Increased risk for ovarian cancer and borderline ovarian tumours in subfertile women with endometriosis

Human Reproduction, Vol.0, No.0 pp. 1 12, 2013 doi:10.1093/humrep/det340 Hum. Reprod. Advance Access published September 5, 2013 ORIGINAL ARTICLE Reproductive epidemiology Increased risk for ovarian cancer and borderline ovarian tumours in subfertile women with endometriosis C.C.M. Buis 1, F.E. van Leeuwen 2, T.M. Mooij 2, and C.W. Burger 1, * on Behalf of the OMEGA Project Group 1 Department of Obstetrics and Gynaecology, Division of Gynaecologic Oncology, Erasmus Medical Center Rotterdam, PO Box 2040, Rotterdam 3000 CA, The Netherlands 2 Department of Epidemiology, Netherlands Cancer Institute, Plesmanlaan 121, Amsterdam 1066 CX, The Netherlands *Correspondence address. Tel: +31 10 4633617; E-mail: c.w.burger@erasmusmc.nl Submitted on March 1, 2012; resubmitted on July 20, 2013; accepted on August 1, 2013 study question: Is ovarian or extra-ovarian endometriosis associated with an increased risk of ovarian cancer and borderline ovarian tumours (BOT)? summary answer: We found a 3- to 8-fold increased risk of ovarian tumours associated with endometriosis: the magnitude of the risk increase depended on the definition of endometriosis. what is known already: There is increasing evidence of an association between endometriosis and increased risk of ovarian cancer. However, most reports were based on self-reported diagnosis of endometriosis. study design, size, duration: We conducted a nationwide historic cohort study among women with subfertility problems between 1980 and 1995. For this analysis we selected all cohort members with endometriosis, and a comparison group of subfertile women (male factor or idiopathic) without endometriosis (total cohort of 8904 women). Median follow-up time was 15.2 for the entire study population. participants/materials, setting, methods: For this analysis we selected all cohort members with (n ¼ 3657) and without (n ¼ 5247) evidence of endometriosis. Seventy-eight per cent of diagnoses of endometriosis were confirmed by pathology report, and 22% was self-reported endometriosis (positive predictive value of 73%). We linked the cohort with the Dutch Pathology Database and the Netherlands Cancer Registry to assess the occurrence of ovarian cancer and BOT between January 1989 and June 2007. main results and the role of chance: We observed a substantially increased risk of all ovarian malignancies combined in women with endometriosis when we based the definition of endometriosis on self-report, medical records information at subfertility treatment and/or the nationwide pathology database (hazard ratio (HR) 8.2; 95% confidence interval (CI) 3.1 21.6). The HR associated with endometriosis was 12.4 (95% CI 2.8 54.2) for ovarian cancer and 5.5 (95% CI 1.5 20.2) for BOT. When we excluded information from the pathology database, HRs were 3.0 (95% CI 1.5 6.1) for all ovarian tumours, 4.3 (95% CI 1.6 11.2) for ovarian cancer and 1.9 (95% CI 0.6 5.8) for BOT. Both ovarian and extra-ovarian endometriosis carried a significantly increased risk for ovarian cancer and BOT. limitations, reasons for caution: We did not have information on oral contraceptive use and parity for 23.4 and 3.4%, of women in the analytic cohort, respectively. Furthermore, a limitation of our study, and also of other studies, is that the date of diagnosis of endometriosis is usually made long after the onset of the disease. Also, the number of cases in the cohort is small (n ¼ 34), resulting in wide CIs. wider implications of the findings: The fact that endometriosis was assessed before diagnosis of ovarian malignancy and the high degree of medical confirmation in our study likely contribute to the validity of our estimate of a 3- to 8-fold increased risk of ovarian tumours associated with endometriosis. The risk of ovarian malignancies associated with endometriosis was much higher in analyses including information on endometriosis from the nationwide pathology database, implying that risk estimates from studies using self-reported information on endometriosis may be too low due to non-differential misclassification bias. study funding/competing interest(s): None. trial registration number: None. Key words: ovary / endometriosis / infertility Description of the Omega Project is given in the acknowledgement section. & The Author 2013. Published by Oxford University Press on behalf of the European Society of Human Reproduction and Embryology. All rights reserved. For Permissions, please email: journals.permissions@oup.com

2 Buis et al. Introduction Only few risk factors for ovarian carcinoma and borderline ovarian tumours (BOT) have been recognized. The most important risk factor for ovarian carcinoma is nulliparity. Childless women have 2- to 3-fold increased risk of ovarian carcinoma compared with parous women (Risch et al., 1994; Hankinson et al., 1995). A second important factor is oral contraceptive (OC) use (Beral et al., 2008) with a significant reduction in ovariancancer incidence from 1.2 to 0.8 per 100 women who have used OCs for 10 years. Some epidemiological studies observe an association with subfertility (Jensen et al., 2008). Ovarian stimulation for IVF may increase the risk of ovarian malignancies, especially BOT (van Leeuwen et al., 2011). Other studies showed no significant increase in risk of ovarian malignancies after IVF treatment (Jensen et al., 2009; Yli-Kuha et al., 2012). It might be that specific causes of subfertility predispose for ovarian cancer and/or BOT. Whittemore et al. (1992) found a non-significantly increased ovarian cancer risk among women with subfertility due to ovulatory abnormalities [odds ratio (OR) ¼ 2.1] and Fallopian tubes dysfunction (OR ¼ 1.3) compared with women without a history of subfertility. Other studies showed no significant increase or decrease in women with a specific type of subfertility compared with the general population rates (Brinton et al., 1989; Modan et al., 1998; Venn et al., 1999). Several studies suggest that endometriosis as a cause of subfertility may be associated with ovarian cancer (Brinton et al., 1997, 2005a,b; Paulson, 1997; Ogawa et al., 2000; Ness et al., 2002) and many studies have documented an increased rate of coincidental endometriosis in women with ovarian cancer, especially endometrioid and clear cell cancer (Heaps et al., 1990; Hitti et al., 1990; DePriest et al., 1992; McMeekin et al., 1995; Sainz de la Cuesta et al., 1996; Jimbo et al., 1997; Komiyama et al., 1999; Nezhat et al., 2008; Vlahos et al., 2010; Wei et al., 2011; Munksgaard and Blaakaer, 2012 ). Endometriosis is a common estrogen-dependent, chronic gynaecologic disorder associated with pelvic pain and subfertility. It is defined as the presence of endometrium-like glandular epithelium and stroma in ectopic locations outside the endometrium. Common anatomic locations for endometriosis are the ovaries, and other locations throughout the pelvis, and even outside the pelvis. The actual prevalence is difficult to determine, since laparoscopy or surgery is required to make a definitive diagnosis. The prevalence of pelvic endometriosis approaches 6 10% in the general female population; in women with pain, infertility or both, the frequency is 35 50% (Giudice and Kao 2004). The prevalence of newly diagnosed endometriosis in premenopausal women who requested a consultation with their general practitioner because of nongyanecological problems is 3.6% (Ferrero et al., 2010). Among adolescent girls undergoing laparoscopic investigation, the overall prevalence of visually confirmed endometriosis is 62%. In adolescent girls with chronic pelvic pain resistant to treatment the prevalence approaches 75% (Janssen et al., 2013). Endometriosis is a disease found almost exclusively in women of reproductive age, with a mean age at diagnosis reported from 25 to 29 years (Olive and Schwartz, 1993). Although several studies reported an association between endometriosis and ovarian cancer risk (Mostoufizadeh and Scully, 1980; Erzen and Kovacic 1989; Moll et al., 1990; Fukunaga and Ushigome 1998; Ness et al., 2000; Thomas and Campbell, 2000; Swiersz, 2001; Modesitt et al., 2002; Horiuchi et al., 2003; Modugno et al., 2004; Steed et al., 2004; Varma et al., 2004; Sayasneh et al., 2011), most studies used a case control design and suffered from various biases. There are only few cohort studies published about the association between endometriosis and ovarian cancer (Hankinson et al., 1995; Simpson et al., 2003; Brinton et al., 2005a,b; Melin et al., 2006; Kobayashi et al., 2007). In most of these reports information on endometriosis was obtained from health questionnaires, without medical information. We examined this association in a large-scale nation-wide cohort study that was initiated in 1995, the OMEGA project. A unique feature of this database is the fact that data on reproductive variables and other risk factors for hormone-related cancers were obtained from questionnaires mailed to the participants, whereas detailed information on subfertility causes and treatment was abstracted from the medical files (van Leeuwen et al., 2011). The majority of women in the OMEGA study underwent a diagnostic laparoscopy for the assessment of subfertility, and consequently detailed information on endometriosis was available. Histological data on endometriosis, when available, were extracted from the nationwide network and registry of histo- and cytopathology in the Netherlands (PALGA). Patients and Methods The present research question was examined in the OMEGA study cohort. First we describe the OMEGA study cohort and subsequently we provide a description of the present analytic cohort selected from the OMEGA cohort. The OMEGA study The study population, study procedures and data collection methods have been described elsewhere (Klip et al., 2001; de Boer et al., 2003; van Leeuwen et al., 2011). In short, the OMEGA study was initiated in 1995 to examine the late effects of hormone stimulation in IVF-treated women. This nationwide cohort study comprised 26 465 women with subfertility problems (Klip et al., 2003). Women were defined as subfertile when they were unable to achieve conception after 1 or more years of frequent unprotected intercourse. The OMEGA study was initiated in 1995 to examine the late effects of hormone stimulation in IVF-treated women who had completed at least one IVF treatment cycle with ovarian stimulation before 1 January 1995 (n ¼ 19 861). Women were treated in 1 of the 12 IVF hospitals with legal permission to provide IVF treatment in the Netherlands and we also included a comparison group of women not treated with IVF in 4 participating clinics that had a computerized registry of all subfertile women evaluated during 1980 1995 (n ¼ 6604). These women received other treatments, such as tubal surgery and/or hormonal treatments, but not IVF. Risk factor questionnaire survey A 23-page questionnaire was sent to 25 353 women between 1997 and 1999 to obtain information on gynaecological disorders before and after subfertility treatment, reproductive factors and several other lifestyle factors. From the initial 26 465 women, 1112 women (4.2%) were not approached because of death (n ¼ 97) or because of unknown, incomplete or foreign addresses (n ¼ 502), emigration (n ¼ 447) or other reasons specifically related to privacy (n ¼ 66). An extra form was attached asking each participant written informed consent for future linkage with the Netherlands Cancer Registry (NCR) and the Dutch pathology registry (PALGA), and data abstraction from the medical records. A total of 16 343 women returned the questionnaire (response rate 65.2%). 4.0% of the women refused linkage with the NCR or PALGA. Medical records Trained research assistants abstracted data from the medical files on gynaecological history, subfertility diagnosis, fertility hormones used prior to IVF

Endometriosis, subfertility and ovarian cancer 3 treatment and detailed information about each subsequent IVF treatment. Due to limited project funding we could only complete medical record abstraction for 9 out of 12 centres (76% of the cohort that gave informed consent to do so). Definition of analytic cohort of women with and without endometriosis To select women with endometriosis from the OMEGA cohort, we linked the cohort with PALGA and we used information from the medical records and risk factor questionnaires. PALGA contains records of all cytological and histological diagnoses made in the Netherlands, with computerized data submission by the individual pathology laboratories (Casparie et al., 2007). PALGA has nationwide coverage since 1989. Linkage of the whole OMEGA cohort with PALGA resulted in 968 women with a histo- or cytopathological diagnosis of endometriosis. Furthermore, based on the medical record data of 13 807 women, 2270 women had evidence of endometriosis (by laparoscopy or laparotomy). Of these 2270 women the diagnosis of endometriosis of 387 women had already been obtained from PALGA. Among women whose medical record could not be abstracted, we also included women who reported endometriosis as a cause of subfertility or as a gynaecological problem in the risk factor questionnaires (n ¼ 806). We chose to do so because the positive predictive value of endometriosis reported in the risk factor questionnaires was 73% for women whose medical record data were available (de Boer et al., 2005). Thus, in total 3657 women (968 + 1883 + 806) were included in the endometriosis group (Fig. 1). In the endometriosis group, 3067 women received IVF treatment and 590 women did not receive IVF treatment. Subsequently, we identified a comparison group of women without evidence of endometriosis in the medical records and PALGA: women with a male cause of subfertility (n ¼ 2619) or with an unexplained cause of subfertility (n ¼ 1834) based on their medical record. The idiopathic or unexplained cause of subfertility group was composed of couples with absolutely no abnormalities found in the subfertility work-up. Comparison of the risk factor questionnaires with the medical records for male subfertility yielded a positive predictive value of 71%, implying that women with a male cause of subfertility were able to recall correctly the cause of subfertility. Therefore, based on the health questionnaires received from women whose medical record had not been abstracted, we also included a total of 794 women with a male cause of subfertility but no report of endometriosis in the comparison group. We excluded 1276 women who had evidence of endometriosis in the medical records or PALGA. These women were included in the endometriosis group. Thus, in total 5247 women were included in the Figure 1 Endometriosis group and comparison group. a n ¼ 1017 actively refused linkage. b n ¼ 387, patients who also had evidence of endometriosis in their medical record. The medical record of n ¼ 383 had not been abstracted. c Evidence of endometriosis in the medical record (laparoscopy or laparotomy). d n ¼ 1276 women with male subfertility problems and unexplained subfertility excluded because of evidence of endometriosis. These patients were included in the endometriosis group. e Patients without medical record information (patients with medical record information are included above).

4 Buis et al. comparison group (Fig. 1). In the comparison group, 4769 women received IVF treatment and 478 women did not receive IVF treatment. Assessment of incidence of ovarian malignancies Cancer incidence between January 1989 and June 2007 was assessed through record linkage with the NCR and PALGA. The NCR is a population-based nationwide registry, established in 1989, and has 96% complete data on invasive malignant neoplasms occurring in the Netherlands (van den Brandt et al., 1990). For the present report on ovarian malignancies, we linked with PALGA since the NCR had incomplete data on BOT. PALGA is one of the main sources of the NCR. The NCR and PALGA granted us permission to not only link responders who gave permission, but also non-responders and deceased women, under additional privacy regulations. Only women who explicitly refused future linkage with disease registries (n ¼ 1017; 4.0% of all women) were excluded from linkage. For each ovarian malignancy we received information on date of diagnosis and morphology. Vital status in June 2007 and date of death of all women alive, when the risk factor questionnaire was mailed, were obtained by linking the cohort with the Central Bureau of Genealogy, which keeps computerized records of all deceased persons in the Netherlands since 1994. Statistical analysis The total analytic study cohort consisted of 8904 women: 3657 women with endometriosis and a comparison group of 5247 women without evidence of endometriosis (see Fig. 1). The observation time for each participant in the endometriosis group started on the date of first diagnosis of endometriosis, or on 1 January 1989 in case endometriosis was first diagnosed before 1989. This was done because PALGA/NCR has a nationwide coverage since 1989. Women diagnosed with ovarian cancer or BOT before the date of first diagnosis of endometriosis (n ¼ 2) were excluded from all analyses. The observation time for each participant in the comparison group started on 1 January 1989 or on the date of first IVF treatment or the first date of clinical visit for subfertility evaluation (if no IVF treatment was given), whichever came last. The observation time for each participant in both groups stopped on the date of end of PALGA follow-up (June 2007), date of first cancer diagnosis, date of bilateral oophorectomy (n ¼ 32) or date of death (n ¼ 42), whichever came first. Two separate analytic approaches were used. In the first approach we included events in women who were diagnosed with ovarian cancer or BOT on the same date or after the date of first diagnosis of endometriosis. In the second and main analytic approach we only included events in women who were diagnosed with ovarian cancer or BOT after the date of first diagnosis of endometriosis. This approach was used in the main analyses because cases with diagnosis of ovarian malignancies have a higher risk to be diagnosed with asymptomatic endometriosis (compared with women without ovarian cancer). Previous studies of the association between endometriosis and ovarian cancer were based on the diagnosis of endometriosis in the medical record or on self-report. Therefore, to compare our results with the literature we also performed an analysis restricted to women who had endometriosis according to their subfertility record or who self-reported endometriosis. The Cox proportional hazards model was used to directly compare ovarian cancer risk and risk of BOT between the endometriosis group and the comparison group without evidence of endometriosis and to explore the effect of concomitant variables on ovarian cancer risk and BOT. Adjustment was made for several variables, including age, OC use, IVF treatment and parity. Information on parity and OC use was derived from the risk factor questionnaire survey. For non-responding women information from the medical records was added when available. Cox models were fitted with the use of SPSS statistical software (SPSS, Inc., Chicago, IL, USA). Results Population characteristics Demographic characteristics of 3657 women in the endometriosis group and 5247 women in the comparison group are presented in Table I. Median follow-up time was 15.2 years for the entire study population. The median follow-up times until diagnosis of ovarian cancer and BOT were 10.9 and 9.5 years, respectively. Seventy-eight per cent of all women in our study were diagnosed with endometriosis by surgical assessment for subfertility cause and/or histological confirmation. The time interval between diagnosis of endometriosis and diagnosis of the ovarian cancer or BOT was 3 cases between 3 and 12 months, 7 cases between 1 and 10 years, 13 cases between 10 and 20 years and 3 cases 20 years or more. Risk of ovarian tumours associated with endometriosis Table II shows the results of Cox regression analysis of the effect of endometriosis on the risks of ovarian cancer, BOT and all ovarian tumours combined. In the first analytic approach, including ovarian cancers and BOTs diagnosed on the date of first diagnosis of endometriosis, we observed 29 cases (17 ovarian cancers and 12 BOT cases) in the endometriosis group and 5 in the comparison group (2 ovarian cancers and 3 BOT cases). Direct comparison of the endometriosis group with the comparison group yielded an age-adjusted hazard ratio (HR) of 9.7 (95% confidence interval (CI) 3.7 25.1) for all ovarian cancers and BOTs combined. For ovarian cancer, the age-adjusted HR associated with endometriosis was 13.4 (95% CI 3.1 58.4) and for BOT the age-adjusted HR associated with endometriosis was 7.3 (95% CI 2.0 26.3). The second and main analytic approach only included cases for whom the date of first diagnosis of endometriosis occurred before the date of diagnosis of the ovarian cancer or BOT. Hereby, we observed 26 cases in the endometriosis group (16 ovarian cancers and 10 BOT cases) and 5 in the comparison group (2 ovarian cancers and 3 BOT). In this approach, direct comparison of the endometriosis group with the comparison group yielded an age-adjusted HR of 8.2 (95% CI 3.1 21.6) for all ovarian cancers and BOT combined. The age-adjusted HR was 12.4 (95% CI 2.8 54.2) for ovarian cancer and 5.5 (95% CI 1.5 20.2) for BOT. We evaluated the potential confounding influence of parity, OC use and IVF treatment on the risk estimates for ovarian cancer and BOT associated with endometriosis. After adjustment for age, OC use, parity and IVF treatment the HR was 8.4 (95% CI 3.2 22.1) for all ovarian cancers and BOT combined, which did not materially differ from the HR only adjusted for age. Separate adjustment for OC use, parity and IVF treatment also did not materially alter the HR. When the first year after the diagnosis of endometriosis was excluded, the adjusted HR associated with endometriosis was 8.7 (95% CI 3.0 25.4) for all ovarian malignancies combined; the adjusted HRs for invasive ovarian cancer and BOT were 11.9 (95% CI 2.7 52.5) and 5.7 (95% CI 1.2 27.2), respectively. So the HRs excluding the first year after the diagnosis of endometriosis are very similar to those only excluding the cases with first diagnosis of endometriosis on the same date as diagnosis of the ovarian tumour. We also examined the association between endometriosis and the risk of ovarian malignancy in an analysis restricted to women who had endometriosis according to their subfertility record or who selfreported endometriosis, ignoring all additional information about

Endometriosis, subfertility and ovarian cancer 5 Table I Population characteristics. Endometriosis Comparison group group (n 5 3657)... (n 5 5247)... No. % No. %... Year of birth 1955 778 21.3 836 15.9 1955 1959 1382 37.8 1819 34.7 1960 1964 1125 30.8 1882 35.9 1965 372 10.2 710 13.5 Age (years) at diagnosis of endometriosis or first visit,25 351 9.6 182 3.5 25 29 1314 35.9 1258 24.0 30 34 1300 35.5 2301 43.9 35 39 527 14.4 1326 25.3 40 165 4.5 180 3.4 Time since diagnosis of endometriosis or first visit (years),5 75 2.1 150 2.9 5 9 209 5.7 238 4.5 10 14 934 25.5 2725 51.9 15 19 1554 42.5 1962 37.4 20 885 24.2 172 3.3 OC use (years) No OC use 426 11.6 708 13.5 1 4 775 21.2 1059 20.2 5 9 1075 29.4 1583 30.2 10 475 13.0 721 13.7 Unknown 906 24.8 1176 22.4 Number of children 0 1510 41.3 2060 39.3 1 2 1775 48.5 2873 54.8 3 160 4.4 226 4.3 Unknown 212 5.8 88 1.7 Main cause of subfertility Tubal 711 19.4 Male 579 15.8 3413 65.0 Unexplained 696 19.1 1834 35.0 Endometriosis 468 12.8 Ovarian 49 1.3 Cervical 19 0.5 Mixed 831 22.7 Unknown 304 8.4 IVF No 592 16.2 478 9.1 Yes 3065 83.8 4769 90.9 and 5 BOT cases) and 15 in the comparison group (7 ovarian cancers and 8 BOT). Table III shows the results of this analysis (restricted to ovarian malignancies occurring after the date of diagnosis of endometriosis). An age-adjusted HR of 3.0 (95% CI 1.5 6.1) was found for all ovarian cancers and BOT combined, with age-adjusted HRs of 4.3 (95% CI 1.6 11.2) for ovarian cancer and 1.9 (95% CI 0.6 5.8) for BOT. After adjustment for all confounders (age, OC use, parity and IVF treatment) the HR was 2.9 (95% CI 1.4 5.8) for all ovarian cancers and BOT combined; 4.1 (95% CI 1.6 10.7) for ovarian cancer and 1.7 (95% CI 0.6 5.4) for BOT separately. When we excluded the first year after the diagnosis of endometriosis, results were very similar. Women with endometriosis are often prescribed OCs. Since OCs are known to reduce the risk of ovarian malignancies (Beral et al., 2008), we wondered whether women with endometriosis who used OC have a lower risk than women with endometriosis who did not. Among women with endometriosis who used OC for 5 years or longer, the HR for ovarian cancer was increased to a lesser extent (HR 2.7(95% CI 0.7 10.4), compared with women without endometriosis) than for women who used OC for,5 years (HR6.8, 95% CI 2.3 20.5). Future studies are needed to explore this effect of OC use on the development of ovarian cancer and BOT and to find out whether OC can be used to diminish the risk of development of ovarian cancer and BOT in women with endometriosis. We also examined whether the association between endometriosis and ovarian malignancy varied by age at cancer diagnosis. Using the second analytic approach (and restricted to women who had endometriosis according to their subfertility record or self-reported), the age-adjusted HRs of ovarian tumours diagnosed before (n ¼ 12) and after (n ¼ 19) the age of 43 (median age at diagnosis) were 2.8 (95% CI 0.9 9.4) and 3.4 (95% CI 1.4 8.5), respectively. The endometriosis group (n ¼ 3657) consisted of 1327 (36.3%) women with ovarian endometriosis and 535 (14.6%) women with extraovarian endometriosis (and no ovarian endometriosis), while for 1795 (49.1%)womenthesiteofendometriosiswasunknown(TableIV). The age-adjusted HR of all ovarian tumours (Table II) was 11.3 (95% CI 4.0 31.8) for ovarian endometriosis and 7.7 (95% CI 2.1 28.7) for extraovarian endometriosis. Comparison of women with endometriosis as (one of) the cause(s) of subfertility with women without endometriosis as a subfertility cause showed an age-adjusted HR for all ovarian malignancies combined (Table III) of 3.4 (95% CI 1.3 8.9) for ovarian endometriosis and 3.6 (95% CI 1.0 12.4) for extraovarian endometriosis. For ovarian cancer the age-adjusted HRs associated with ovarian and extraovarian endometriosis were 4.8 (95% CI 1.4 16.6) and 7.6 (95% CI 2.0 29.7), respectively. For BOT the age-adjusted HR associated with ovarian endometriosis was 2.2 (95% CI 0.5 10.6). The morphologies of BOTs and ovarian cancers in the endometriosis group and the comparison group are given in Table V. Comparisons are hindered by small numbers of events in the comparison group. endometriosis from the PALGA linkage. In this analysis we compared theriskof ovarian malignancies between women for whom endometriosis was (one of) the cause(s) of subfertility (n ¼ 2288) with women who had other causes of subfertility (n ¼ 6616). In this analysis, we included 17 cases in the endometriosis group (12 ovarian cancers Discussion We found a 3- to 8-fold increased risk of ovarian tumours associated with endometriosis: the magnitude of the risk increase depended on the sources of information used to assess the presence of endometriosis.

6 Buis et al. Table II Risk of ovarian tumours associated with endometriosis. All cases (n 5 34)... Ovarian cancer (n 5 19)... BOT (n 5 15)... HR 95% CI HR 95% CI HR 95% CI... First analytic approach No endometriosis (n ¼ 5247) 1.0 Ref. 1.0 Ref. 1.0 Ref. Any endometriosis (n ¼ 3657) Crude 7.9 3.0 20.3 11.6 2.7 50.2 5.4 1.5 19.1 Age adjusted 9.7 3.7 25.1 13.4 3.1 58.4 7.3 2.0 26.3 All cases (n ¼ 31) Ovarian cancer (n ¼ 18) BOT (n ¼ 13) HR 95% CI HR 95% CI HR 95% CI Second analytic approach a Any endometriosis Crude 7.0 2.7 18.3 10.9 2.5 47.4 4.4 1.2 16.1 Age adjusted 8.2 3.1 21.6 12.4 2.8 54.2 5.5 1.5 20.2 Adjusted for all confounders b,c 8.4 3.2 22.1 12.7 2.9 55.5 5.5 1.5 20.4 Ovarian endometriosis d 11.3 4.0 31.8 15.0 3.1 72.4 8.9 2.2 35.7 Extraovarian endometriosis d 7.7 2.1 28.7 19.1 3.5 104.5 e e Unknown location endometriosis d 6.0 2.0 18.1 8.1 1.6 41.8 4.7 1.0 21.5 BOT, borderline ovarian tumour; HR, hazard ratio; CI, confidence interval; OC use, oral contraceptive use; first analytic approach: including the cases with first diagnosis of endometriosis on the same date as diagnosis of the ovarian tumour; second analytic approach: only cases with first diagnosis of endometriosis before the date of diagnosis of the ovarian tumour. a The time interval between diagnosis of endometriosis and diagnosis of the ovarian cancer or BOT: 3 cases (3 and 12 months), 7 cases (1 and 10 years), 13 cases (10 and 20 years), 3 cases (20 years or more). b After adjustment for age (two decimal places), OC use (,5 versus 5 years), child (Y/N), IVF (Y/N). c OC use was missing in similar proportions of women in the endometriosis group (24.8%) and the comparison group (22.4%). However, parity was missing among 5.8% of the women in the endometriosis group but only among 1.7% of the women in the comparison group; biased loss of these data may have slightly affected our estimates. d Subgroup analyses, after adjustment for age. e Numbers are not sufficient for analysis. The risk was increased both for ovarian cancer and BOT, and could not be explained by potentially confounding factors. Furthermore, both ovarian endometriosis and extra ovarian endometriosis carried a significantly increased risk for ovarian cancer and BOT. Strikingly, the risk of ovarian malignancies associated with endometriosis was much higher in analyses including lifetime information on endometriosis from the nationwide pathology database, implying that risk estimates from studies using self-reported information on endometriosis may be too low due to non-differential misclassification bias. Several other epidemiologic investigations support a link between endometriosis and ovarian cancer. A cohort study published by Brinton et al. (1997) on 20 686 women hospitalized for endometriosis found an increased number of ovarian cancers among women with a long-standing history of ovarian endometriosis [standardized incidence ratio (SIR) 4.2; 95% CI 2.0 7.7]. Hospitalizations were identified through the nationwide Swedish Inpatient Registrar, and outcomes were identified through the National Swedish Cancer Registry after a mean of 11.4 years of follow-up. A limitation of this study was an inability to control for the confounding factors, parity or OC use. Furthermore, therewas no precise information on the types of gynaecologic operations (for example the removal of the ovaries) in the history of the women with endometriosis. There was also no information on the histology of the ovarian cancers. In a subsequent retrospective cohort study of 12 193 subfertile women, Brinton et al. (2004) found that women with endometriosis had the highest risk of developing ovarian cancer (SIR 2.48; 95% CI 1.3 4.2), compared with women of the female population. Although the number of women diagnosed with endometriosis was smaller than in our study (n ¼ 1919 versus n ¼ 3657), all endometriosis was diagnosed by surgery. There was no histological confirmation of endometriosis. There was also no information on the exact anatomic localization of the endometriosis and the histology of ovarian cancer. The median length of follow-up among subjects included in analysis was very long, i.e. 18.8 years. Stewart et al. (2013) examined in a retrospective cohort study (n ¼ 21 646) the risk of invasive epithelial ovarian cancer in women seeking treatment for infertility in the years 1982 2002. They examined the effect of IVF treatment, endometriosis and parity on risk of ovarian cancer. Women diagnosed with endometriosis who remained nulliparous had a 3-fold increase in the rate of ovarian cancer (HR 3.11; 95% CI 1.13 8.57); the HR in parous women was 1.52 (95% CI 0.34 6.75). The strength of this study was that they explored potential confounding by tubal ligation, hysterectomy and unilateral oophorectomy or salpingo-oophorectomy. A limitation is that there was no confounding control for OC use, and there was no information on the histology of the ovarian cancers or information on the exact anatomic location of the endometriosis. In the largest published case control study, Ness et al. (2002) pooled 5207 cases and 7705 controls (from eight studies) to investigate the influence of subfertility on risk of ovarian cancer. Endometriosis was associated with a significantly increased risk of ovarian cancer (OR 1.73; 95% CI 1.10 2.71). The association between ovarian cancer and endometriosis persisted after adjustment for multiple confounding factors, such as parity and OC use. This study suffers from the same limitations as any other (pooled) case control

Endometriosis, subfertility and ovarian cancer 7 Table III Risk of ovarian tumours in women with endometriosis as one of the cause(s) of subfertility versus women with other causes of subfertility. All cases (n 5 31) Ovarian cancer BOT (n 5 13)... (n 5 18)...... HR 95% CI HR 95% CI HR 95% CI... Second analytic approach No endometriosis (n ¼ 6616) 1.0 Ref. 1.0 Ref. 1.0 Ref. Any endometriosis (n ¼ 2288) Crude 2.7 1.3 5.5 3.9 1.5 10.3 1.6 0.5 5.0 Age adjusted 3.0 1.5 6.1 4.3 1.6 11.2 1.9 0.6 5.8 Adjusted for all confounders a,b 2.9 1.4 5.8 4.1 1.6 10.7 1.7 0.6 5.4 Ovarian endometriosis c 3.4 1.3 8.9 4.8 1.4 16.6 2.2 0.5 10.6 Extraovarian endometriosis c 3.6 1.0 12.4 7.6 2.0 29.7 e Unknown location of endometriosis c 2.5 1.0 6.3 2.9 0.8 10.2 2.2 0.6 8.6 Med. confirmed endometriosis d 3.0 1.3 6.6 3.5 1.2 10.4 2.4 0.7 8.0 Endometriosis with OC use,5 years 3.3 1.3 8.5 6.8 2.3 20.5 e Endometriosis with OC use 5 years 2.1 0.8 5.9 2.7 0.7 10.4 e BOT, borderline ovarian tumour; HR, hazard ratio; CI, confidence interval; OC use, oral contraceptive use; first analytic approach: including the cases with first diagnosis of endometriosis on the same date as diagnosis of the ovarian tumour; second analytic approach: only cases with first diagnosis of endometriosis before the date of diagnosis of the ovarian tumour. a After adjustment for age (two decimal places), OC use (,5 versus 5 years), child (Y/N), IVF (Y/N). b OC usewas missing in similar proportions of women in the endometriosis group (22.6%) and the comparison group (23.7%). However, parity was missing among 0.9% of the women in the endometriosis group and among 4.2% of the women in the comparison group; biased loss of these data may have slightly affected our estimates. c Subgroup analyses, after adjustment for age. d Subgroup analyses, with evidence of endometriosis by surgery and/or pathology (exclusion of self-reported endometriosis), after adjustment for age. BOT (n ¼ 11), Ovarian cancer (n ¼ 14). e Numbers are not sufficient for analysis. Table IV Anatomic localization of endometriosis. Anatomic localization of the Endometriosis group a (n 5 3657) Ovarian cancer (n 5 17) Borderline tumour (n 5 12) endometriosis (%) (%) (%)... Ovarian 1327 (36.3) 7 (41.2) 7 (58.3) Extra-ovarian 535 (14.6) 4 (23.5) 1 (8.3) Unknown 1795 (49.1) 6 (35.3) 4 (33.3) Endometriosis group b (n ¼ 2288) (%) Ovarian cancer (n ¼ 12) (%) Borderline tumour (n ¼ 5) (%) Ovarian 708 (30.9) 4 (33.3) 2 (40) Extra-ovarian 312 (13.6) 3 (25.0) 0 Unknown 1268 (55.4) 5 (41.7) 3 (60) a Women diagnosed with endometriosis. b Women with endometriosis as the (or one of the) cause(s) of subfertility. study, since in all studies endometriosis was self-reported by the women, after the diagnosis of ovarian cancer had been made. Only one previous cohort study investigated the relationship between endometriosis and BOT (Brinton et al., 2005a,b). In this study, no increased risk of BOT was found in patients previously diagnosed with endometriosis. In our study, ovarian endometriosis did not carry a significantly higher risk of ovarian cancer than extra-ovarian endometriosis. Both localizations of endometriosis showed significantly increased risks, although there was insufficient power to explore this effect because the numbers of cases are small. Localization of endometriosis was unknown for 49.1% of women in the total analytic study cohort and for 33.3% of women with BOT and 35.3% of women with ovarian cancer. The ovaries are a common site of implants of endometriosis, with literature data available that 54.9% of women in the reproductive age have either unilateral or bilateral involvement (Jenkins et al., 1986). However, the prevalence by site and subtype varies dramatically by diagnostic pathway. With magnetic resonance imaging (MRI), deep infiltrating endometriosis is more prevalent. Superficial peritoneal endometriosis is more prevalent with laparoscopy, and can be missed with MRI.

8 Buis et al. Table V Histology of ovarian tumours. Histology Endometriosis Comparison group group... Women diagnosed with endometriosis Ovarian cancers n ¼ 17 n ¼ 2 Serous 7 (41.2) 1 (50) Mucinous 1 (5.9) 1 (50) Endometrioid 4 (23.4) Clear cell 2 (11.8) Mixed epithelial 1 (5.9) Adenocarcinoma 2 (11.8) NOS a Borderline tumours n ¼ 12 n ¼ 3 Serous 6 (50) 2 (66.7) Mucinous 6 (50) 1 (33.3) Women with endometriosis as the (or one of the) cause(s) of subfertility. Ovarian cancers n ¼ 12 n ¼ 7 Serous 5 (41.7) 3 (42.9) Mucinous 1 (8.3) 1 (14.2) Endometrioid 1 (8.3) 3 (42.9) Clear cell 2 (16.7) Mixed epithelial 1 (8.3) Adenocarcinoma 2 (16.7) NOS a Borderline tumours n ¼ 5 n ¼ 8 Serous 2 (40) 5 (60) Mucinous 3 (60) 3 (40) a NOS, not otherwise specified. We also wanted to examine whether endometriosis was stronger associated with the endometrioid and clear cell histology of ovarian cancer. Since only two cases of ovarian cancer arose in the comparison group, we could not investigate this in our own database. However, according to the NCR, an average of 11.5% of epithelial ovarian cancers is endometrioid and 4% is of clear cell histology. In the literature, the percentage of women with endometrioid ovarian cancer and associated endometriosis has ranged from 8 to 38%, and the percentage of women with clear cell ovarian carcinoma and associated endometriosis has ranged from 21 to 55% (DePriest et al., 1992; Vercellini et al., 1993; McMeekin et al., 1995; Sainz de la Cuesta et al., 1996; Jimbo et al., 1997; Komiyama et al., 1999; Erzen et al., 2001; Heintz et al., 2001; Stern et al., 2001; Kobayashi et al., 2008). Although the number of cases in our study was small, we appear to confirm these earlier findings since endometrioid and clearcell cancers accounted for 35.3% of our cases in the endometriosis group. The subtypes of ovarian cancer associated with endometriosis (clear cell and endometrioid) present at an earlier age than those not associated with endometriosis (Goodman and Howe, 2003). With a median follow-up time of 15.2 years, this may also be an explanation of the number of clear cell and endometrioid subtypes finding in our study population. The association of endometriosis with endometrioid and clear cell carcinoma of the ovary seems to support the hypothesis that endometriosis represents a preneoplastic condition for these two types of ovarian cancer. However, current research does not fully support this hypothesis. Although 30% of endometriosis lesions display changes in chromosomal regions that are also involved in ovarian carcinogenesis (Simpson et al., 2003), only few chromosomal mutations have been reported. Furthermore, epigenetic changes shared between endometriosis and ovarian cancers have not been investigated (Vigano et al., 2006). Several studies have evaluated pathways believed to be important in ovarian carcinogenesis and determined that endometrioid and clear cell tumours differ from the traditional serous tumours with regard to the presence of p53 mutations, PTEN expression, estrogen receptors, galactose-1-phosphate uridyltransferase and gluthathione S-transferase activity (Obata et al., 1998; Baxter et al., 2001; Fujimura et al., 2001; Kobayashi et al., 2011; Kurman and Shih, 2011; Wei et al., 2011). In conclusion, an association between endometriosis and endometrioid and clear cell tumours of the ovary has been found, but further work is required to explore this association. Endometriosis as a risk factor to develop ovarian cancer has, in theory, a biological basis. The endometriosis hypothesis, put forward by Paulson (1997), postulated that endometriosis may act to promote the development of ovarian cancer if endometriotic implants cause an inflammatory reaction. The cause of endometriosis was hereby thought to be retrograde menstruation, i.e. the menstrual fluid flows backwards through the Fallopian tubes into the pelvic cavity. This is supported by the decreased risk of ovarian cancer in women who underwent tubal ligation or hysterectomy (Irwin et al., 1991; Sightler et al., 1991; Hankinson et al., 1993; Parazzini et al., 1993; Cornelison et al., 1997; Green et al., 1997; Loft et al., 1997; Miracle-McMahill et al., 1997; Rosenblatt and Thomas, 2004; Cibula et al., 2011; Rice et al., 2012). Furthermore, both endometriosis and ovarian cancers are associated with unopposed estrogens. Possibly, an estrogen-rich, progesterone-poor hormonal environment may encourage the growth of endometriosis and promote its malignant transformation to ovarian cancer (Ness, 2003; van Gorp et al., 2004). Hypothetically, parity and OC use may have beneficial effects on immune balance, inflammation and retrograde menstruation (Paulson, 1997; Ness, 2003; Giudice and Kao 2004), for example by reducing the number of menstrual cycles and the amount of menstrual flow. These factors may explain, in part, how OCs and parity reduce ovarian cancer risk among women with endometriosis. When comparing our results with the literature, several strengths and limitations of our study should be considered. Our study is one of the few published cohort studies evaluating the risk of ovarian cancer in subfertile women diagnosed with endometriosis. Most previous investigations were case control studies, with self-reported diagnosis of endometriosis after the diagnosis of ovarian cancer, which may have led to recall bias and non-differential misclassification bias. One of the strengths of our study is that we were able to collect extensive medical record data on the presence of endometriosis of the participating women. Another unique feature is the linkage of the OMEGA cohort with PALGA and the NCR, even for deceased women and non-responders. Therefore, follow-up in our cohort study was complete for 96% of the women and we were able to add confirmation by pathology of the diagnosis of endometriosis for a large group of women. Through this linkage and the extensive medical record data on subfertility causes and gynaecological disorders, 78% of all women in our study were diagnosed with endometriosis by pathology report. In comparison with the few previously published cohort studies, one of the strengths of our study is

Endometriosis, subfertility and ovarian cancer 9 that, in view of the known protective effect of parity on the risk of hormone-related cancers, we also collected data from the women themselves on reproductive events and OC use. We also obtained data on the anatomic localization of endometriosis, but information about exact localization of endometriosis was available for only 50.9% of the women. A unique feature of our study is that even women with minor endometriosis could be included, because a large proportion of our cohort had had a diagnostic laparoscopy. To compare our study with previous cohort studies (Brinton et al., 1997, 2005a,b; Paulson, 1997; Ogawa et al., 2000; Ness et al., 2002) we also did an analysis with women who had endometriosis as a cause of subfertility (n ¼ 2288). With this analysis we still found an increased risk (3-fold) of ovarian tumours associated with endometriosis, but the magnitude of the risk increase was much lower implying that risk estimates from studies using self-reported information on endometriosis may be too low due to non-differential misclassification bias. A limitation of our study is that, due to limited funding, we were not able to abstract information from the medical records of all women in the endometriosis group. There was no selection bias, because the medical records of 24.0% of the cohort that had given informed consent, were the women of the last three participating clinics. However, among women whose medical record data were available, comparison of the medical records with the risk factor questionnaires gave a positive predictive value of 73% for self-reported endometriosis. Therefore, we also included women with self-reported endometriosis whose medical record could not be abstracted (22.0% of the endometriosis group). To evaluate potential bias from this approach, we did an analysis without the 806 women without medical confirmation of endometriosis. The age-adjusted HRs in the second analytic approach of developing ovarian cancer and BOT (Table III) were still high (HR 3.5; 95% CI 1.2 10.4 and HR 2.4; 95% CI 0.7 8.0, respectively), which renders misclassification unlikely. Another limitation is that we did not have information on OC use and parity for 23.4 and 3.4%, of women in the analytic cohort, respectively. Furthermore, an important limitation of our study, and also of other studies, is that the date of diagnosis of endometriosis is usually made long after the onset of the disease. Endometriosis is a chronic disease associated with a wide variety of symptoms, and in many patients even asymptomatic. Because laparoscopy or surgery is required to make a definitive diagnosis, many women in our study without specific complaints of endometriosis were not diagnosed with endometriosis before their assessment for subfertility, or gynaecological surgeries for other reasons. Because women with ovarian cancer have a higher risk to be diagnosed with asymptomatic endometriosis (compared with women without ovarian cancer), our second analysis (Table II) only included women who were diagnosed with ovarian cancer or BOT after the date of first diagnosis of endometriosis (HR 8.2; 95% CI 3.1 21.6). Most patients were diagnosed with endometriosis years before the malignancy; 23 of the 26 cases had a time interval of more than a year (13 cases even between 10 and 20 years). The minimum time interval between diagnosis of endometriosis and diagnosis of the ovarian cancer case was between 3 and 12 months (only 3 of 26 cases). When the first year after diagnosis of endometriosis was excluded, the adjusted HR associated with endometriosis was 8.7 (95% CI 3.0 25.4) for all ovarian malignancies, compared with an HR of 8.4 (95% CI 3.2 22.1) when weonly excluded the cases with first diagnosis of endometriosis on the same date as diagnosis of the ovarian tumour. So the HRs excluding the first year are very similar to those only excluded the cases with first diagnosis of endometriosis on the same date as diagnosis of the ovarian tumour. Although our HRs from the second analytic approach are the most valid risk estimates, it is not possible to predict how HRs would change if the exact exposure time from start of endometriosis to tumour development in our cohort were known. It has been reported that there is a 6 7 year delay between onset of symptoms and a surgical diagnosis (Nnoaham et al., 2011). Therefore, we analysed the data excluding all ovarian cancers diagnosed within 6 years of follow-up after endometriosis. Although the number of cases available for this analysis was substantially reduced (n ¼ 21), the increased risk associated with endometriosis remained [HR of 8.3 (95% CI 2.4 28.4)]. Because the number of cases in our study was relatively small, there was insufficient power to compare the effects of endometriosis on the development of ovarian malignancies and BOT. Another concern is that because of the complaints associated with endometriosis, there is a risk of ascertainment bias. In theory, women with endometriosis associated subfertility have a higher risk of undergoing surgery with coincidental findings of ovarian malignancies and BOT. Another limitation is the difference between endometriosis diagnostic criteria used during 1980 1995 and the current criteria. Within the exposure timeframe for this cohort, endometriosis was typically diagnosed when there were superficial lesions on the ovaries, serosal surfaces and peritoneum (black, dark-brown or bluish lesions). With the current knowledge of the broad presentation of endometriosis (for example atypical or subtle lesions of red implants and serous or clear vesicles, white plaques, scarring and yellow-brown peritoneal discoloration of the peritoneum), endometriosis may have been underdiagnosed in the past. This may have led to underestimation of the ovarian cancer risk associated with endometriosis. We identified the comparison group of women without evidence of endometriosis by selecting women with a male cause of subfertility and with an unexplained cause of subfertility in the medical records. In this group, 1276 women with male subfertility and unexplained subfertility were excluded because of evidence of endometriosis in PALGA. These women were included in the endometriosis group. We selected women with these two causes of subfertility for the comparison group because they had the lowest risk of having undiagnosed endometriosis. In the Netherlands, in the study period, women with unexplained subfertility and couples with mild and moderate male infertility received the complete diagnostic work-up, mostly including diagnostic laparoscopy. In conclusion, our observations confirm previous reports that endometriosis increases the risk of ovarian cancer. The fact that endometriosis was assessed before diagnosis of ovarian malignancy and the high degree of medical confirmation in our study likely contribute to the validity of our estimate of a 3- to 8-fold increased risk of ovarian tumours associated with endometriosis. The risk of ovarian malignancies associated with endometriosis was much higher in analyses including information on endometriosis from the nationwide pathology database, implying that risk estimates from studies using self-reported information on endometriosis may be too low due to non-differential misclassification bias. Because this cohort is restricted to subfertile women, there is a generalizability issue. The majority of women with endometriosis do not experience fertility problems and from our study we cannot tell whether the association with ovarian cancer in a fertile population is as strong as in subfertile women. Also, the number of cases in the cohort is small, resulting in wide CIs. Future research is needed to explore if the risk associated with endometriosis is similarly increased in OC users and non-oc users,