Survival in ovarian cancer patients by histology and family history

Abstract

Introduction. Earlier studies suggest that histology has no prognostic significance in patients with invasive ovarian tumors. Studies about the effect of family history on survival have given conflicting results, which we try to clarify in this study. As an additional question, we examined whether family members share survival experience. Methods. We used the nation-wide Swedish Family-Cancer Database to estimate hazard ratios (HRs) for cause-specific and overall survival in ovarian cancer patients by histology and family history. HRs show the probability of death in the study group compared to the reference group. Results. A total of 6 049 ovarian cancer patients with specific histologies were retrieved from our Database from years 1993 to 1999. Compared to women with epithelial ovarian cancer, women with borderline epithelial tumors had the best survival (HR 0.02 and 0.14 for cause-specific and overall survival). Good survival was also noted for patients with sex cord-stromal tumors and germ cell tumors. Among specific subtypes of epithelial ovarian cancers, good survival was noted for women with clear cell and endometrioid carcinomas and mucinous cystadenocarcinoma. The study covered 80 mother-daughter pairs with a family history. Patients with a family history had a poorer survival than sporadic cases in both maternal and offspring generations. When the survival was analyzed according to the probands’ length of survival, there was a non-significant concordance of prognosis. Conclusion. Our data showed that histology and family history are prognostic factors for ovarian tumors. Patients with a family history had a more aggressive course than the sporadic cases.

Ovarian cancers are a heterogeneous group of malignancies that arise from the various cell types that comprise the organ, including epithelial, germ cell and sex cord-stromal ovarian cancers [1]. They impose a great clinical challenge because most patients are asymptomatic until the disease has metastasized [2]. Epithelial ovarian cancer is the leading cause of death from gynaecological malignancies [2], [3]. Borderline epithelial ovarian tumors are a subset of epithelial ovarian tumors with a very good prognosis, and they comprise approximately 15% of all epithelial ovarian tumors. The mean age of occurrence is approximately 10 years younger than that of women with frankly malignant ovarian cancer. These tumors are no longer considered as malignant with the implementation of ICD-O-3, and they were recorded as in situ tumor in the Swedish Cancer Registry [4]. Earlier studies have shown that clinicopathologically related factors, such as tumor stage, histological grade, and residual disease after primary surgery, have prognostic significances [5], [6]. Data about the effect of histologies on survival are largely limited [5], [7], [8]. However, a few studies have found that the tumor responsiveness to chemotherapy, the main treatment modality for ovarian tumors, differed by histologic subtypes [9], suggesting survival differences by histology. Thus a comprehensive study on specific histologies is needed.

About 4% of the patients with ovarian cancers report a family history in the first degree relatives in Sweden [10], and they are usually diagnosed at younger ages compared to patients without a family history [11]. Unselected patients with invasive ovarian cancer have an 11% chance of carrying a germline mutation of BRCA1 or BRCA2, but for those with a family history the risk of germline mutations in these genes increases to 19% [12]. However there is a large variation in the reported frequency of BRCA1 and BRCA2 carriers among ovarian cancer patients [13]. Although survival in familial ovarian cancer has been examined in some previous studies [14–16], the results have been inconsistent, mainly because of the limited numbers of familial cases.

We posit that survival in ovarian cancer patients could differ by histology and family history. To examine these issues we used the nation-wide Swedish Family-Cancer Database to estimate hazard ratios in ovarian cancer patients by specific histologies and by family history. As an additional question, we examined whether family members share survival experience, i.e. good or poor survival. Earlier study from this Database has shown a familial survival in breast cancer [17]. To prove this hypothesis, the survival time in probands was stratified and analyzed separately.

Patients and methods

The Family-Cancer Database was created by linking information from the Multigeneration Register, national censuses, Swedish Cancer Registry and death notifications [18]. Data on family relationships were obtained from the Multigeneration Register, in which children born in Sweden in 1932 and later were registered with their biological parents as families. Thus, the second generation (daughters) had reached an age of 72 years; the ages in the first generation (mothers) were not limited. Information retrieved from the various registers was linked at the individual level via the national 10-digit civic registration number assigned to each person in Sweden for her lifetime. Prior to inclusion in the Database (MigMed2), civic registration numbers were replaced by serial numbers to ensure anonymity of all individuals. The Swedish Cancer Registry is based on compulsory reports of diagnosed cases provided by physicians (clinical report) and by pathologists or cytologists (pathology report); the coverage of the cancer registration is currently considered to be close to 100%. The Database was updated in 2006 to include cancer cases from years 1958 to 2004 [18].

Only cases classified as primary neoplasms of the ovary (International Classification of Diseases, ICD, 7th revision, code 175) were considered in the present study. Because histology has been recorded according to the College of American Pathologists Systemized Nomenclature of Medicine (SNOMED) since 1993, the analyses of histology-specific survival included only women (mothers and daughters) diagnosed between years 1993 and 1999. The Swedish Cancer Register also provides a three-digit code (PAD code) with information on histological subtype, which was available from 1958 onward. A total of 6 049 ovarian tumors were noted, including 4 494 epithelial, 101 germ cell, 286 sex cord-stromal and 1 168 borderline epithelial tumors. However, for specific histologies, only the common subtypes were examined. The analyses of survival in patients with a family history included only invasive cancer, and they were done separately in daughters and mothers. To decrease the possible confounding by incidence and survival changes, cancer cases in daughters were included only women diagnosed between years 1990 and 1999, and cases in mothers included only women diagnosed before year 1990. A total of 3 105 daughters were diagnosed with ovarian cancer, 80 of whom had a mother affected before year 1990. For maternal ovarian cancers, 13 821 cases were noted before year 1990. Follow-up for survival began at diagnosis and terminated at death, emigration or the closing date of the study, 31 December, 2004, whichever came first. In order to adjust the numbers of fatal events between the subgroups, the probands' survival time was divided differently between maternal and offspring familial cases. For daughters with a family history, the mothers’ survival time was divided at <12, 12–59 and more than 59 months so that the fatal events in the three subgroups largely matched. For familial cases in mothers, we tried to separate daughters’ survival time using the same definition; the fatal events were few for these daughters who survived less than 12 months, so the daughters’ survival time was divided at 60 months to match the numbers of fatal events in the two subgroups.

Cox's proportional hazard regression models were used to estimate the hazard ratios (HRs) for disease-specific and overall survival. HR indicates the relative risk of dyeing in the defined period compared to a reference group; when it is below 1.0, the risk is lower than in the reference group. HRs were adjusted for age at diagnosis, period, socioeconomic status and diagnosis region. For the analysis of disease-specific survival, deaths as a result of ovarian cancer (recorded as underlying cause of death) were the primary end point; deaths due to other causes were censored. For the analysis of overall survival, deaths due to any cause were the end point. The proportional hazard assumption for the covariates was tested by Schoenfeld residuals and by plotting the log of the negative log of the survival function versus the log of time; covariates were stratified in the models if they did not meet the assumption. All statistical analyses were performed using SAS software (version 9.1; SAS Institute, Cary, NC) [19].

Results

Survival in ovarian cancer patients was analyzed by histology (Table I). Compared to women with epithelial ovarian cancer (HR 1.0), women with borderline epithelial tumors had the best survival (HR 0.02 and 0.14 for cause-specific and overall survival). Good survival was also noted for patients with sex cord-stromal tumors and germ cell tumors. For specific histologies in epithelial cancer, we used papillary serous cystadenocarcinoma, the largest entity in epithelial cancer, as a reference (HR 1.0). The best survival was noted for women with clear cell and endometrioid carcinomas and mucinous cystadenocarcinoma. For patients with granulose cell carcinoma and mullerian tumor, both of them belonging to sex cord-stromal tumors, they had different survival experience, the former favorable (HR 0.11) and the latter very unfavorable (HR 1.57). For specific borderline epithelial tumors, survival was similar among patients with serous, papillary serous and mucinous cystadenoma. The median year of diagnosis ranged from 1996 to 1999; the median survival times ranged from 12 to 94 months.

Table I.  Hazard ratios for ovarian cancer patients diagnosed between 1993–1999 according to histologies.

				Cause-specific suurvival					Overall survival
Histological types	Cases	Median year of diagnosis	Median survival time (months)	Deaths	Percent	HR	95%	CI	Deaths	Percent	HR	95%	CI
Epithelial ovarian cancer, reference*	4494	1996	39	2580	57.4	1.00			3067	68.2	1.00
Papillary serous cystadenocarcinoma	1407	1996	34	940	66.8	1.00			1062	75.5	1.00
Undifferentiated carcinoma	154	1996	31	98	63.6	0.96	0.78	1.18	114	74.0	0.96	0.79	1.17
Clear cell carcinoma	207	1996	66	91	44.0	0.59	0.47	0.73	104	50.2	0.58	0.48	0.71
Endometrioid carcinoma	607	1996	65	268	44.2	0.55	0.48	0.63	329	54.2	0.58	0.51	0.66
Serous cystadenocarcinoma	469	1996	40	286	61.0	0.88	0.77	1.01	318	67.8	0.86	0.76	0.97
Mucinous cystadenocarcinoma	473	1996	70	171	36.2	0.49	0.41	0.57	232	49.0	0.56	0.48	0.64
Others	1177	1996		726	61.7	1.04	0.94	1.15	908	77.1	1.14	1.04	1.24
Sex cord-stromal tumors	286	1996	77	80	28.0	0.42	0.34	0.53	121	42.3	0.53	0.44	0.63
Granulosa cell carcinoma	145	1996	94	13	9.0	0.11	0.07	0.19	30	20.7	0.21	0.15	0.30
Mullerian tumor	55	1997	12	36	65.5	1.57	1.13	2.18	47	85.5	1.70	1.27	2.26
Others	86	1995		31	36.0	0.59	0.41	0.84	44	51.2	0.70	0.52	0.94
Germ cell tumors	101	1996	84	21	20.8	0.42	0.27	0.65	24	23.8	0.41	0.27	0.61
Teratoma	41	1996	90	5	12.2	0.22	0.09	0.53	6	14.6	0.22	0.10	0.49
Others	60	1996		16	26.7	0.52	0.32	0.85	18	30.0	0.49	0.30	0.77
Borderline epithelial tumors	1168	1996	92	23	2.0	0.02	0.02	0.04	159	13.6	0.14	0.12	0.16
Serous cystadenoma	348	1995	91	2	0.6	0.01	0.00	0.03	43	12.4	0.12	0.09	0.16
Papillary serous cystadenoma	269	1997	92	4	1.5	0.02	0.01	0.05	36	13.4	0.13	0.10	0.19
Mucinous cystadenoma	541	1996	93	17	3.1	0.04	0.02	0.06	77	14.2	0.13	0.11	0.17
Others	10	1996		0	0.0				3	30.0	0.32	0.08	1.28

HR, hazard ratios, adjusted for age, period, region and socioeconomic status. Percent, percent of deaths. Bold type, 95%CI does not include 1.00.

*, for subtypes of epithelial ovarian cancer, using papillary serous cystadenocarcinoma as reference.

Survival in daughters diagnosed between years 1990 and 1999 is shown in Table II. Compared to daughters with sporadic ovarian cancer, patients with a family history had a poor cause-specific (HR 1.49) and overall survival (HR 1.38). When survival in daughters was analyzed according the mothers’ length of survival (mothers as probands), there was a concordance of good survival: the longer the survival time of the mother, the better the survival of the daughter (p for trend 0.04 for overall survival). In this table, the median year of diagnosis was 1995; the median survival times ranged from 33 to 67 months.

Table II.  Hazard ratios for cause-specific and overall survival in daughters diagnosed with ovarian cancer in the 1990s whose mothers were diagnosed before 1990.

				Cause-specific survival					Overall survival
Parameters	Cases	Median year of diagnosis	Median surviva1 time (months)	Deaths	Percent	HR	95%	CI	Deaths	Percent	HR	95%	CI
Sporadic cases, reference	3025	1995	67	1392	46.0	1.00			1598	52.8	1.00
Mothers with ovarian cancer before 1990	80	1995	51	50	62.5	1.49	1.12	1.98	53	66.3	1.38	1.05	1.81
Mothers' survival time (month)
< 12, reference	27	1994	33	20	74.1	1.00			22	81.5	1.00
12-59	27	1995	51	17	63.0	0.93	0.43	2.03	18	66.7	0.84	0.39	1.79
> = 60	26	1995	67	13	50.0	0.78	0.32	1.89	13	50.0	0.65	0.27	1.56
P value for trend test							0.08					0.04

HR, hazard ratios, adjusted for age, period, region and socioeconomic status. Percent, percent of deaths.

Bold type, 95%CI does not include 1.00.

Reversing the analysis and deriving HRs for mothers who were diagnosed before year 1990 (daughters as probands) showed that HRs were significantly higher for patients with a family history compared to those with sporadic disease (Table III). The HR for mothers showed a borderline decrease if their daughters had survived more than 60 months, indicating correlation for survival experience. Median survival times ranged from 14 to 30 months. Because histologies were associated with survival in ovarian cancer patients, we tried to adjust them for the analyses of survival by family history. However, because histological classification by SNOMED was available only from 1993 onward in the Swedish Cancer Registry, we could not use SNOMED for adjustment. Instead we adjusted the data using the PAD code, which was available from year 1961 to1999, and the results did not change.

Table III.  Hazard ratios for cause-specific and overall survival in mothers diagnosed with ovarian cancer before 1990 whose daughters were diagnosed in the 1990s.

				Cause-specifie survival					Overall surviva1
Parameters	Cases	Median year of diagnosis	Median survival time (months)	Deaths	Percent	HR	95%	CI	Deaths	Percent	HR	95%	CI
Sporadic cases, reference	13741	1978	30	8290	60.3	1.00			11171	81.3	1.00
Daughters with ovarian cancers in the 1990s	80	1977	18	60	75.0	1.37	1.06	1.76	75	93.8	1.34	1.07	1.68
Daughters' survival time (month)
< 60, reference	45	1976	14	38	84.4	1.00			42	93.3	1.00
> = 60	35	1977	30	22	62.9	0.54	0.28	1.01	33	94.3	0.62	0.35	1.12

HR, hazard ratios, adjusted for age, period, region and socioeconomic status. Percent, percent of deaths.

Bold type, 95%CI does not include 1.00.

Discussion

In the present survival study, histology is shown to be a prognostic factor for ovarian tumors, which is contrary to the earlier observations on invasive tumors [3], [20], [21]. A similar observation was noted in the recent SEER publication from USA [22], but this analyses did not adjust for possible confounding factors and it showed only the 5-year relative survival rates. The data quality in this study is high because it is based on the nation-wide databases in Sweden with registered family structures, medically verified cancers, registered underlying causes of death and completed follow-up of cancer patients, guaranteeing reliable data.

Because of the low potential of spreading and invading other tissues, women with borderline epithelial ovarian tumors presented the best survival among all histologies. It should be noted that different types of borderline cystadenoma had similar survival ability. Sex cord-stromal neoplasms are composed of various cell types derived from gonadal stroma and sex cords [23]. Unlike patients with epithelial cancer, in which 75% are considered to be diagnosed at stage III or IV [2], about 70% patients with these tumors are diagnosed at stage I because they usually have specific symptoms, such as infrequent menstrual periods and vaginal bleeding in postmenopausal women. However, granulosa cell carcinoma and mullerian tumor, the common subtypes of sex cord-stromal neoplasms, showed totally different survivals. Because of the limited number of mullerian tumors, studies on its etiology and prognosis have been limited, but its poor survival could be explained by its specific feature of poor response to the treatment and late presentation at diagnosis [24]. Germ cell neoplasms are thought to be derived from primitive germ cells of the embryonic gonad [25]. Although these tumors can occur in women at any age, the peak incidence is seen during the early 20s, which may explain the observed survival advantage compared to epithelial cancer. For specific epithelial ovarian cancers, women with papillary serous cystadenocarcinoma, the commonest subtype, had the worst survival, which was in agreement with an earlier study [20]. The distinction between serous cystadenocarcinoma and papillary serous cystadenocarcinoma is based on histological verification of whether papillary features are noted in the diagnostic description. Both serous and papillary serous cystadenocarcinoma had adverse survival compared to non-serous carcinoma, which may be related to the later diagnosis [20]. Stage at diagnosis and treatment, important prognostic factors for ovarian tumors, are not available in our Database, which is a limitation in this study. However, stage at diagnosis may be one of the biological characteristics for specific histologies, i.e., different histologies may be diagnosed at different stages. Therefore, matching for stage at diagnosis may reduce the true survival difference between histologic subtypes.

Another finding in this study is that patients with a family history had a more aggressive course than women with sporadic disease, which is in agreement with an earlier study from UK [14]. The results were consistent in showing the survival difference for both maternal and offspring generations. The worse prognosis among familial cases compared to sporadic cases may reflect the real biological differences. Surveillance bias may confound the results, for example, women with a familial history may have an increased surveillance compared to sporadic cases, because of more active search for health service. However, such bias cannot explain the high risk for mothers who were diagnosed two decades before their daughters. In this study we could also show a familial concordance in survival experiences between family members: family members appeared to share either good or bad prognoses. HRs were lower for cause-specific and overall survival when the probands had a better survival. The results were consistent in both maternal and offspring generations, which further supported this observation. However, because the numbers of familial cases were limited, these associations were of borderline significance. Further studies are needed to verify these findings.

In summary, our data showed that histology and family history are prognostic factors for ovarian tumors, which may have important implications for the management and prevention of ovarian tumors. Familial ovarian cancers had a more aggressive course than sporadic cases. The data suggest familial concordance in either good or poor survival among family members.

Acknowledgements

Supported by Deutsche Krebshilfe, the Swedish Cancer Society, the EU, LSHC-CT-2004-503465 and the Swedish Council for Working Life and Social Research. The Family-Cancer Database was created by linking registers maintained at Statistics Sweden and the Swedish Cancer Registry.