsistent with the one other study that reported inverse associations with first pregnancy for both dominant (ovarian) and non-dominant (tubal) tumors ( 16 ). This finding with serous fallopian tube cancers is likely due to chance. However, the generally stronger inverse relationship of being parous being inversely associated with risk of ovarian cancer is in line with most previous studies of ovarian ( 13 , 19 ) and fallopian tube ( 4 , 6 , 7 ) cancers; a positive association with risk of primary peritoneal, relative to ovarian, cancer was reported in a case-only analysis ( 3 ). Given that the association for subsequent pregnancies (after the first) were very similar across anatomic sites, it seems unlikely that there is a strong differential relationship by parity. We observed heterogeneity in associations for early adult and baseline BMI (for serous only), and height (for high grade serous only). While taller height was positively associated with risk for all cancer subtypes (4–30% increased risk per 5 cm), associations were stronger for peritoneal and fallopian tube cancers, suggesting a role for growth factors and earlier life exposures (e.g., energy restriction) for these tumor types. BMI, either in early adulthood or in adulthood, was positively associated solely with peritoneal cancer risk. Few studies have examined early adult BMI and risk by ovarian tumor subsite, although a recent study observed that BMI increase from age 10 to 18 years was positively associated with risk of ovarian/primary peritoneal cancer combined, although the association was suggestively stronger among cases of non-serous disease ( 20 ). Adult BMI is not a strong risk factor for ovarian cancer, with modest associations observed for the mucinous and endometrioid subtypes ( 13 , 21 ). Previous studies have suggested no association of adult BMI with fallopian tube cancers ( 4 , 16 ) and positive associations for primary peritoneal cancers ( 6 , 9 ). Obesity is a state of chronic low-grade inflammation ( 22 ), both systemically and at the local tissue level (i.e., in the visceral (omental) adipose tissue), and inflammation is associated with ovarian cancer risk ( 23 – 25 ). Further, adipose tissue is the predominant source of estrogens in obese postmenopausal women ( 26 ), and sex steroid hormones are associated with higher risk of non-serous ovarian cancer ( 27 – 29 ). Finally, ovarian cancer frequently metastasizes to the adipose-tissue rich omentum with immune cell aggregates (so-called “milky clusters”) within this adipose tissue acting in immune modulation and identified as sites of metastatic colonization ( 30 ). Given that primary peritoneal cancers develop more proximal to the omentum, relative to ovarian and fallopian tube cancers, an inflammatory and sex steroid hormone-rich tumor microenvironment, together with omental adipose tissue-related promotion of metastases, represent plausible mechanisms linking BMI more strongly with primary peritoneal cancers. A limitation of this study is potential misclassification of primary peritoneal and fallopian tube cancers as primary ovarian cancer because the classification of these is largely based on a subjective determination of disease spread at the time of surgery and registry-based cases were categorized based on the coding rules used by cancer registries. Nonetheless, despite potential misclassification, differences in risk associations by anatomic site for selected risk factors have generally been consistent across studies. Pathology protocols including detailed pathological evaluation of the fallopian tubes (i.e., the SEE-FIM protocol) and greater awareness of and surveillance for STICs contribute to improved classification of anatomical site at diagnosis, and likely account for the recent increase in diagnosed fallopian tube cancers (e.g., 16.2% annual percentage change from 2002–2012 in the US) ( 12 ). These relatively recent advances will provide improved classification with respect to site of origin (rather than progression) for future prospective studies. Future studies may also consider risk factors beyond those included in the present analysis (e.g., anti-inflammatory analgesic use). Overall, our findings suggest that tumors identified as ovarian, primary peritoneal, and fallopian tube cancers have both shared and distinct etiologic pathways, although most risk factors appear to have similar associations by anatomic site. This is in contrast to previous investigations by histotype ( 13 ), which is likely to be more reflective of the cell-of-origin than the anatomic location of tumor presentation. The risk factors that did differ, particularly adiposity measures, suggests that these exposures may be important for determining the vulnerability of the anatomic site to tumor growth. Patterns were generally similar in analyses restricted to (high grade) serous cancers, suggesting that the heterogeneity in associations was not explained by histotype. Enhanced understanding of these differential risk patterns, and mechanistic studies toward a more refined understanding of the underlying physiologic processes leading to this heterogeneity, may inform prevention efforts for these cancers.

Epithelial ovarian cancer is often investigated as a composite outcome including ovarian, primary peritoneal, and fallopian tube cases, given commonalities (e.g., histologic subtypes, pathological staging) and potentially shared tissues of origin (e.g., serous tumors predominantly from the fallopian tube, endometrioid and clear cell from endometriosis and endometrioid adenofibromas) ( 1 ). Relatively few studies have investigated risk factors by primary anatomic site ( 2 – 10 ). To date, these studies have suggested potential heterogeneity in associations by primary site for pregnancy-related and anthropometric characteristics, hysterectomy, and family history of cancer. Prospective studies ( 3 , 7 ) are sparse given the relative rarity of cancers diagnosed as fallopian tube or primary peritoneal cancer (e.g., incidence of ovarian, fallopian tube, and primary peritoneal cancers are estimated at 6.6 per 100,000, 0.62 per million, and 6.78 per million women per year, respectively ( 5 , 11 , 12 )). Given the limited evidence to date, the aim of the current study was to evaluate whether the association between risk factors for invasive epithelial cancers arising in the peritoneal cavity differ by anatomic site at diagnosis (i.e., ovarian, primary peritoneal, fallopian tube), and, if differences were observed, to investigate whether these differences persist after restriction to (high-grade) serous tumors, given recognized heterogeneity in risk factors by histologic subtype ( 13 ). This study was conducted using the Ovarian Cancer Cohort Consortium (OC3), including 3,738 ovarian, 337 primary peritoneal, and 176 fallopian tube incident cancer cases accrued from ~892,000 women in 15 prospective cohorts.