Ovarian cancer survival by tumor dominance, a surrogate for site of origin.

Despite extensive study, the etiology of epithelial ovarian cancer is not fully understood and minimal progress has been made in prevention and treatment, leading to little change in mortality and morbidity associated with this disease [ 1 – 3 ]. Prognosis is largely based on stage at detection, volume of disease, optimal debulking, histologic subtype, grade and response to chemotherapy [ 4 , 5 ]. Along with efforts aimed at early detection, strategies that impact morbidity and mortality are warranted. Ovarian tumors are traditionally thought to arise from the ovarian surface epithelium or related ovarian inclusion cysts, and are classified based on pathologic features such as histology, grade and stage [ 2 ]. Recent pathology studies suggest that a proportion of high-grade serous tumors may actually originate in the fallopian tube fimbriae [ 6 – 10 ], which along with risk-factor diversity [ 11 ] suggests that ovarian cancer is not a homogeneous disease [ 12 , 6 ]. Detailed sectioning of the fallopian tube can in some cases identify tumors that are of tubal origin and have fimbrial involvement; however, this approach is labor intensive, applies essentially to localized disease, and cannot be used in cases identified after surgery [ 13 ]. Tumor “dominance” represents an alternative method for attempting to assign a greater probability of a tubal versus ovarian origin based on whether the largest tumor mass is on one of the ovaries [ 6 ]. In this model, a dominant ovarian mass would imply the tumor has a greater likelihood of having started at that site. In contrast, absence of a dominant ovarian mass would be presumed to occur if the tumor initiated elsewhere and randomly spread throughout the peritoneal cavity. With this model, tumors that are described as originating entirely in one ovary or tumors where one ovary exceeds the other by at least two-fold in size are classified as “dominant” and are more likely to have an ovarian site of origin. Tumors that are comprised of microscopic surface deposits, spread equally across peritoneal cavity, or have only tumor foci on ovaries are classified as “non-dominant” and are presumed to favor a fallopian tube origin [ 6 ]. Kotsopoulos et al. , recently reported differences in risk factor associations by site of origin [ 14 ]. Specifically, they found that reproductive factors were more important for dominant vs. non-dominant tumors, suggestive of potentially independent carcinogenic pathways. There are no studies, to our knowledge, that have evaluated whether tumor dominance is associated with survival. Thus, the goal of the current study was to explore the association between tumor dominance and survival following a diagnosis of epithelial ovarian cancer.

Of the 1,386 women patients included in the present analysis, we observed 1,160 (84%) with dominant and 226 (16%) non-dominant tumors ( Table 1 ). The dominant tumors were primarily of serous (46%) and endometrioid histology (26%), while the non-dominant tumors were mostly comprised of serous subtype (87%)( P < 0.0001). Approximately half of the dominant tumors (57%) were stage III/IV vs. 92% of the non-dominant tumors ( P < 0.0001). Compared to women with dominant tumors, those with non-dominant tumors were significantly more likely to carry BRCA1/2 mutations (23% vs. 10%; P < 0.0001), have first degree relatives with history of ovarian or breast cancer (31% vs. 21%; P = 0.003), to be parous (88% vs. 81%; P = 0.02), have higher mean parity (2.8 vs. 2.4; P = 0.008) more likely to have ever used intra-uterine devices (IUD) (21% vs 15%; P = 0.03) and more likely to have used estrogen hormone replacement therapy (30% vs. 22%) P =0.01). In our study sample, significantly more women died by the end of follow-up among women with non-dominant vs. dominant tumors (66% vs. 36%; P < 0.0001). Among the entire study population, women with non-dominant tumors experienced significantly worse survival compared to women with dominant tumors ( P < 0.0001) ( Figure 1 ). The findings were similar in the analyses stratified by serous ( P = 0.0002) ( Figure 2 ) vs. non-serous tumors ( P = 0.0001) ( Figure 3 ). The HRs and 95% CIs for ovarian cancer-specific mortality associated with tumor dominance are presented in Table 2 . In the reference model (only adjusted for age at diagnosis), a non-dominant tumor was associated with a significant increased risk of death (HR = 2.41; 95%CI 1.98–2.95) compared to non-dominant tumors. The results were similar, but attenuated, in the models additionally adjusted for histology and stage (Model 1 HR = 1.29; 95%CI 1.05–1.59) and subsequently for BRCA1/2 mutation status and past use of estrogen hormone replacement therapy (Model 2 HR = 1.28; 95%CI 1.02–1.60). In the analyses limited to serous subtype, non-dominant tumors were associated with increased mortality risk when compared to dominant tumors (reference model HR = 1.55; 95%CI 1.24–1.93 and again, attenuated in model 1 (HR = 1.28; 95%CI 1.03–1.61) and model 2 (HR = 1.28; 95%CI 1.01–1.63) ( Table 2 ). Among the non-serous subtypes, non-dominant tumor was also associated with increased risk of death (reference model HR = 3.06; 95%CI 1.75–5.35), although attenuated in models 1 and 2 (model 1 HR = 1.49; 95%CI 0.85–2.62 and model 2 HR = 1.40; 95%CI 0.77–2.53). Finally, we examined mortality risks stratified by early (i.e., stage I/II) vs. late stage (i.e., stage III/IV) tumor ( Table 2 ). Among early stage tumors, in the reference model, non-dominance was associated with significant increased risk of death (HR = 3.93; 95%CI 1.40–11.05) but was attenuated in model 1 (HR = 2.39; 95%CI 0.82–6.92) and model 2 (HR = 0.91; 95%CI 0.20–4.12). Among late stage tumors, women with non-dominant tumors had significant increased risk of death (reference model HR = 1.45, 95%CI 1.17–1.78), which was attenuated but still significant in model 1 (HR = 1.28; 95%CI 1.04–1.58) and model 2 (HR = 1.30; 95%CI 1.04–1.63).

The overall goal of the current study was to evaluate whether survival following diagnosis of epithelial ovarian cancer was associated with dominance, a potential tool for assigning probable for tumor site of origin. In this study of 1,386 women, women with non-dominant tumors had significant 28% increased mortality risks with adjustment for important covariates. These findings add to the growing literature that ovarian cancer is not a homogenous disease and that the fallopian tube represents a potential source of serous cancers [ 1 , 20 , 21 ]. We observed significant differences in specific tumor and clinical characteristics by dominance. For example, non-dominant tumors were more likely to be serous, stage III/IV, occur in women with inherited BRCA1 or BRCA2 mutations, as well as family histories of breast or ovarian cancer, and were associated with greater parity and with use of estrogen hormone replacement therapy. In contrast, dominant tumors were more evenly split between the serous and endometrioid subtypes and had a more even distribution of stage I/II and stage III/IV tumors. This is comparable to an earlier report which found that non-dominant tumors were more likely to be serous, higher grade, and later stage while dominant tumors were more likely to be mucinous, endometrioid or clear cell [ 14 ]. Kotsopoulos et al. , also reported that specific factors were associated with the risk of ovarian-originating tumors (i.e., dominant tumors) but not tumors originating in the fallopian tubes (i.e., non-dominant): tubal ligation and two or more births decreased risk while endometriosis increased risk. Collectively, these two studies have provided preliminary evidence that risk factor profiles and survival may differ by tumor site of origin and are suggestive of different developmental pathways of ovarian tumors. Considering combinations of case characteristics may be important for understanding predictors of survival. Evidence from pathology studies suggests that as many as half of high-grade serous tumors may originate in the distal fallopian tube [ 6 – 9 , 22 ]. A step-wise progression to the development of invasive cancer in the fallopian tube has been supported from studies carefully examining the distal fallopian tubes of BRCA mutation carriers (as well as non-carriers) undergoing risk-reducing bilateral salpingo-oophrectomy [ 13 , 23 , 12 ]. This model includes an early or "latent" precursor lesion (i.e., p53 signature) in the fallopian tube that occurs in at least one half healthy women yet contains conserved p53 mutations and is located in the distal fallopian tube, similar to tubal intraepithelial carcinomas (TICs), which are considered immediate precursors to invasive or metastatic serous cancer [ 24 , 13 , 25 ]. The frequency of the p53 signature is similar in women with and without inherited BRCA mutations and appears to be an early pathogenic event in the fallopian tube that is largely independent of BRCA mutation status and like most precursors, rarely progresses to malignancy [ 26 , 8 , 27 ]. We observed that a significantly higher proportion of women with non-dominant vs. dominant tumors had BRCA mutations (23% vs. 10%; P < 0.0001) which is consistent with studies indicating the distal fallopian tube as the site of origin of high-grade serous cancers among BRCA mutation carriers [ 28 , 23 , 29 , 25 , 13 ]. Findings were similar, although not significant likely due to small numbers, when we limited this analysis to high-grade serous cancers, raising the question of whether dominant serous cancers have different pathogenic pathways compared to their non-dominant counterparts (data not shown)[ 30 , 31 ]. Further investigation of the p53 signature-to-invasive carcinoma pathway may lead to a better understanding of how to prevent and treat epithelial ovarian and fallopian tube cancers [ 32 , 5 , 33 , 20 , 1 ]. Examining ovarian cancer by putative tumor site of origin may help to obtain more precise risk estimates for survival [ 5 ]. Within our study follow up period 66% of patients with non-dominant tumors had died, compared to 36% of patients with dominant tumors, and patients with non-dominant tumors had significantly shorter total study follow up time than those with dominant tumors (5.57 vs. 8.14 years). Accordingly, women are beginning to be offered bilateral salpingectomy at the time of pelvic surgery among those at average population risk of developing ovarian cancer, with the objective the prevention of aggressive subtypes of this disease [ 34 , 35 ]. Given the high lifetime risks associated with BRCA1 and BRCA2 mutations, such high-risk women are advised to undergo prophylactic bilateral salping-oophrectomy at age 35 or when childbearing is complete [ 23 , 36 ]. It has been put forth that salpingectomy with oophorectomy delayed until the time of natural menopause be offered to this high-risk population as a way to minimize the impact of early surgical menopause on overall morbidity [ 37 ]. However, this topic remains under scrutiny given the lack of evidence examining the clinical benefit of such a procedure [ 38 , 35 ]. The current study has several strengths and weaknesses. The primary strength is its novelty, in that it represents, to our knowledge, the first study to investigate in greater depth the relationship between tumor site of origin and survival using tumor dominance as proxy. Furthermore, our study population was representative of all epithelial ovarian cancer cases diagnosed in Ontario during the study recruitment period and we were able to adjust for important covariates. Finally, one study author classified tumor dominance from all of the pathology reports, which maximized consistency of the determination. Our primary limitation was that the approach to classify tumor site of origin according to dominance has low sensitivity and low specificity. Prior studies using the classification of Roh et al. , reported that 33%–40% of tumors were non-dominant [ 14 ] and 69% in a sample that was limited to serous carcinoma cases [ 6 ]. In contrast, we found that 16% of all of the tumors and 27% of the serous subtypes were classified as non-dominant. One reason for this disparity may be inconsistencies in the quality of pathology reports in the current study. However, more fundamental explanations must be considered. First, the arbitrary assignment of "dominance" based on a single measurement (diameter) is highly unlikely to reliably distinguish the actual site of origin. Rather it provides an estimate. Second, some tumors arising at the distal tube could invade directly into the adjacent ovary, creating a "dominant" mass. Another drawback is that we did not have complete information about duration and type of chemotherapy treatment received or debulking status, two factors that have previously been shown to be significant predictors of survival [ 39 ]. In fact, it is possible that non-dominant tumors may be associated with worse survival because they are more difficult to debulk optimally than dominant tumors. In summary, we have demonstrated that the application of a tumor dominance model for site of origin, as defined by parameters of pathology reports, is associated with ovarian cancer-specific survival. Specifically, tumors putatively originating in the fallopian tubes were associated with significantly worse prognoses and support emerging evidence for a novel paradigm that classifies ovarian tumors into distinct entities based on pathologic and molecular evidence as well as site of origin [ 40 , 41 ]. Based on these findings, prevention strategies including salpingectomy in women at general population risk undergoing pelvic surgery may have likelihood to decrease incidence and mortality of the more lethal subtypes of the disease. Future epidemiologic studies of ovarian cancer focusing on risk factors, treatment and prognosis should consider tumor heterogeneity by conducting stratified analyses.

All Ontario, Canada residents diagnosed with invasive epithelial ovarian cancer from January 1995 through December 1999 or January 2002 through December 2004, were identified through the Ontario Cancer Registry (OCR) [ 15 , 16 ]. Briefly, investigators reviewed pathology records for each case, to determine subject eligibility and tumor histology. Patients were between 20 and 79 years of age at the time of diagnosis of a new primary epithelial ovarian tumor. Through a risk factor questionnaire, information on known or suspected ovarian cancer risk factors and demographic data were collected by standardized telephone script. Styrofoam-packed venipuncture kits with consent forms were mailed to subjects, who had blood samples drawn locally and, along with signed consent forms, returned to the study center by pre-paid courier. All participants were offered the option to receive their genetic testing results in the context of a counseling clinic with the study team or elsewhere in the province as previously described [ 15 ]. Patient medical records were reviewed to obtain information about clinical staging, treatments received, and treatment outcomes. More information about the study design and data collection approach can be found elsewhere [ 15 , 16 ]. Hospital medical records were requested covering a minimum of one year following diagnosis to ensure completion of primary surgery and chemotherapy. Information was extracted regarding clinical staging, treatments received, and response to treatment. A second chart review was undertaken five years after initial treatment, if necessary. The FIGO system was primarily used for staging, as this is the widely accepted practice in Ontario, but was substituted by TNM staging where the necessary information was not available. Survival status, date of death, and cause of death were determined from death certificates linked with OCR computerized records, as well as via chart reviews at local hospitals. The OCR contains information about cancer incidence and vital status in Ontario. More information and an evaluation of using OCR record linkage for estimates of death statistics is available [ 17 ] and the automated procedures appear to be more accurate than manual approaches. Based on available pathology reports, 1,423 women were eligible for inclusion in the current study. Patient pathology reports were reviewed to classify all cases by tumor dominance as previously described [ 6 ]. Dominant tumors were defined as those limited to one ovary or where one involved ovary exceeded the other in dimension by at least two-fold (n = 1,160) [ 6 ]. Non-dominant tumors were defined as the remainder (n = 226). This method of classification of site of origin was proposed by Roh et al. [ 6 ] and previously used by Kotsopoulos et al. [ 14 ]. Also, 454 tumors were classified as unknown, due to the lack of dimension information presented in the pathology reports. These subjects were excluded from our analyses, and the final study sample included data from 1,386 participants. Differences between clinical, demographic, and other characteristics were compared by tumor dominance, using the χ 2 test or t-test, as appropriate. The primary outcome was ovarian cancer-specific survival, defined as the duration of survival from time of diagnosis to date of death from ovarian cancer. Case survival was censored when death occurred from another cause, or on September 30, 2010, which was the most recent limit of available death-certificate information. We performed a left-truncated analysis to reduce the extent of survivorship bias present in our study population [ 18 , 16 ]. We employed Cox proportional hazards models to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) associated with tumor dominance. The reference model was adjusted for age at diagnosis (continuous). Multivariate modeling took into account both the plausibility of biological effects of each covariate as well as statistical evidence of confounding and used a 10% change in the parameter of interest as the criterion to select covariates into the final model [ 19 ]. Multivariate model 1 was adjusted for age at diagnosis (continuous), histologic subtype (serous, mucinous, endometrioid, other [clear cell, mixed histology, and epithelial not otherwise specified]), and stage (I, II, III, IV). Although none of the variables that differed significantly by dominance and survival as outlined in Table 1 confounded the main association by more than 10%, the authors included BRCA1/2 mutation status (carrier/non-carrier) and estrogen hormone replacement therapy use (ever/never) into model 2 based on the association of these exposures with ovarian cancer risk and survival. Presence of a first degree relative with a history of breast or ovarian cancer was not included in the model due to collinearity with BRCA1/2 carrier status. The proportional hazards assumption was examined for all variables included in the final model. None of the variables, except for BRCA1/2 mutation status, violated the proportionality assumption. Since BRCA1/2 mutation status is not a time-dependent variable, it was allowed to remain in the proportional hazards model despite some evidence of non-proportionality. We also stratified our analyses by histology (i.e., serous, non-serous) and stage, where stage was combined into early stage (i.e., stages I and II) and late stage (i.e., stages III and IV) tumors. All analyses were performed using SAS version 9.2 (SAS Institute, Cary, NC, USA). All P -values are two-sided.