{"paper_id":"771f58b8-ae45-43a5-8d45-120325b2398d","body_text":"The study was conducted in line with the recommendations from the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines ( Liberati  et al , 2009 ). For the meta-analysis, we searched PubMed, EmBase and the Cochrane Central Register of Controlled Trials (CENTRAL) in the Cochrane Library for relevant studies published online between January 1990 and December 2012. The search terms used were the following: ‘ovarian tumor and endometriosis', ‘ovarian neoplasm and endometriosis', ‘ovarian carcinoma and endometriosis' and ‘ovarian cancer and endometriosis'.\nWe included relevant studies that met the following criteria: (1) epithelial ovarian cancer; (2) case–control or two-arm cohort studies comparing ovarian cancer risk between women with endometriosis and those without endometriosis; (3) single-arm cohort studies comparing ovarian cancer risk between observed and expected events of ovarian cancer in only women with endometriosis; and (4) studies comparing progression-free survival, overall survival and clinicopathologic characteristics between EAOC and non-EAOC patients. However, we excluded studies as follows: (1) review articles; 2) case reports or editorials or letters to the editor not including original data; (3) studies not meeting the selection criteria; and (4) non-English literature.\nTwo of the authors (HSK and HHC) independently evaluated potential eligibility of all studies retrieved from the database according to the predetermined selection and exclusion criteria, and the third author (YSS) resolved disagreement between the two authors after discussion. As a result, a total of 1625 studies were identified, and we excluded 89 duplicates and an additional 624 including reviews ( n =294), case reports ( n =157), non-English literature ( n =145), editorials or letters to the editor ( n =25), and relevant pooled analyses where we could not obtain individual data from each study, and data from some studies overlapped with those included in the meta-analysis ( n =3) ( Ness  et al , 2002 ;  Modugno  et al , 2004 ;  Pearce  et al , 2012 ). In addition, we excluded 860 studies because of non-ovarian cancer ( n =640), no endometriosis ( n =87) and no data about clinicopathologic characteristics, ovarian cancer risk or prognosis ( n =133). Furthermore, 17 were also excluded because of no appropriate comparator ( n =16), and not enough data to calculate survival ( n =1). Finally, 20 case–control ( Ness  et al , 2000 ;  Ziogas  et al , 2000 ;  Royar  et al , 2001 ;  Erzen  et al , 2001 ;  Borgfeldt and Andolf, 2004 ;  Glud  et al , 2004 ;  Pike  et al , 2004 ;  Terry  et al , 2005 ;  Risch  et al , 2006 ;  Merritt  et al , 2008 ;  Moorman  et al , 2008 ;  Rossing  et al , 2008 ;  Cunningham  et al , 2009 ;  Wu  et al , 2009 ;  Lurie  et al , 2010 ;  Balogun  et al , 2011 ;  Bodmer  et al , 2011 ;  Kumar  et al , 2011 ;  Ness  et al , 2011 ;  Vitonis  et al , 2011 ) and 15 cohort studies including 444 255 patients were included in the meta-analysis ( McMeekin  et al , 1995 ;  Brinton  et al , 1997 ;  Komiyama  et al , 1999 ;  Olson  et al , 2002 ;  Brinton  et al , 2004 ;  Brinton  et al , 2005 ;  Melin  et al , 2006 ;  Kobayashi  et al , 2007 ;  Melin  et al , 2007 ;  Orezzoli  et al , 2008 ;  Aris, 2010 ;  Melin  et al , 2011 ;  Cuff and Longacre, 2012 ;  Katagiri  et al , 2012 ;  Wang  et al , 2013 ;  Supplementary Figure 1 ).\nData extraction was also performed by the two authors (HSK and THK), and any discrepancies were addressed by a joint reevaluation of the article with the third author (YSS). The following data were independently extracted from each study for the meta-analysis: the first author; period of enrollment; study design; assessment of endometriosis; age; numbers of women with endometriosis and those without endometriosis in case–control or two-arm cohort studies; numbers of observed and expected events of ovarian cancer, sample size and a number of person-years in single-arm cohort studies; adjustment for potential confounding factors; the International Federation of Gynecology and Obstetrics (FIGO) stage; grade; nulliparity; optimal debulking surgery; histology; numbers of EAOC and non-EAOC patients; and progression-free survival or overall survival. When there was a lack of the relevant data in some studies, we could obtain the formation from some authors whom we contacted or databases suggested from systematic reviews or pooled analyses ( Ness  et al , 2002 ;  Modugno  et al , 2004 ;  Sayasneh  et al , 2011 ;  Pearce  et al , 2012 ).\nWe assessed the quality of each study using the Newcastle–Ottawa Scale (NOS) for included case–control and cohort studies (Wells  et al ). The NOS consists of three parameters of quality: selection, comparability and exposure (for a case–control study) or outcome (for a cohort study). It assigns a maximum of four points for selection, two points for comparability and three points for exposure or outcome. In the current study, we considered a study with NOS score ⩾7 as a high-quality study because it has been used as the criteria of high-quality study in spite of no standard criteria ( Myung  et al , 2009 ;  Castillo  et al , 2011 ). In case–control studies, 15 (75%) were of high quality with an average NOS score of 6.9 ( Supplementary Table 1 ), and 10 (66.6%) showed high quality with an average NOS score of 7.6 in cohort studies ( Supplementary Table 2 ).\nDichotomous data eligible in each study were shown as a risk ratio (RR) with its 95% confidence interval (CI) in case–control or two-arm cohort studies. In the meta-analysis using single-arm cohort studies, standard incidence ratio (SIR), which was computed as the observed number of events divided by the expected number of events in only women with endometriosis, and 95% CI were calculated. Moreover, we performed survival analyses using the statistical procedure described by  Tierney  et al  (2007 ). Heterogeneity was assessed using Higgins  I 2  that measures the percentage of total variation across studies that is due to heterogeneity rather than chance ( Higgins  et al , 2003 ). An  I 2 >50% was considered to represent substantial heterogeneity, and we used the random effects model using the DerSimonian and Laird method. On the other hand, the fixed effect model using the Mantel–Haenszel method was used in this meta-analysis when the  I 2  was ⩽50% because it indicated no heterogeneity.\nFor identifying publication bias, funnel plots were represented that were scatter plots of hazard ratios (HRs) or RRs or SIRs of individual studies on the X axis against the standard error of the log HR or log RR or log SIR of each study on the Y axis. As a result, all funnel plots resembled symmetric inverter funnels that suggested no publication bias in this meta-analysis. Furthermore, we also found no publication bias using Egger's test ( P >0.05) ( Supplementary Figure 2 ). For this analysis, we used Comprehensive Meta-analysis Version 2.0 (Biostat Inc., Englewood, NJ, USA), and a  P <0.05 was considered statistically significant.\n\nSupplementary Tables 3 and 4  show general characteristics of 18 case–control or three two-arm cohort studies including 314 421 women with or without endometriosis, and five single-arm cohort studies including 79 388 women with endometriosis. Potential confounding factors including age, parity, history of tubal ligation and use of oral contraceptive were adjusted in most of studies. As a result, ovarian cancer risk increased in women with endometriosis when compared with those without endometriosis in case–control or two-arm cohort studies (RR, 1.265; 95% CI, 1.214–1.318;  Figure 1A ), and single-arm cohort studies (SIR, 1.797; 95% CI, 1.276–2.531;  Figure 1B ). When we performed subgroup analyses based on study design, assessment of endometriosis, quality of study and adjustment for potential confounding factors, all results also showed that endometriosis was associated with an increased risk of ovarian cancer ( Table 1 ).\nNext, we compared progression-free survival and overall survival between EAOC and non-EAOC patients in eight relevant studies with NOS score ⩾7 that included 47 047 patients, the characteristics of which are summarised in  Supplementary Table 5 . In most of the studies, patients with EAOC were relatively young in comparison with those with non-EAOC. In terms of survival, there was no difference in progression-free survival between EAOC and non-EAOC (HR, 1.023; 95% CI, 0.712–1.470;  Figure 2A ), whereas EAOC was associated with a better overall survival that non-EAOC in crude analyses (HR, 0.778; 95% CI, 0.655–0.925;  Figure 2B ). However, there were no differences in progression-free survival and overall survival between EAOC and non-EAOC in subgroup analyses based on histology, assessment of endometriosis, FIGO stage and adjustment for potential confounding factors ( Table 2 ).\nFinally, we evaluated clinicopathologic characteristics between EAOC and non-EAOC in six cohort studies including 46 563 patients and 15 case–control studies including 8417 patients. General characteristics are depicted in  Supplementary Table 6 . In crude analyses, FIGO stage I–II disease (RR, 1.959; 95% CI, 1.367–2.807;  Figure 3A ), grade 1 disease (RR, 1.319; 95% CI, 1.149–1.514;  Figure 3B ) and nulliparity (RR, 1.327; 95% CI, 1.245–1.415;  Figure 3C ) were more common in EAOC, whereas there was no difference in probability of optimal debulking surgery between EAOC and non-EAOC (RR, 1.403; 95% CI, 0.915–2.152;  Figure 3D ). In subgroup analyses according to study design, assessment of endometriosis, quality of study and adjustment for potential confounding factors, the results were similar except no difference in grade 1 disease in studies with NOS score <7 (RR, 1.087; 95% CI, 0.518–2.280;  Table 3 ).\nIn terms of histology, crude analyses showed that serous carcinomas were less frequent in EAOC than in non-EAOC (RR, 0.733; 95% CI, 0.617–0.871;  Figure 3E ), and there was no difference in the risk of mucinous carcinomas between the two groups (RR, 0.805; 95% CI, 0.584–1.109;  Figure 3F ), whereas endometrioid carcinomas (RR, 1.759; 95% CI, 1.551–1.995;  Figure 3G ) and clear cell carcinomas (RR, 2.606; 95% CIs, 2.225–3.053;  Figure 3H ) were more common in EAOC than in non-EAOC. These findings were more definite in subgroup analyses based on study design, quality of study, assessment of endometriosis and adjustment for potential confounding factors except no difference in the risk of serous carcinoma in studies where endometriosis was assessed with histology (RR, 0.408; 95% CI, 0.064–2.585;  Table 4 ).\n\nRecent studies suggest the possibility that genetic and nongenetic factors potentially contribute to the neoplastic progression of endometriosis, where the following five typical factors have been suggested to increase ovarian cancer risk by endometriosis: atypical endometriosis as a precursor of malignancy; genetic alteration in endometrial tissues; heme or free iron-induced oxidative stress; chronic inflammation; and steroid hormones including oestrogen and progesterone ( Del Carmen  et al , 2003 ;  Somigliana  et al , 2006 ;  Mandai  et al , 2009 ;  Kokcu, 2011 ;  Munksgaard and Blaakaer, 2012 ). For supporting the possibility of the malignant transformation of endometriosis, a recent pooled analysis has reported that the association of a history of ES with an increased risk of ovarian cancer may be clear, in particular, for low-grade serous, endometrioid and clear cell carcinoma, showing the consistency with laboratory evidence of related molecular and genetic alterations ( Pearce  et al , 2012 ).\nHowever, relevant reviews and pooled analyses have some limitations as follows: first, some case–control or cohort studies include only women with moderate or severe endometriosis that thereby can overestimate ovarian cancer risk. Second, definite information about well-known preventive factors of ovarian cancer such as duration of hormonal agent use, infertility and gynaecologic treatment are missing, although potential confounding factors have been reported to be controlled. Third, hospital- or community-based control groups and interview or self-report without medical records can act as selection or recall bias. Furthermore, different regimen of adjuvant chemotherapy after surgery can also be a limitation for comparing prognosis between EAOC and non-EAOC patients.\nAlthough the meta-analysis could not overcome these limitations completely, and most of include studies did not show the definite relation between ovarian cancer and endometriosis in spite of the suggested criteria for the diagnosis of ovarian cancer arising from endometriosis ( Sampson, 1925 ), it has major advantages as follows. We included the greatest number of relevant studies, and performed subgroup analyses according to study design, assessment of endometriosis, histology, FIGO stage, quality of study and adjustment for potential confounding factors to minimise bias. As a result, we obtained the following meaningful results in the meta-analysis.\nFirst, endometriosis increased ovarian cancer risk by ∼27% in case–control or two-arm cohort studies, and ∼80% in single-arm cohort studies. These findings are consistent with the results from previous reviews ( Sayasneh  et al , 2011 ;  Pearce  et al , 2012 ;  Heidemann  et al , 2014 ). Furthermore, these findings were similar in subgroup analyses to minimise bias, suggesting the epidemiologic evidence than endometriosis may be strongly associated with the increased risk of ovarian cancer. Second, early-stage disease, low-grade disease and endometrioid and clear cell carcinomas were strongly associated with EAOC and non-EAOC. Recently, a dualistic model for ovarian carcinogenesis has been suggested. Type I ovarian tumours are clinically indolent and usually present with low-grade carcinoma, showing  KRAS ,  BRAF ,  ERBB2 ,  PTEN ,  CTNNB1  and  PIK3-CA  mutations. These mutations exhibit the continuum of tumour progression between benign cystic neoplasms and the corresponding carcinomas such as endometrioid, clear cell and low-grade serous carcinomas, often through precursor lesions such as ES and borderline tumours ( Cho and Shih, 2009 ). On the other hand, type II ovarian tumours are highly aggressive and almost always present in advanced-stage disease, showing  TP53  mutation ( Bast  et al , 2009 ). Our meta-analytic results show the epidemiologic evidence that EAOC may have favourable characteristics of type I ovarian tumours. Furthermore, we found that the risk of EAOC increased in relatively young or nulliparous women, and this also suggests the epidemiologic evidence that the retrograde menstruation and activation of oncogenic pathways in eutopic endometrium may permit endometrial tissues to implant and invade on ovarian and peritoneal surfaces that leads to type I ovarian tumours ( Bulun, 2009 ).\nIn particular, a specific histology such as endometrioid or clear cell carcinoma supports the hypothetical pathogenesis of malignant transformation of endometriosis. In the hypothesis, the carcinogenic process in an oestrogen-rich, progesterone-poor hormonal environment primarily gives rise to endometrioid carcinoma ( Ness, 2003 ;  Mandai  et al , 2009 ). Moreover, a high-level of heme and free iron induces persistent oxidative stress that results in stress-resistant type such as clear cell carcinoma ( Mandai  et al , 2009 ). Furthermore, genetic mutations in  hepatocyte nuclear factor-1β  ( HNF-1β ) and  ARID1A  are known to be related with the onset of endometrioid or clear cell carcinoma from endometriosis ( Kato  et al , 2006 ;  Wiegand  et al , 2010 ). Nevertheless, we found a relatively low incidence of serous carcinoma in EAOC, and no impact of endometriosis on the risk of mucinous carcinoma.\nOn the other hand, the recent pooled analysis showed that endometriosis was not associated with the risk of mucinous carcinoma of the ovary (odd ratio (OR), 1.02; 95% CI, 0.69–1.50), whereas it increased the risk of low-grade serous carcinoma (OR, 2.11; 95% CI, 1.39–3.20) and did not affect the risk of high-grade serous carcinoma in the recent pooled analysis (OR, 1.13; 95% CI, 0.97–1.32) ( Pearce  et al , 2012 ). These conflicting results on the meta-analysis are because of a number of included studies, study design, quality of study and adjustment for potential confounding factors. When compared with the previous pooled analysis using 13 case–control studies, more studies (15 case–control and two cohort studies) for histology were included in this meta-analysis, and all results were obtained in both crude and subgroup analyses for minimising bias that made the results more persuasive. Furthermore, the result that endometriosis was associated with a lower risk of serous adenocarcinoma is reasonable in this meta-analysis when we considered that endometriosis was related with the increased risk of endometrioid and clear cell carcinomas, and mucinous carcinoma was not associated with endometriosis.\nThird, endometriosis did not affect prognosis of ovarian cancer. Although there was no difference in progression-free survival between the two groups, EAOC was associated with better overall survival than non-EAOC in crude analyses. These findings explain why previous studies have suggested better prognosis of EAOC with favourable characteristics including early-stage disease, low-grade disease and a specific histology up to now ( Erzen  et al , 2001 ). However, there were no differences in both progression-free survival and overall survival between the two groups in subgroup analyses based on histology, assessment of endometriosis, disease status and adjustment for potential confounding factors. These findings mean that endometriosis may not affect prognosis of ovarian cancer in spite of favourable characteristics of type I ovarian tumours, and previous studies have also demonstrated no benefit of survival in patients with EAOC when controlled with FIGO stage ( McMeekin  et al , 1995 ;  Komiyama  et al , 1999 ;  Kumar  et al , 2011 ). Moreover, the impact of endometriosis on probability of optimal debulking surgery, the most important prognostic factor in ovarian cancer, was not determined in the meta-analysis, suggesting no benefit of survival in patients with EAOC indirectly.\nIn conclusion, endometriosis is strongly associated with the increased risk of ovarian cancer risk. Furthermore, favourable factors of EAOC including early-stage disease, low-grade disease and a specific histology such as endometrioid or clear cell carcinoma belong to type I ovarian tumours showing less invasiveness and slow growth, which supports the epidemiologic evidence linking endometriosis to a precursor lesion of ovarian cancer. In spite of favourable characteristics of EAOC, there was no difference in prognosis between EAOC and non-EAOC when adjusted with stage and a specific histology that suggests that endometriosis may not affect the progression after the onset of ovarian cancer.\nThese results from this meta-analysis suggest the possibility of no difference in the efficacy of primary standard treatment including cytoreductive surgery and adjuvant taxane- and platinum-based chemotherapy between EAOC and non-EAOC. Thus, prospective clinical trials are required to determine the surgical extent to remove endometriosis as well as tumour, and the optimal regimen and cycles of adjuvant chemotherapy based on clinicopathologic characteristics of EAOC for improving its prognosis.","source_license":"CC0","license_restricted":false}