Polycystic Ovary Syndrome, Oligomenorrhea, and Risk of Ovarian Cancer Histotypes: Evidence from the Ovarian Cancer Association Consortium

Polycystic ovary syndrome (PCOS) is a complex endocrine disorder characterized by oligomenorrhea and abnormal hormone levels including hyperandrogenism, hyperinsulinemia, and gonadotropin imbalance. While PCOS has been examined as a risk factor for ovarian cancer in multiple studies (1–9), results have been inconsistent, due in part to the inherent limitations in assessing PCOS in a population-based study (10). More objective measures, like measurement of circulating androgen levels, are not generally available. However, oligomenorrhea, defined as infrequent or irregular periods, is estimated to occur in 75%–85% of women with PCOS and less than 18% of women without PCOS, and can be assessed by questionnaire (10–12). Oligomenorrhea has also been examined as an ovarian cancer risk factor but results from studies of this association are also inconsistent (2, 4, 7, 13–19). Some studies have reported a reduced risk of ovarian cancer among women with oligomenorrhea (4, 19), consistent with the incessant ovulation hypothesis (20), which proposes that repeated damage and repair to the ovarian surface increases ovarian cancer risk. Thus, among women with infrequent or irregular cycles, the resulting reduction in lifetime ovulations would be expected to reduce ovarian cancer risk. However, other studies have observed increased risks of ovarian cancer among women with oligomenorrhea (7, 14). Hormonal alterations in women with oligomenorrhea (21–23) including elevated androgen levels, a common characteristic of women with PCOS, have been hypothesized to explain this increased ovarian cancer risk (24, 25). Small sample sizes in previous individual studies in combination with subtype heterogeneity may in part explain the inconsistent results in studies examining PCOS and oligomenorrhea with ovarian cancer risk. To address these issues, we sought to examine the associations between PCOS, menstrual cycle length and irregularity, and risk of ovarian cancer, overall and by histologic subtype among 14 studies from the Ovarian Cancer Association Consortium (OCAC).

Individual-level data were obtained from 14 case–control studies participating in the OCAC. To be included, studies must have collected data on menstrual cycle irregularity, menstrual cycle length, and/or PCOS. All studies had ethics approval, and all study participants provided informed consent. More details on OCAC are provided elsewhere (26). Characteristics of the 14 studies included in the analysis are presented in Table 1. Nine studies were conducted in the United States: the Connecticut Ovarian Cancer Study (CON), the Diseases of the Ovary and their Evaluation Study (DOV), the Hawaii Ovarian Cancer Case-Control Study (HAW), the Hormones and Ovarian Cancer Prediction Study (HOP), the North Carolina Ovarian Cancer Study (NCO), the New England Case-Control Study of Ovarian Cancer (NEC), the New Jersey Ovarian Cancer Study (NJO), the University of California, Irvine Ovarian Cancer Study (UCI), and the University of Southern California Study of Lifestyle and Women's Health (USC); two studies were conducted in Canada: the Southern Ontario Ovarian Cancer Study (SON) and Familial Ovarian Tumor Study (TOR); two were conducted in Europe: the Danish Malignant Ovarian Tumor Study (MAL) and Polish Ovarian Cancer Case-Control Study (POL); and one was conducted in Australia: the Australian Ovarian Cancer Study (AUS). Participants included 16,594 women with invasive ovarian cancer (n = 13,719) or borderline ovarian disease (n = 2,875) and 17,718 controls.

History of PCOS, menstrual cycle irregularity, and menstrual cycle length were self-reported in all studies. The wording of the questions for each of the three exposure variables varied between studies (Table 1). PCOS, polycystic ovary disease, Stein Levanthal syndrome, and polycystic ovaries were reported by the participants (n = 10 studies). For menstrual cycle irregularity (n = 12 studies), we defined women as having irregular cycles if they reported that their cycles were: “sometimes irregular,” “very irregular,” “fairly irregular,” “never regular,” “variable in length,” or were “unable to predict when next period would start.” Menstrual cycle length in days was reported by the participants as a continuous variable or in categories (n = 13 studies). When menstrual cycle length was reported in categories the midpoint of the category was used to assign numerical menstrual cycle length. Menstrual cycle length was dichotomized into >35 days and ≤35 days (reference). Sensitivity analysis were also conducted excluding women with a cycle length <21 days from the reference group. When available, individual studies provided information on other characteristics associated with PCOS and hyperandrogenism (11) including amenorrhea (absence of a menstrual period for 3 months or longer; n = 6 studies), hirsutism/excess body hair (n = 6 studies), and adult acne (n = 8 studies). Information on known and suspected risk factors for ovarian cancer was collected in each study including age, race/ethnicity, oral contraceptive use, parity, tubal ligation, family history of breast or ovarian cancer, and body mass index (BMI).

For analysis including all cases, we calculated study-specific adjusted ORs and 95% confidence intervals (CIs) for the association between each exposure variable and ovarian cancer risk using logistic regression. Study-specific ORs were combined using random-effects meta-analysis to obtain a summary OR. Between study heterogeneity was assessed with Cochran Q statistic. All analyses were adjusted for age (continuous), oral contraceptive use (never use, <2, 2–<5, 5–<10, or 10+ years), parity (0, 1, 2, 3, 4+), history of tubal ligation (yes, no), family history of ovarian or breast cancer (yes, no), BMI (<20, 20-<25, 25-<30, 30+), and race/ethnicity (white, nonwhite). If an individual study did not have enough subject exposure variation (<5 cases and/or controls reporting a given exposure) to calculate study-specific ORs, the study was excluded from the meta-analysis but included in the pooled analyses (described below).

For histology-specific analyses, the data from individual studies were pooled and polytomous logistic regression were used to simultaneously estimate separate ORs and 95% CIs for each histotype (serous borderline, mucinous borderline, low-grade serous, high-grade serous, mucinous invasive, clear cell, and endometrioid; ref. 27). Likelihood ratio statistics were used to calculate P values for heterogeneity comparing models allowing separate associations for each histotype to models assuming the same association for all histotypes (28). All of the included covariates were allowed to vary across histotypes. In addition to adjustment by the covariates listed above, pooled analyses were additionally adjusted for study site. To evaluate effect modification by BMI (<25, ≥25), oral contraceptive use (never, <2 years, 2–<5 years, ≥5 years), age (<50, ≥50), and parity (nulliparous, parous), data from individual studies were pooled. Likelihood ratio statistics were used to compare models with interaction terms and main effects to a model with main effects only. We also conducted the following sensitivity analyses: (i) excluding studies (n = 3) that only asked about diagnosis of polycystic ovaries and not PCOS; (ii) limiting to studies (n = 7) that asked specifically about menstrual cycle characteristics in women in their 20s and 30s; and (iii) excluding the three OCAC studies that had previously published on menstrual cycle characteristics and/or PCOS by histotype (AUS, HAW, and NEC); (iv) excluding women with a cycle length <21 days from the reference group. We also examined the association between amenorrhea (no period for > 3 months; n = 6 studies), hirsutism/excess body hair (n = 6 studies), and adult acne (n = 8 studies) and ovarian cancer among the studies with these data available. All P values were based on two-sided statistics and were considered statistically significant if P < 0.05. Statistical analyses were performed using SAS Version 9.4 (SAS Institute Inc) and Stata 9 (StataCorp).

Overall, 16,594 women with invasive ovarian cancer (n = 13,719) or borderline ovarian disease (n = 2,875) and 17,718 controls were included in the analysis. Irregular cycles was the most commonly reported of the three exposures followed by typical cycle length >35 days and self-reported PCOS (Table 1). The prevalence of these exposures among controls differed between study sites. For PCOS, the prevalence was highest in the NEC study (2.3%) and lowest among women in the HAW study (0.4%). The percentage of women reporting irregular cycles was highest in the AUS study (24.8%) and lowest in MAL (5.9%). The highest prevalence of reported cycle lengths >35 days was in the HAW study (9.1%) while in three studies (MAL, NEC, and USC) the prevalence was only 1.3%. Women reporting irregular cycles were more likely to report longer cycle lengths (mean cycle length = 32.3) than those reporting regular cycles (mean cycle length = 28.3).

Self-reported PCOS was associated with a nonsignificant decrease in risk of invasive ovarian cancer (summary OR = 0.87; 95% CI = 0.65–1.15; Fig. 1). Menstrual cycle irregularity was associated with a 17% decrease in risk of invasive ovarian cancer compared with regular cycles (95% CI = 0.76–0.89; Fig. 2). Reported typical menstrual cycle length >35 days was associated with a 30% decrease in risk of invasive ovarian cancer compared with women reporting cycle length ≤35 days (95% CI = 0.58–0.84; Fig. 3). There was no evidence of study heterogeneity for PCOS, menstrual cycle length, or menstrual cycle irregularity (P_{study heterogeneity} > 0.46; Figs. 1–3). Among the studies, six studies with information on periods of amenorrhea (no menstrual period for >3 months), we observed a decreased risk of ovarian cancer for women with this exposure (summary OR = 0.88; 95% CI = 0.78–0.99). No association was observed between hirsutism (summary OR = 0.93; 95% CI = 0.80–1.08; n_studies = 6) or adult acne (summary OR = 0.94; 95% CI = 0.80–1.10; n_studies = 8), and ovarian cancer risk. Results were similar in sensitivity analyses that excluded studies that only asked about diagnosis of polycystic ovaries and not PCOS, when restricted to studies that asked specifically about menstrual cycle characteristics in women in their 20s and 30s, when the two OCAC studies that had previously published on menstrual cycle characteristics were excluded, and when women with a cycle length <21 days were excluded from the reference group.

The associations between menstrual cycle characteristics differed by histotype (Table 2). Menstrual cycle irregularity was associated with increased risk of the serous borderline (OR = 1.34; 95% CI = 1.16–1.55) and mucinous borderline (OR = 1.21; 95% CI = 1.02–1.43) histotypes but with decreased risk of high-grade serous, endometrioid, and clear cell histotypes, with corresponding ORs (95% CI) of 0.86 (0.78–0.95), 0.84 (0.72–0.98), and 0.68 (0.55–0.84), respectively (P_{heterogeneity} < 0.001). No associations were observed between cycle irregularity and low-grade serous and invasive mucinous histotypes. Associations with menstrual cycle length also showed heterogeneity by histotype (P_{heterogeneity} = 0.006). Menstrual cycle length >35 days was associated with the lowest risk for low-grade serous (OR = 0.48; 95% CI = 0.25–0.92), invasive mucinous (OR = 0.38; 95% CI = 0.19–0.76), and high-grade serous (OR = 0.62; 95% CI = 0.48–0.80) histotypes, but only a suggestion of decreased risk among the other invasive histotypes (Table 2). No significant associations were found between self-reported PCOS and individual ovarian cancer histotypes (Table 2).

In analyses stratified by BMI, the association between irregular menstrual cycles and risk of the serous borderline histotype was significantly stronger among women who were overweight (BMI ≥25: OR = 1.66; 95% CI = 1.36–2.03) compared with normal weight women (BMI<25: OR = 1.14; 95% CI = 0.88–1.47; P_interaction = 0.009). The inverse associations observed between menstrual cycle irregularity and invasive histotypes were generally stronger among overweight women; however, the P values for interaction did not reach statistical significance (all P_interaction > 0.15; Supplementary Table S1). No differences in effect estimates by BMI were observed for cycle length (Supplementary Table S1) or PCOS. Generally, the PCOS stratified analyses were limited by small numbers in many of the histotype groups.

Significant differences by age were observed for the clear cell histotype, with stronger inverse associations for both irregular cycles and cycle length >35 for women under age 50 years compared with women 50 years and older. Women with a cycle length of >35 days who were <50 years had an 87% (95% CI = 0.02–0.92) decreased risk of clear cell ovarian cancer while women ages 50 years and older had a 7% (0.53–1.66) decreased risk (P_interaction = 0.01; Supplementary Table S2). No differences were observed by age for PCOS with ORs for the association between PCOS and invasive ovarian cancer of 0.88 (95% CI = 0.58–1.35) and 0.93 (95% CI = 0.64–1.35) for <50 and ≥50, respectively. When the associations were examined stratified by oral contraceptive use (never use, < 2 years, 2–<5 years, ≥5 years) no consistent patterns emerged (Supplementary Table S3). No significant differences by parity (nulliparous vs. parous) were observed.

In this large consortium study including over 16,000 women with invasive ovarian cancer or borderline ovarian disease, and 17,000 controls, we observed that both irregular and longer (>35 days) menstrual cycles were associated with decreased risks of invasive ovarian cancer. There was also an inverse association between self-reported PCOS and risk of invasive ovarian cancer, but this result was not statistically significant. Among overweight women who reported irregular cycles, an increased risk of the serous borderline histotype was noted.

Our results for menstrual cycle irregularity are consistent with two prior studies, both of which are included in the current analyses, that examined the associations by histotype. The population-based NEC study of ovarian cancer reported that irregular cycles were associated with a decreased risk of the high-grade serous (OR = 0.68; 95% CI = 0.49–0.95; n_cases = 846) and clear cell histotypes (OR = 0.39; 95% CI = 0.56–1.36) although the small numbers of women with clear cell cancer (n_cases = 116) meant the result did not reach customary statistical significance. An increased risk of serous borderline tumors with menstrual cycle irregularity that was stronger among women who were overweight (OR = 2.29; 95% CI = 1.32–3.98) was also observed (4). In a population-based case–control study conducted in Hawaii and Southern California, that overlapped with the HAW study in the current analyses, Tung and colleagues also reported a significant inverse association between menstrual cycle irregularity and invasive ovarian cancer overall (OR = 0.7; 95% CI = 0.5–0.9; n_cases = 431). When examined by histotype, this association remained significant for only clear cell disease (OR = 0.3; 95% CI = 0.1–0.7; n_cases = 48; ref. 19). In contrast, a cohort analysis [Child Health and Development Studies (CHDS)] found the risk of ovarian cancer was increased for women who reported irregular menstrual cycles, a result that was statistically significant among women age 70 years and older. For high-grade serous disease, the only histotype examined in this study, a nonsignificant increased risk was observed for women reporting irregular cycles (HR = 2.1; 95% CI = 0.9–5.0; n_cases = 30). The CHDS cohort enrolled women that were currently pregnant, including 60% of subjects who were multiparous at study enrollment. CHDS women were relatively young (median enrollment age 26 years), had not used oral contraceptives in the period prior to their pregnancy (96%), and investigators did not have information on established ovarian cancer risk factors that occurred after the study observation period (e.g., tubal ligation, oral contraceptive use; ref. 14). In contrast, the 14 studies included in our analyses included nulliparous and parous women among whom oral contraceptive use was not uncommon. In addition, we were able to adjust for tubal ligation, family history of cancer, and lifetime oral contraceptive use. How such differences in study population characteristics and covariate adjustment could have produced results with opposite directions of association is unclear.

To date, eight studies and one meta-analysis have assessed the association between PCOS and ovarian cancer with primarily null and nonsignificant increased risks reported (29). Only two studies, which are included in the current analysis (AUS and NEC), have examined the association by histotype subtype with both studies reporting an increased risk with the serous borderline histotype which is consistent with our results (4, 5). We also observed an increased risk of the serous borderline hisotype among women with both long and irregular menstrual cycles; however, this association was only statistically significant for women reporting irregular cycles. Androgen excess among women with oligomenorrhea may explain these observations. Women with menstrual cycle irregularity often have higher total and free testosterone levels (30), with potential effects on ovarian tissue. Serous borderline ovarian tumors can express androgen receptors (31) indicating that androgens could exert physiologic effects on these tumors. We observed that the increase in risk associated with irregular cycles was greatest among women with who were overweight or obese (BMI≥25). In addition to elevated androgen levels women who have irregular cycles and are additionally overweight or obese are more likely to suffer other metabolic abnormalities including insulin resistance, and limited cross-sectional evidence suggests that elevated insulin levels may be associated with ovarian cancer risk (32, 33).

Oligomenorrhea was associated with decreased risk of most invasive histologic subtypes. Fewer ovulatory cycles or more anovulatory cycles among women with long and irregular menstrual cycles is a possible explanation for the observed decreased risks. While an association between calculated number of ovulatory cycles and ovarian cancer risk has been established for decades (34–43), few previous studies have examined the association by histologic subtypes. Those studies all found an inverse association that was strongest for the endometrioid histotype (38, 39, 43). While we observed a decreased risk for endometrioid tumors for both menstrual cycle irregularity and longer length (>35 days), the strongest histotype-specific associations were observed with clear cell (for both cycle irregularity and length) and mucinous invasive (for menstrual cycle length). To our knowledge, only the NEC study has examined ovulatory cycles separately for serous borderline and serous invasive cases, reporting a significant increased risk of serous invasive ovarian cancer with increasing number of ovulatory cycles, and a nonsignificant decreased risk of serous borderline disease with increasing cycles (43); this is consistent with the opposite direction of associations that we observed for the serous borderline and serous invasive histotypes for both measures of oligomenorrhea.

In this analyses, PCOS and menstrual cycle characteristics were self-reported. Estimates of PCOS prevalence in reproductive age women range from 4% to 21%, influenced by both population characteristics and the diagnostic criteria used (44–50). Percentage estimates for PCOS in the included studies ranged from 0.4% to 2.3%, well below the expected population levels. While PCOS was initially described in 1935 (51), modern clinical definitions did not emerge until 1990 (NIH criteria), 2003 (Rotterdam criteria), and 2006 (Androgen Excess and PCOS Society criteria; ref. 52). Given the majority of the women participating in our studies were of reproductive age prior to 1990, before PCOS was as commonly clinically recognized, it is likely that PCOS was underdiagnosed in our study population. Among women with clinically apparent PCOS, approximately 80% will have oligomenorrhea (11), thus many women who reported irregular and/or long menstrual cycles may have PCOS.

In case–control studies, recall bias is a common concern with self-reported exposures. Cases could be more likely to recall PCOS or menstrual cycle irregularities than controls. However, given the differing results we observed by histologic subtype, recall bias is an unlikely explanation for our results, particularly for ORs less than unity. More plausible is that underdiagnosis of PCOS in both cases and controls attenuated the results. Another limitation of our study was the differences between the studies in how the question on menstrual cycle irregularity was worded and how answers were categorized. Some studies include a “yes” or “no” response to menstrual cycle irregularity, while others offered more than two answer choices (e.g., very regular, sometimes irregular, very irregular). This likely resulted in some nondifferential misclassification of this exposure.

Strengths of our study include the large number of participants, which provided us with greater statistical power than previous studies to examine the associations by histologic subtype. In addition, previous harmonization of the covariates allowed for consistent adjustment for confounders across the included studies. The associations observed were generally consistent across studies providing further support for our results.

In conclusion, in this large consortium analysis of case–control studies we observed a decreased risk of most invasive epithelial ovarian cancer histologic subtypes among women reporting longer or irregular menstrual cycles. In contrast, such cycles were associated with an apparent increased risk of serous borderline tumors. Further research is needed to understand the mechanisms underlying these bidirectional associations.

No potential conflicts of interest were disclosed.

Conception and design: H.R. Harris, F. Modugno, J.R. McLaughlin, H. Anton-Culver, A. Ziogas, K.L. Terry

Development of methodology: H.R. Harris, M.T. Goodman, K.B. Moysich, J.R. McLaughlin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Babic, P.M. Webb, C.M. Nagle, H.A. Risch, M.A. Rossing, J.A. Doherty, M.T. Goodman, F. Modugno, K.B. Moysich, S.K. Kjaer, E. Høgdall, A. Jensen, J.M. Schildkraut, A. Berchuck, D.W. Cramer, E.V. Bandera, N. Wentzensen, J. Kotsopoulos, S.A. Narod, C.M. Phelan, J.R. McLaughlin, H. Anton-Culver, A. Ziogas, A.H. Wu, K.L. Terry

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H.R. Harris, A. Babic, S.J. Jordan, H.A. Risch, K.B. Moysich, A. Berchuck, N. Wentzensen, A. Ziogas, A.H. Wu, K.L. Terry

Writing, review, and/or revision of the manuscript: H.R. Harris, A. Babic, P.M. Webb, C.M. Nagle, S.J. Jordan, H.A. Risch, M.A. Rossing, J.A. Doherty, M.T. Goodman, F. Modugno, K.B. Moysich, S.K. Kjaer, E. Høgdall, A. Jensen, A. Berchuck, D.W. Cramer, E.V. Bandera, N. Wentzensen, J. Kotsopoulos, S.A. Narod, C.M. Phelan, J.R. McLaughlin, H. Anton-Culver, C.L. Pearce, A.H. Wu, K.L. Terry

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H.R. Harris, S.J. Jordan, H.A. Risch, F. Modugno, E. Høgdall, J. Kotsopoulos, J.R. McLaughlin

Study supervision: F. Modugno, K.B. Moysich, S.K. Kjaer, A. Berchuck, J.R. McLaughlin, K.L. Terry

Other (supplied data, edited, and reviewed): R.B. Ness

This work was supported by grant K22 CA193860 from the National Cancer Institute, NIH (to H.R. Harris) and Department of Defense grant W81XWH-10-1-0280 (to K.L. Terry). Support for the Ovarian Cancer Association Consortium was provided by donations from family and friends of the Kathryn Sladek Smith to the Ovarian Cancer Research Fund. In addition, these studies were supported by the NIH (R01-CA054419, to D.W. Cramer; P50-CA105009, to D.W. Cramer; R01-CA063678, to S.A. Narod; R01-CA074850, to H.A. Risch; R01-CA080742, to H.A. Risch; R01-CA112523, to M.A. Rossing; R01-CA087538, to M.A. Rossing; R01-CA058598, to M.T. Goodman; N01-CN-55424, to M.T. Goodman; N01-PC-67001, to M.T. Goodman; R01-CA095023, to R.B. Ness; P50-CA159981, to K.B. Moysich; K07-CA080668, to F. Modugno; R01-CA061107, to S.K. Kjær; R01-CA076016, to J.M. Schildkraut; K07-CA095666, to E.V. Bandera; R01-CA083918, to E.V. Bandera; K22-CA138563, to E.V. Bandera; P30-CA072720 and R01-CA063678, to S.A. Narod; R01-CA063682, to H.A. Risch; R01-CA058860, to H. Anton-Culver; P01-CA17054, to A.H. Wu; P30-CA14089, to A.H. Wu; R01-CA061132, N01-PC67010, and R03-CA113148, to C.L. Pearce; R03-CA115195, to C.L. Pearce; N01-CN025403 and R01-CA149429, to C.M. Phelan; Intramural Research Program N. Wentzensen), the U.S. Department of Defense (DAMD17-01-1-0729, DAMD17-02-1-0669, to F. Modugno; DAMD17-02-1-0666, to A. Berchuck), National Health & Medical Research Council of Australia (199600 and 400281, to P.M. Webb), Cancer Councils of New South Wales, Victoria, Queensland, South Australia and Tasmania, Cancer Foundation of Western Australia (multi-state applications 191, 211, and 182, to P.M. Webb); Danish Cancer Society (94 222 52, to S.K. Kjær), the Mermaid I project (to S.K. Kjær), the Cancer Institute of New Jersey, National Health Research and Development Program Health Canada (6613-1415-53, to H.A. Risch), Lon V. Smith Foundation (LVS-39420, to H. Anton-Culver), and the California Cancer Research Program (00-01389V-20170; to C.L. Pearce; 2II0200, to A.H. Wu).

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.