The Risk of Ovarian Cancer Increases with an Increase in the Lifetime Number of Ovulatory Cycles: An Analysis from the Ovarian Cancer Cohort Consortium (OC3).

The study population included women participating in 20 prospective studies from North America, Europe, Asia, and Australia ( Supplementary Table 1 ). Eligible studies included cohort studies and/or clinical trials with prospective follow-up of women with determination of ovarian cancer endpoints through questionnaire/medical record-based follow-up or confirmation by cancer registries, as well as follow-up for death. Women were excluded from primary analyses if they had a history of cancer (other than non-melanoma skin cancer) at baseline, bilateral oophorectomy prior to study entry, were pre- or peri-menopausal at baseline, or were missing baseline age or age at natural menopause, duration of oral contraceptive use, or number of pregnancies lasting greater than six months (referred to as number of pregnancies in the sections that follow). Our analysis included studies that collected information on age at menarche, age at menopause, number of pregnancies, and oral contraceptive use and the study population was limited to naturally menopausal women at study enrollment (n=558,709), as such women with hysterectomy as their reason for menopause were excluded. All studies obtained ethics approval at their respective institution(s); participants provided either written informed consent or implicit consent through return of the study questionnaire. The OC3 Data Coordinating Center and analytic approaches were approved by the institutional review board of the Brigham and Women’s Hospital. Reproductive factors, including number of pregnancies lasting greater than six months, history and duration of oral contraceptive use, and ages at menarche and menopause, were self-reported at enrollment and previously harmonized as part of a core dataset.( 19 ) We utilized the following formula to calculate number of LOC: L O C = [ ( menstrual span ) ∗ 13 ] − [ ( O C months + preg months ) 12 ] ∗ 13 where ‘menstrual span’ was calculated as the difference between age at natural menopause and age at menarche; ‘OC months’ = duration (in months) of oral contraceptive use; and ‘preg months’=estimated number of months pregnant (calculated from reported number of pregnancies lasting greater than six months*9 months). We included incident epithelial ovarian, peritoneal, or fallopian tube tumors; as described previously.( 19 ) We first evaluated associations of LOC and its component factors with all invasive tumors combined (ovarian, peritoneal, and fallopian tube; n=3,246). Information on histotype was extracted from surgical pathology reports or through cancer registries. The date of or age at death (if applicable) during follow-up was extracted from death registries or reported via family members; all studies reported at least 95% mortality follow-up.( 20 ) We further evaluated associations by the four most common tumor histotype categories: serous (n=2,045, including tumors coded as poorly differentiated), endometrioid (n=319), mucinous (n=184), clear cell (n=121), and a category for missing/unknown histology (n=577). Serous carcinomas are recognized as two distinct diseases ( 21 , 22 ), low- and high-grade serous carcinoma. We used a combination of histology and tumor grade to further define low-grade serous (grade 1 or well-differentiated; n=70) and high-grade serous (≥ grade 2 or moderately differentiated; n=1375). However, nearly a third of the serous carcinomas (600 of 2045 serous carcinomas) were missing tumor grade. Due to the considerable proportion of unknown grade tumors and the likelihood that these tumors are high-grade, we repeated the analyses excluding only the known low-grade tumors (n=70, leaving n=1975 presumed high-grade serous tumors) and additionally excluding unknown grade (n=600, leaving n=1375 confirmed high-grade serous tumors). We also evaluate associations by tumor aggressiveness among presumed high grade serous tumors (n=1975): highly aggressive (lived <1 year post-diagnosis, n=302), very aggressive (lived 1-<3 years, n=625), moderately aggressive (lived 3-<5 years, n=283), less aggressive (lived 5+ years, n=454), and unknown (n=311).( 20 ) We calculated hazard ratios (HR) and 95% confidence intervals (95% CI) using Cox proportional hazards regression to evaluate the association between LOC or its component factors and risk of ovarian cancer. Women entered the analysis at age at study entry and contributed person-time until the age at first diagnosis of ovarian cancer (event), death (censored), or end of follow-up (censored). In primary analyses, we pooled data from all cohorts, stratifying on cohort to account for potential differences in baseline hazards. A priori adjustment factors included baseline age (continuous), body mass index (BMI; 0–5, >5–10, >10 years). We assessed between-study heterogeneity using meta-analysis of cohort-specific estimates. To estimate the impact of LOC independent of its component factors, we mutually adjusted for oral contraceptive use and pregnancies in our final models. We also evaluated the influence of menstrual span as well as the individual factors age at menarche and age at menopause but could not adjust for all factors simultaneously because of collinearity with calculated LOC. In models adjusting for LOC and pairwise combinations of the individual component factors, oral contraceptive use and number of pregnancies explained the most variation in ovarian cancer risk, therefore we present results for LOC with and without adjustment for oral contraceptive use and number of pregnancies. We evaluated possible deviations from linearity of the LOC-ovarian cancer association using a five-knot spline with knots at the 10 th , 25 th , 50 th , 75 th , and 90 th percentiles of LOC. Associations between LOC or its component factors and ovarian cancer case characteristics (e.g., histotypes and tumor aggressiveness) were calculated using fixed effects competing-risks Cox regression.( 23 ) Statistical heterogeneity of associations across case characteristics was assessed via likelihood ratio test comparing a model that assumed different associations for the exposure of interest by case characteristics (full model) to a model with a single estimate for the case characteristics (reduced model).( 24 ) Effect modification by baseline age, oral contraceptives use (never/≤1 year vs. >1-year use), parity (vs. nulliparity)), and commonly measured factors that may influence inflammation (i.e., smoking, BMI, categorized as indicated previously) were evaluated in stratified models, with statistical significance assessed by a likelihood ratio test comparing a model with versus without a multiplicative interaction term. Confounding by other factors that may be associated with LOC and ovarian cancer risk was also assessed, including race, tubal ligation, endometriosis, aspirin/NSAID use, as well as confounding by first degree family history of breast or ovarian cancer. To understand the pattern of LOC across its component factors better, we generated summary tables of LOC based on commonly observed reproductive characteristics and utilized shading to represent low (blue, reference), moderate (white, HR ~1.4 times compared to blue), and high (red, approximately double risk compared to blue) hazard ratios. We then estimated average absolute risk in each category (blue, white, red) based on the 4 components of LOC using a published calculation to put these associations in context.( 25 ) In brief, the absolute risk calculation was previously developed using data from non-Hispanic white women aged 50 or older from two large population-based US cohorts (NIH-AARP Diet and Health Study and the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial).( 25 ) Estimates of relative and absolute risk were combined with age-specific population incidence and competing mortality rates to create the published absolute risk calculator.( 25 ) For ovarian cancer, this model included family history of breast or ovarian cancer, duration of menopausal hormone therapy use, parity, and oral contraceptive use and was validated in a third large population-based US cohort (Nurses’ Health Study cohort).( 25 ) To calculate absolute risk estimates for the current study using the published model, we assumed no history of menopausal hormone therapy use, as that most accurately reflected most of the study population. Survivor function plots for exposures were parallel suggesting no deviation from proportional hazards. All statistical tests were two-sided, and p-values<0.05 were considered statistically significant; analyses were performed using SAS v9.4 (Cary, North Carolina).

The distribution of baseline characteristics overall and by percentile category of LOC are presented in Table 1 . The study population included predominantly white women (>90%). The average baseline age of the study participants increased slightly across LOC category: 58.6, 60.3, 60.8, 61.8, 62.6, 63.5 years for <10 th , 10 th -<25 th , 25-<50 th , 50-<75 th , 75-<90 th , ≥90 th percentile of LOC, respectively. In models evaluating the overall LOC effect (without adjustment for component factors), women in the 90 th percentile of LOC (≥514) were almost twice as likely to be diagnosed with ovarian cancer during follow-up than women in the 10 th percentile (<294 cycles) [HR (95% confidence interval (CI): 1.92 (1.60–2.30), p-trend<0.0001] ( Table 2 ). The association between LOC and ovarian cancer risk was log-linear ( Figure 1 ); associations between LOC and individual histotypes (i.e., serous, endometrioid, mucinous, and clear cell) were also predominantly log-linear. For each five-year increase in LOC (60 cycles), ovarian cancer risk increased by 14% [95% CI: 1.10–1.17]. There was no between study heterogeneity. Adjusting for LOC-components, duration of oral contraceptive use and pregnancy, the LOC-ovarian cancer association remained but was attenuated [per 5 years of LOC: 1.08 (1.04–1.12)] ( Table 2 ). The association was heterogenous by histotype [P heterogeneity=0.01] ( Table 3 ); each five-year increase in LOC was associated with increased risk of serous [1.13 (1.09–1.17)], endometrioid [1.20 (1.10–1.32)], and clear cell [1.37 (1.18–1.58)], but not mucinous [0.99 (0.88–1.10)] tumors. Interestingly, after further adjusting for number of pregnancies lasting greater than six months and oral contraceptive use, hazard ratios were not significantly heterogeneous across histotype [P heterogeneity=0.15] across histotypes [serous 1.08 (1.03–1.12), endometrioid 1.11 (0.99–1.26), clear cell 1.26 (1.08–1.48), mucinous 0.94 (0.81–1.08)], although associations were only statistically significant for serous and clear cell tumors ( Table 3 ). Associations were similar limiting to presumed high-grade serous [1.12 (1.08–1.17); adjusted for pregnancy and oral contraceptive use: 1.07 (1.02–1.11)] and confirmed high-grade serous tumors [1.10 (1.06–1.15); adjusted 1.05 (1.00–1.11), results not tabled]. The risk of presumed high-grade serous ovarian carcinoma per five-year increase in LOC was not statistically heterogeneous across categories of tumor aggressiveness [P heterogeneity=0.60] ( Supplementary Table 2 ). LOC-ovarian cancer associations were not significantly modified by age, prior oral contraceptive use, parity, or by factors potentially related to inflammation [i.e., smoking, BMI] ( Table 4 ). LOC-ovarian cancer associations were unchanged in analyses adjusting for additional potential confounders listed above. The pattern of LOC across commonly observed component factors is graphically presented in Figure 2 . Based on the categorical model in Table 2 , the relative risk varies almost two-fold comparing the 90 th (red shading) to 10 th (blue shading) percentile of number of LOC. The range of LOC exposure values observed in the current study is most clearly reflected when considering menstrual span in the context of no oral contraceptive use and no pregnancy history (top left panel of Figure 2 ). Accounting for increasing number of pregnancies (moving left to right across panels) and increasing duration of oral contraceptive use (moving top to bottom across panels) changes the pattern of relative risk to a scenario where virtually all menstrual span combinations reflect values less than 294 LOC representing the lowest absolute risk (bottom right panel of Figure 2 ). Using these variations in reproductive history and assuming no history of menopausal hormone therapy use, the 10-year absolute risk of ovarian cancer averaged approximately 0.31% (range: 0.21%−0.44%) for 55-year-old women with fewer than 300 LOC (blue); in contrast the 10-year absolute risk averaged approximately 0.59% (0.40%−0.84%) for 55-year-old women with 500 or more LOC (red). The 10-year absolute risk for a 55-year-old woman with the median number of LOC (n~435) averaged approximately 0.42% (range: 0.28%−0.59%) (white).

The present analysis represents the largest prospective study to date of LOC and ovarian cancer risk. The results not only confirm prior studies in showing that higher LOC is associated with an increase in subsequent ovarian cancer risk, they provide data that increases are specific to serous and clear cell and suggestive for endometrioid histotypes. While there was statistical heterogeneity in the risk estimates by histotype when considering LOC overall, after accounting for the impact of oral contraceptive use and pregnancies, the effect sizes were more similar for these three histotypes. The results further show a striking dose-response curve that is largely log-linear. Although the 10-year absolute risk of ovarian cancer is small, it roughly doubles as the number of cycles rises from 300 to 500. We estimated that the 10-year absolute risk of ovarian cancer for a 55-year-old postmenopausal woman varies from approximately 0.31% to 0.59% across the spectrum of LOC values calculated herein. Together with data from other study designs, these results suggest a unifying underlying mechanism of action, such as inflammation that occurs with each ovulation.( 4 ) The association between LOC and ovarian cancer risk has been evaluated in a number of prior studies ( 5 – 10 , 24 , 26 – 33 ), with a comparison of the different algorithms reported in a recent study.( 33 ) Risk estimates varied across these studies; however, most showed an increased risk for greater number of LOCs. In studies that evaluated associations by quantiles, risk estimates were in the range of 1.6 to 1.9 for comparing the top to bottom tertile and 2.1–2.8 comparing the top to bottom quartile of LOC. Most of the published ovarian cancer studies evaluated LOC (or ovulatory years) without further adjustment for the components of LOC, although some studies adjusted for at least one component factor. As a result, it is not well established whether LOC is associated with ovarian cancer risk beyond the contributions made by the factors that define LOC. In the current study, however, we demonstrated that higher LOC is associated with a subsequent increase in the risk of ovarian cancer, independent of the associations with duration of oral contraceptive use and number of pregnancies, or more generally irrespective of the cause of anovulation. Evaluations of the LOC association by histotype ( 24 , 32 ) and tumor aggressiveness are limited; both published studies reported increased risk of serous and endometrioid tumors. Neither study evaluated risks for clear cell tumors given limited numbers. Given that the current paradigm regarding the origin of ovarian cancers suggests that the majority of serous, endometrioid, and clear cell tumors do not originate from the ovarian surface epithelium ( 3 ), it is likely that the increased risk with increasing LOC for these histotypes may be due in part to increased exposure to an acute proinflammatory environment associated with a greater number of LOC.( 34 ) Altered gonadotropin/steroid hormone exposure may also be relevant to clear cell and endometrioid tumors given evidence suggesting etiologic heterogeneity of androgen and estrogen exposure with endometrioid or non-serous histotypes, and null associations for serous tumors.( 35 – 37 ) Alternatively, we cannot rule out other unknown mechanisms that may explain associations between parity and/or oral contraceptive use on ovarian cancer risk reductions. The absence of heterogeneity in the LOC-histotype associations after adjustment for number of pregnancies and duration of oral contraceptive use likely suggests that the heterogeneity was contributed by these factors, however, it could also reflect an inability to detect heterogeneity given the small effect sizes. The prospective design of the pooled studies in this analysis precludes recall bias. Additional strengths of the study include the large sample size, the ability to identify deaths as well as capture losses to follow-up, and the ability to account for many known and suspected risk factors for ovarian cancer. Further, we limited our evaluation to naturally postmenopausal women, which allowed us to compute more reliable estimates of lifetime number of menstrual cycles. Measurement error is inherent to any estimator of LOC and our method is no exception. Multiple algorithms have been used to calculate LOC; the majority include a calculation of menstrual span and then subtract an estimate of cycles when no ovulations are occurring (i.e., on oral contraceptives, pregnant) ( 5 – 10 , 24 , 26 – 32 ) as we did. Some algorithms have also subtracted other times when ovulation is suppressed/ceased including 1) breast feeding ( 5 , 7 – 10 , 24 , 29 , 30 , 32 ), 2) pregnancy loss ( 7 , 27 , 29 , 30 ), 3) preterm birth ( 7 ), and 3) amenorrhea (either postpartum or missed/irregular periods) ( 6 , 8 , 28 , 31 , 32 ), and some have accounted for average cycle length.( 6 , 10 , 30 ) A recent summary of these published algorithms, however, demonstrated that they are all highly correlated (correlations≥0.88, average correlation across algorithms=0.96) ( 33 ), therefore we used a calculation that would maximize the number of potential studies that could be evaluated while minimizing missing data. For example, information on breast-feeding duration was only collected in 8 out of 20 cohorts, representing less than 32% of postmenopausal women included. As such, the calculation we used in the current analysis is likely comparable with other methods on a relative scale but may not reflect absolute values of LOC. Information on irregular menstrual cycles, polycystic ovarian syndrome, early pregnancy losses, etc., were not included in our exposure definition. Additional limitations include potential for residual confounding by age-related factors; however, we did not observe substantial differences in associations across age-strata. We utilized a published absolute risk estimate for illustrative purposes ( 25 ); however, this model was developed using US white women, and as such may have limited generalizability to non-white/non-US study populations. In conclusion, higher numbers of LOC were associated with increased risk of ovarian cancer overall and serous and clear cell tumors, independent of the associations with oral contraceptive use duration and number of pregnancies. Our findings support the hypothesis that ovulation may be a common etiologic factor for most types of ovarian cancer, suggesting this as an important area for identifying intervention strategies. It is plausible, that individual mechanisms for the components used to estimate LOC may also influence other factors beyond ovulation. Future research should examine, in detail, the common biologic mechanisms by which ovulation events influence ovarian cancers.

Ovarian cancer is the most fatal gynecologic cancer. One of the leading hypotheses for epithelial ovarian cancer development is incessant ovulation.( 1 , 2 ) Approximately 80% of high-grade serous ovarian cancers likely originate in the fallopian tube ( 3 ), which are likely also susceptible to the impact of ovulation. Notably, an acute pro-inflammatory environment is created following ovulation; both the surface of the ovary and distal fallopian tube are bathed in follicular fluid containing inflammatory cytokines, reactive oxygen species, and steroids, creating a DNA damage-rich environment.( 4 ) The incessant ovulation hypothesis is supported by the consistent positive association between the lifetime number of ovulatory years or cycles and ovarian cancer risk.( 5 – 10 ) Lending further support, reduced ovarian cancer risk has been observed for reproductive factors that interrupt ovulation (e.g., pregnancy, use of oral contraceptives, and breastfeeding).( 8 , 11 – 15 ) The mechanism as to how the ovulatory process contributes to carcinogenesis is unknown, yet several theories have been proposed, including: an acute proinflammatory environment, altered gonadotropin and/or steroid hormone exposure, or direct damage to the ovarian surface epithelium.( 16 – 18 ) It is difficult to measure number of ovulations directly; however, estimates of cumulative lifetime ovulatory cycles (LOC) can be obtained through algorithms that calculate the time between menarche and menopause (menstrual span) subtracting out presumed anovulatory cycles, due to duration of oral contraceptive use and pregnancy. Prior individual studies, however, have not had sufficient numbers to evaluate the role of ovulation on risk independent of the contributors to ovulatory cycle counts over the life course, particularly by histotype. Thus, we investigated the association of LOC overall and independent of its component factors with subsequent risk of ovarian cancer using prospective individual-level data from the Ovarian Cancer Cohort Consortium (OC3). We further evaluated associations by histotype and tumor aggressiveness among high-grade serous tumors (estimated by time between diagnosis and death), given demonstrated heterogeneity in risk factor associations for these tumor subtypes.( 19 , 20 )