Ovarian cancers have shown a slowly declining incidence in Sweden over the past 2 decades, and with an incidence of less than 20/100,000, they rank sixth among all female cancers in Sweden.1 Tumors are derived from 3 tissue types; the most common are adenocarcinomas from surface epithelium, while other less common ones are germ cell tumors and sex cord-stromal tumors.2 Epithelial tumors present in many histologic forms and they are often cystic, known as cystadenocarcinomas, with serous or mucinous secretions. Both malignant and benign tumors present in these histologic forms. The low-malignant forms are often called borderline tumors, and these are classified as in situ tumors in the Swedish Cancer Registry. In contrast to invasive ovarian cancers, notification of the in situ cases has increased in recent years, when they have amounted to one-third the number of invasive cases. Ovarian cancer is a relatively early-onset cancer with a peak incidence in the 50s. International differences in incidence are large, with a gradient from low risk in developing countries to high risk in western countries.3 In Sweden, all immigrant groups have lower incidence than Swedes in the first generation, but the differences have disappeared in the second generation, indicating the involvement of environmental factors.4, 5, 6 However, apart from reproductive, hormonal and familial risk/protective factors, little is known about causes of ovarian cancer.7, 8, 9, 10, 11
Ovarian cancer is a manifestation in families with BRCA1/2 mutations and in hereditary nonpolyposis colorectal cancer (HNPCC).12, 13, 14, 15 Some 5–10% of ovarian cancers are often cited to be familial,7, 8 but the Canadian source of this figure states 2.9–6.9%.16 There are probably differences in the prevalence of familial ovarian cancer between populations because the familial percentage has been lower in other studies: less than 1% of cases were attributed to family history in 1 study, and in another study, 2.7% of the cases had an affected parent.17, 18 The prevalence of BRCA1/2 mutations, 11.7%, in unselected cases of invasive ovarian cancer has also been highest in a Canadian population.14 In population-based studies with medically verified diagnosis, the familial risk of invasive ovarian cancer has ranged between 2.0 and 3.0 and no increase has been observed for borderline tumors.19, 20, 21, 22, 23, 24, 25 The risk has been higher at young age.26 A metaanalysis of family studies found an unexplained difference between mothers of cases (risk 1.1) and daughters of cases (3.8).27 Most family studies on ovarian cancer have been based on interviews of cases and controls about cancer in their relatives, and 1 source of discrepancy may be the difficulty in accurately reporting intraabdominal cancers diagnosed far back in history.
We examine here familial risks in ovarian cancer using the nationwide Swedish Family-Cancer Database,28 which was updated in 2000 to include over 10 million individuals and over 1 million registered tumors, retrieved from the Swedish cancer registry from year 1958 to 1998. The cancer registry started to use the International Classification of Diseases (ICD) coding system ICD-O-2, with detailed histopathologic information since 1993. Even though the specific histologic classification only covers 6 years, a reasonable number of familial cases can be recorded because the familial cases accumulate towards the end of the follow-up period because of the aging second generation. The Database offers unique possibilities for reliable estimation of familial risks, because the data on family relationships and cancers were obtained from registered sources of practically complete coverage. We want to answer specifically the questions about familial risks by proband status and tumor histopathology in invasive and borderline ovarian cases and their links to breast cancer. Additionally, we determine the population-attributable fraction (PAF) for familial risk, which gives the preventable proportion of ovarian cancer if familial causes could be eliminated.
SUBJECTS AND METHODS
The Swedish Family-Cancer Database was initially created in the mid-1990s by linking an administrative family register on all Swedish families to the Swedish cancer registry.28, 29, 30 For each child there are data on both parents at the time of birth. Each person has been assigned a unique technical identification number (different from the national identification number, or “personal number”), allowing construction of families for example through the mother. The Database includes all persons born in Sweden after 1931 plus their biologic parents, totaling over 10.2 million individuals. It was updated in 2000 to include cancers from the nationwide Swedish cancer registry from years 1958 to 1998.28 The Database is organized in 3.2 million families, including parents and offspring.
The completeness of cancer registration in the 1970s has been estimated to be over 95% and is now considered to be close to 100%. The percentage of cytologically or histologically verified cases of ovarian cancers has been close to 100%.1 The Family-Cancer Database has an incomplete linkage from deceased offspring to parents, particularly among those offspring born between 1932 and 1940. Of a total of 7.0 million offspring, 216,000 have died by the end of followup (December 31, 1998). Parental information was missing from 15,600 dead offspring who had a diagnosis of cancer (9.9% of all offspring cancers); most of these were born in the 1930s. This deficit is unlikely to cause any appreciable effect on familial risk estimates in the present study but may cause a small uncertainty to the estimated sibling risks. The effects of missing parental links are discussed in more detail elsewhere.31
The Swedish cancer registry is based on compulsory notification of cases.1 A 4-digit anatomic site code according to the 7th revision of the ICD-7 has been used since 1958, together with a code for histologic type (WHO/HS/CANC/24.1 histology code). This histology code divided histology into main subgroups, such as adenocarcinoma and squamous cell carcinoma, and an overwhelming proportion of ovarian cancers were classified as adenocarcinoma. From 1993 onwards, ICD-O-2/ICD-10 site codes and histopathologic/morphology data according to the International Classification of Diseases for Oncology, the World Health Organization (WHO) and the Systematized Nomenclature of Medicine (SNOMED, http://snomed.org) have been used; we refer to this classification as “SNOMED or histopathology”. This classification was used to group epithelial and other (germ cell, sex cord, teratoma) tumors separately.
Family history information was collected on all first-degree relatives (parents, siblings and children) but only the mother-daughter, sister-sister and mother-2 daughters relationships were used in the present study. All tumor incidence rates were based on the data in the Family-Cancer Database. The risk of histopathology-specific ovarian cancer was calculated for daughters whose mother presented with any invasive ovarian cancer, and the risk was compared to the rate of histopathology-specific ovarian cancer among all daughters. Follow-up was started at birth, immigration or January 1, 1961, whichever came latest. Follow-up was terminated on diagnosis of first invasive or borderline cancer, death, emigration or the closing date of the study, December 31, 1998. Standardized incidence ratios (SIRs) were calculated as the ratio of observed (O)-to-expected (E) number of cases. The expected numbers were calculated from 5-year-age-, sex-, tumor type-, period- (10 year bands), parity- (6 groups), age at first birth- (5 groups), socioeconomic status- (6 groups) and residential area- (3 groups) specific standard incidence rates for all daughters. Confidence intervals (95% CI) were calculated assuming a Poisson distribution.32 We had no data on the use of oral contraceptives, but the adjustment for period should eliminate some of the possible effects of lack of information.
The population-attributable fraction (PAF) of cases with a family history of ovarian cancer was estimated as follows: proportion of cases with a family history x (familial SIR-1)/familial SIR, as defined by Miettinen33 and cited in a textbook as formula 16-21.34 For sisters, the SIRs were calculated only in families of 2 or more sisters, as given earlier.32
The Family-Cancer Database covered years 1961 to 1998 from the Swedish cancer registry and included 15,040 mothers and 4,135 daughters with ovarian cancer. The numbers for women with borderline tumors was 1,788 and 1,648, respectively. We first considered all ovarian cancers without histologic specification. Familial risks for ovarian cancer in daughters are shown in Table I by familial ovarian and breast cancer in mutually exclusive proband categories. The data were adjusted for age, period, parity, age at first birth, socio-economic index and region. Only invasive cancers were considered in probands. When a mother had ovarian cancer, the SIR was 2.68 and when a sister had ovarian cancer it was 2.94. These figures are not fully comparable because sisters were younger (all below 67 years) than mothers (any age). If a mother's age was limited to less than 67 years, the SIR for daughter's ovarian cancer was 3.21 (2.57–3.96). If both a mother and a sister were affected, the SIR was 24.03. Even borderline ovarian tumors were increased but the SIRs were not significant. Breast cancer in a mother increased the risk of ovarian cancer in the daughter to 1.21, and ovarian cancer in a mother and breast cancer in a sister resulted in an SIR of 7.52. No other comparison was significant, and the SIRs for borderline ovarian cancer from breast cancer probands were close to unity.
Table I. SIR for Familial Ovarian Cancer in Daughters1
Bold type, 95% CI does not include 1.00. Period of followup: 1961–1998.
SIR adjusted for age, period, parity, age at first birth, SEI, residential region.
Mother and sister
Mother and sister
Mother and sister
The age-specific incidence of ovarian cancer is shown in Figure 1, for mother and sister history of invasive ovarian cancer and for those lacking a family history. All the data points for familial invasive cases are well above the curve for “no history,” and sister and mother history appear similar, considering the numbers of cases at each data point. The divergence of the familial curves at the highest age is probably due to the low number of cases (3 for mother and 2 sister history). The age-specific SIR for family history is shown in Figure 2. The stars show data points that differ significantly from 1.00. The highest familial risks were noted for age groups 35 to 55 years of age, when the SIRs were about 3.0 from an affected mother and almost 4.5 from an affected sister. Daughters showed an early-onset component at ages 15 to 19 years, which appeared to be lacking for those with a sister history. The number of cases (6) was too few for showing data for the families in which 2 sisters and a mother were affected (see Table I). Of the 6 sisters in these families, 3 were diagnosed at ages 40 to 44 years (SIR 71.35, 13.45–171.93, data not shown) and 1 each at ages 45–49, 50–54 and 55–59 years. We do not show data for borderline ovarian cancer because no age specific SIR was significant. However, 4 of 7 affected sisters were diagnosed before age 40 years, which may be an explanation for the apparent lack of young sisters among invasive cancers.
The SNOMED codes have been used systematically since 1993 in the Swedish cancer registry and the analysis in Table II is based on histopathology of ovarian tumors in daughters by any invasive ovarian or breast cancer in probands. All histologic types of ovarian cancer showed a familial risk when a mother was affected. The SIR for epithelial tumors was 3.50 (95% CIs 2.67–4.49) and that for germ cell and sex cord tumors 2.11 (0.40–6.23). Seropapillary cystadenocarcinoma showed the highest SIR (4.49) and endometroid carcinoma the lowest SIR (3.14). A limited number of histopathology-specific cases were recorded from the sister history, and although the SIRs were increased, none were significant for a specific histology; all epithelial tumors showed an SIR of 2.98 (1.08–7.38). The only histology that was increased from maternal breast cancer was the nonspecific “adenocarcinoma”; all epithelial tumors showed an SIR of 1.30 (1.05–1.58).
Table II. SIR for Ovarian Cancer in Daughters According to SNOMED1
Bold type, 95% CI does not include 1.00. Period of follow-up; 1993-1998.
SIR adjusted for age, period, parity, age at first birth, SEI, residential region.
Family member has ovarian cancer
Germ cell and sex cord
Family member has breast cancer
Germ cell and sex cord
The overall SIR of ovarian cancer, 3.39, from a mother history in Table II can be used to estimate the PAF for family history. Among all daughters with ovarian cancer, 64/1,788 (3.58%) had an affected mother. Thus, the PAF was 2.52%.
The present study is the largest on familial ovarian cancer published to date, and the number of affected mother-daughter pairs (119 + 6) is about as large as that of 4 previous cohort studies combined (132).19, 20, 22, 23 Also, ours is the only study comparing mutually exclusive proband categories, which is of both clinical and scientific advantage. In clinical counseling, the exact family situation is relevant, whether only sister, only mother or both have been diagnosed with cancer. Scientifically, mixing the high-risk families where a mother and 2 sisters were affected (here SIR 24.03) with either mothers or sisters would increase SIRs in these groups. Particularly, the risk for sisters (30 affected with an SIR 2.94) would have increased substantially if 6 sisters were added from the high-risk families of an SIR 24.03. In the present study, the risk from a sister (SIR 2.94) was equally high as that from an affected mother (2.68), particularly when mothers' age was limited to ages below 67 years (3.21), similar to the age of daughters. The genetic interpretation of the results on different proband histories would be that the mother history could be due to dominant heritable effects, sister history due to recessive effects or effects transmitted from the paternal grandmother and mother and sister history due to high penetrant dominant effects.35, 36 However, a more plausible interpretation may be that the first 2 histories show low penetrant dominant effects, particularly because the SIRs were similar. The PAF for a mother history was 2.52%, which is in line with the previous studies from Europe, but it may be less than that found for a Canadian population, with a high prevalence of BRCA1/2 mutations.14, 16, 17, 18, 23 As a comparison, the PAF for breast cancer from a mother history is 3.61% in the Swedish Database.37 Of course, the meaning of PAF for family history is conceptually different from PAF for environmental exposure and because of less than 100% penetrance, familial PAFs may underestimate the magnitude of heritable effects.37 However, familial PAFs are meaningful for etiologic reasoning and guiding of health policy.
The size of the study matters, and we now had a chance to analyze the effect of age on familial risk, on which only limited previous data are available. These suggest an increase in risk at early diagnostic age, in line with many types of familial cancer.23, 26, 27 Even our age-specific data were sparse but they suggest 3 tentative conclusions: (i) age distribution of familial risk is unusual, peaking in the 40s; (ii) risks from mother and sister are approximately equal; and (iii) 1 or more discrete early age components exist. The age distribution was quite different, for example, from breast cancer, for which the familial risk is about 4.0 from mother or sister in the 20s; it then declines asymptotically towards 1.7 in the 60s.37 One speculative explanation to maximal risk relatively late in life, contrary to many heritable cancers, may be that the triggering mechanisms are age-dependent, such as sexual maturity and related hormonal changes. The discrete early-onset component(s) appeared to be unrelated to BRCA1/2 mutations because, for BRCA1 and BRCA2, the average age of onset has been reported as 51.2 and 57.5 years, respectively,14 but it could be related to some interacting genes, such as H-Ras changing the penetrance of BRCA1.15 If the median age of onset were equal in the Swedish BRCA1/2 families, it would appear that only a small proportion of the observed familial cases would be due to these genes, because the risks were highest at earlier ages. In the high-risk families with 2 sisters and a mother affected, half of the sisters were diagnosed at ages 40 to 44 years, which appears too young to be from BRCA1/2 families. These data suggest that other, yet unknown high-penetrant genes may predispose to ovarian cancer. Even another observation suggests that BRCA1/2 mutations account for a minor portion of the Swedish familial ovarian cancer. In Canadian ovarian cancer families with BRCA1/2 mutations, the risk of ovarian cancer was about 2 times higher than that for breast cancer.14 Assuming in the present study that all excess breast cancer in ovarian cancer families was due to BRCA1/2 mutations, we can estimate the share of these mutations in familial ovarian cancer. The SIR for breast cancer was 1.31, giving thus a calculated SIR of 1.62 for ovarian cancer, quoting the Canadian risk ratios. As the SIR for ovarian cancer was 3.39, some 26% of the excess risk of ovarian cancer could be explained by BRCA1/2 mutations. Genetic testing for BRCA1/2 mutations is available and appears as the first recommendation in clinical counseling for ovarian cancer.15, 38, 39, 40, 41 However, there will be a high risk for familial ovarian cancer even when the test is negative, and a clinical follow-up may be helpful.
Ovarian cancer presents in many histologic forms,2 but little is known about histopathology in familial cancers. An interesting finding in a recent case-control study on ovarian cancers showed a protective effect of oral contraceptives on serous tumors but an opposite effect on mucinous tumors.42 In the BRCA1/2 families, the tumors display small histopathologic distinctions, mainly as being nonmucinous.13, 14, 23, 43, 44 The present SIRs were highest from a mother history for the commonest histopathology, seropapillary cystadenocarcinoma (SIR 4.49) and lowest for endometrioid carcinoma (3.14). The group “other” with mucinous epithelial tumors and with nonepithelial tumors showed an SIR of 2.03. However, the number of cases for each specific histopathology was so small that no definite statements can be made for differences between the histologic types, particularly between sisters. A further point to note is that the histopathology of the proband's tumor could not be specified because of the small number of cases; thus the true histopathology-specific differences are likely to be larger than those presently observed. Borderline ovarian tumors have not shown a familial risk in an earlier study and no BRCA1/2 mutations were found in them.14, 21 Our data showed no significant familial risks but the numbers were small. However, the SIRs were above unity and, particularly for sister history, there was an excess of early-onset cases, perhaps diagnosed when familial risks were suspected.
In summary, familial risks for ovarian cancer were high and about equal from an affected mother and sister. Seropapillary cystadenocarcinoma showed the highest familial risk, but the effect of histopathology could not be fully assessed because of lack of data in probands. Age-specific data showed some early-onset components and an unusual maximal incidence in the 40s. Quoting data from an earlier study on BRCA1/2 mutations and relative risks of ovarian and breast cancer, the present results suggest that these mutations could account for 26% of the familial aggregation of ovarian cancer. It would appear that further gene identification efforts should target families with histopathology-specific, early-onset tumors.
The Family-Cancer Database was created by linking registers maintained at Statistics Sweden and the Swedish Cancer Registry.