Evaluation of the Family History Collection Process and the Accuracy of Cancer Reporting among a Series of Women with Endometrial Cancer¹

Family-based cancer genetics research has made possible the discovery of numerous cancer susceptibility genes, pointed to molecular genetic pathways that are disrupted in tumor initiation and progression, and offered insights into normal cellular development. Family history data are essential for this research as a combination of linked molecular and family/medical history data are needed to determine the penetrance and expressivity of putative susceptibility alleles. Family history data are critical not only in the study of hereditary cancer syndromes but also in research conducted to identify cancer risk modifier genes. Modifier genes lead to a modest elevation of cancer risk and may play a role in a far greater percentage of the population than the rare, highly penetrant cancer susceptibility alleles. Family history data are also valuable in surveillance and chemoprevention trials of individuals at increased risk for disease, as the efficacy of these interventions cannot be measured without an accurate assessment of individuals’ family-based risk.

Few studies have examined the accuracy of cancer family history reporting in the United States (1, 2, 3, 4, 5). In other countries, centralized cancer registries have facilitated research focused on determining the accuracy of cancer family history reporting (6, 7, 8, 9). Some investigators have studied high-risk populations or individuals attending cancer risk clinics (4, 8, 9, 10). In most studies, the estimation of reporting accuracy has been limited to first-degree relatives or a specific cancer type (6, 7, 9, 11, 12, 13, 14). Furthermore, few studies have examined both the process to collect and verify family history information, as well as the accuracy of the cancer reporting in a research setting (1, 9).

Studies to evaluate the identification, collection, and reporting accuracy of more than one cancer type, using multiple generation family history data, are needed. The purpose of this study was to analyze the process for collecting and verifying reported cancer family histories, as well as to identify reporting inaccuracies among a series of women with endometrial cancer.

The study population was derived from a cohort of 80 women with histologically confirmed endometrial cancer who were treated in the Division of Gynecologic Oncology at Washington University School of Medicine from 1993 to 1998. Participants consented to family history and tumor studies as part of a protocol approved by the Washington University School of Medicine’s Human Studies Committee. The 80 women included in this study were selected from a series of 231 women from whom tumor tissue was available for DNA studies. Cases were selected without knowledge of family history information, age at diagnosis, or other clinical features suggestive of inherited susceptibility to cancer. The study was designed to obtain equal numbers of cases with tumor microsatellite instability (defective DNA mismatch repair) and tumors lacking microsatellite instability (40 in each group). Participants were the probands for the family history study and are referred to as “probands” throughout.

Family history data were obtained at two different times. First, as part of their routine clinical care, participants completed a medical history questionnaire that included a single question regarding the health of first-degree relatives. These data, as well as information documented in the initial medical encounter, comprise the screening family history (Fig. 1).

A genetic counselor later obtained a detailed, three-generation pedigree from each proband, either face-to-face or over the telephone. Collection of the detailed family history did not begin until 1998, usually when a proband returned for clinical follow-up. The following data were obtained for all individuals in the pedigree: current age or age at death, history of cancer (including site, age at diagnosis, and city or hospital where treated), and other pertinent medical history (including cause of death, history of chronic illness, and surgical history). Directed inquiries regarding breast, gynecologic, and gastrointestinal screening or illness were made. These data comprise the detailed family history (Fig. 1).

Medical record verification was sought for 289 cancers reported in the detailed family histories. Medical records were also pursued for 82 noncancer procedures, including hysterectomy, gynecologic procedures, colonoscopy, colon surgery, and breast biopsies. As is standard practice, the genetic counselor requested the proband contact the identified relative or, if deceased, the closest living family member, to obtain permission to release their medical records. Medical records and/or death certificates were requested for verification. The genetic counselor contacted the proband periodically by telephone to assist with the collection of the signed medical release forms. Signed release forms were returned to the genetic counselor, then forwarded directly to the hospital. If permission to release medical records was not granted, the reason stated by the proband was documented and assigned to one of three categories: (a) proband unwilling to approach identified relative with cancer; (b) the identified relative was unwilling to release records; and (c) insufficient information was known by the proband to request medical records (Fig. 1).

All records received were independently reviewed by a medical geneticist (A. W.) and a cancer research investigator (P. G.). A cancer reported by a proband was considered to be inaccurate if on medical record review the cancer site was incorrectly reported, the cancer site was unknown to the proband, or the proband did not report the cancer. The medical geneticist (A. W.) used the family history to assess the family-based cancer risk, focusing on malignancies associated with HNPCC.³ Reported cancers were only considered in the risk assessment if they were verified with medical records. The criteria used to classify each family history are shown in Table 1.

Statistical comparisons using the two-sample z test for binomial proportions, the t test for the difference in two means, and the χ² test for difference in multiple proportions were made. All Ps were two tailed, and Ps ≤ 0.05 were considered statistically significant.

Eighty women treated surgically for endometrial cancer comprise the study population. The average age at the time of surgery was 63 years (38–87 years). Seventy-four women were Caucasian, and 6 were African-American. Ten women had a synchronous or metachronous malignancy, and 2 women had a third primary cancer.

The 80 probands reported the detailed family history information to the genetic counselor and assisted with medical record collection. Information on 634 first-degree, 1501 second-degree, and 790 third-degree relatives was provided. Probands were unable to report the total number of third-degree relatives or consistently provide detailed medical information for such relatives.

Probands reported 11 additional cancers among themselves and 278 cancers among first-, second-, and third-degree relatives for a total of 289 reported cancers. Although breast (39 reports) and lung (29 reports) cancer was reported frequently, cancer of unknown site (63 reports) was the most common reported malignancy (Table 2). For the 100 cancer reports among first-degree relatives, only 4 cancers were reported among the probands’ children. Cancer was significantly more likely to be reported in first-degree relatives (100 of 634) than second-degree relatives (116 of 1501; P < 0.001; Fig. 1).

In addition to the 289 reported cancers, 10 additional cancers, not reported initially by probands, were identified later during medical record collection and review. Four of the 10 cancers that were not reported by the probands were among first-degree relatives. Five of the unreported cancers were second primaries in relatives identified by the proband as having only one malignancy. Two cancers not reported by the proband (colon and skin cancer) were identified during review of medical records for noncancer procedures (Table 3).

The genetic counselor attempted to contact probands every 3 weeks to follow-up on requests for release of medical records. A total of 327 contacts were made to 78 probands, for whom additional records were requested. The fraction of medical records retrieved from hospitals by the number of contacts made with each proband was evaluated. Success in obtaining medical records peaked after three contacts. There was a statistically significant relationship between the number of contacts made with probands and the fraction of medical records that were retrieved (χ² = 23.68, d.f. = 7, P = 0.001; Fig. 2).

Seventy-eight probands were asked to contact identified relatives to initiate retrieval of medical records. Nineteen probands (24%) refused to contact relatives or were unsuccessful in obtaining permission to release medical records. There was no significant difference between obtaining permission from relatives to release cancer records than noncancer records or to release cancer records from identified female relatives than male relatives. However, permission to release cancer medical records was significantly more likely to be obtained from first-degree relatives (72 of 100) than from combined second- and third-degree relatives (75 of 178; P < 0.001; Table 2).

Once permission was granted, attempts were made to retrieve medical records from the identified hospital. Of the 156 cancer and 54 noncancer records requested, 102 (35%) and 42 (51%) medical records, respectively, were retrieved successfully. Noncancer procedure medical records (42 of 82) were significantly more likely to be retrieved than cancer records (102 of 289; P = 0.01). The great majority, however, of noncancer procedure reports was reported for either the proband or for first-degree relatives. Both cancer and noncancer records were also significantly more likely to be retrieved from a hospital if the identified relative was alive (48 of 108) than dead (63 of 220; P = 0.001). This significance held when only cancer records were examined as well (P = 0.01). Cancer reports in which the proband identified a specific anatomical site or cancer type (91 of 226) were significantly more likely to be retrieved than cancer reports of unknown anatomical site (11 of 63; P = 0.001). Furthermore, a significantly greater number of cancer reports was verified with medical records among first-degree relatives (47 of 100) than combined second- and third-degree relatives (46 of 178; P < 0.001). The likelihood medical records were successfully retrieved when requested from a hospital declined with the age of record, such that <50% of records were retrieved when records were ≥40 years old (Fig. 3). There was a significant relationship with age of medical record and the fraction of records obtained from hospitals (χ² = 37.24, d.f. = 5, P < 0.001).

Medical records were retrieved for 112 cancers (102 cancers reported by the probands and 10 additional cancers identified by medical record review). Thirty-two of the 112 (28.6%) verified cancers were identified to be inaccurately reported by 19 probands (Table 3). One proband had four separate inaccurate reports. Only one cancer report (pancreatic cancer) was verified not to be a cancer (over reporting). Probands with inaccurate cancer reports had a significantly higher average number of relatives reported with cancer (mean = 7.1 relatives) than probands with no cancer reporting inaccuracies (mean = 2.9 relatives; P < 0.001).

Cancer reporting errors were classified into two groups. The first group consisted of 11 cancer reports in which the cancer site was reported incorrectly by the proband (Table 3). These inaccurate reports lower the number of true positives for a given site and, thus, the overall reporting accuracy rate. None of the cancers reported represented a metastasis from a primary site. The most common site inaccurately reported was “stomach cancer.” The second group of inaccurate reports included both cancers reported as “unknown site” and malignancies that were not reported by the proband but were identified during medical record review. This type of inaccuracy lowers the sensitivity of the family history interview process. The cancers reported as “unknown” for which the type was verified by medical records represent a variety of cancer sites, with cervical cancer being the most common. Colon carcinoma was the most frequent cancer not reported (Table 3). Inaccurate reports were more common among second- and third-degree relatives (21 of 56) than first-degree relatives (10 of 55; P = 0.02). There was no difference in the probands’ reporting accuracy for identified male and female relatives with cancer (Table 3).

Detailed family history data were used to classify probands’ family-based cancer risk into the following risk categories: 50 with baseline family cancer risk, 19 with low family cancer risk, 8 with moderate family cancer risk, and 3 with high family cancer risk. All three families classified as high cancer risk met Amsterdam I or II criteria for HNPCC. The preponderance of families with “baseline” cancer risk is expected, given the population was chosen from a cohort of women who were being treated in a gynecologic oncology office, not a high-risk clinic.

Family histories were reevaluated after the identification of inaccurate cancer reports. For the 19 probands with inaccurate cancer reports, 2 (11%) probands had an increase in their clinical cancer risk assessment. One family moved from baseline up to low risk, with little consequence in clinical cancer screening recommendations. However, the other family moved from moderate to high risk, with the clinical diagnosis of HNPCC being made. None of the 19 probands with inaccurate cancer reports had a decrease in their clinical cancer risk assessment.

Family-based cancer genetics research and clinical risk assessment both depend on the allocation of substantial resources to collect and verify family history data. As research moves from highly penetrant single gene disorders in high-risk populations to lower penetrant modified genes in the general population, reliable methods of family history collection must be identified. Most studies published to date used high-risk populations or national cancer registries. In anticipation of research and clinical cancer genetics moving toward lower risk groups, we undertook this analysis to better understand the efficacy of the family history collection process and the cancer reporting accuracy in an older, lower risk population.

General research populations may have limited knowledge of or motivation to provide family history information. Timely follow-up contacts with the study population are necessary to obtain permission from relatives to release medical records. In our study, the number of contacts with probands was significantly related to the eventual receipt of medical records. With as many as five timely contacts, the success for obtaining medical records was slightly >50%. A portion (24%) of probands refused or had insufficient data to track records or contact identified relatives with cancer. This finding is consistent with other investigations, in which insufficient detail to collect medical records was the most common reason for failing to verify reported cancers (5, 10).

As expected, probands had a lack of health knowledge of and personal contact with distantly related relatives. Cancers and noncancer procedures were reported significantly more frequently among first-degree relatives than more distant relatives. There was a greater likelihood to receive permission for release of records from first-degree relatives than combined second- and third-degree relatives, consistent with other investigations (4, 9, 10). Second- and third-degree relatives, however, are critical in the determination of the expressivity and penetrance of cancer genes, and as such, investigators may need to contact distant relatives directly so detailed medical history information and medical records may be obtained (1, 2).

The ability to verify cancer reports with medical records varies from study to study. Investigations using populations from high-risk cancer clinics have tended to achieve high verification rates, ranging from 42 to 94% among first-degree relatives with cancer (4, 8, 10). Studies that use cancer registries also have good confirmation rates, ranging from 43 to 97% (6, 8, 9). For the research population we investigated, medical records were retrieved for 35% of all cancers reported and nearly 50% of cancers reported among first-degree relatives. The success in retrieving medical records was greatest for cancer reports in which the cancer site was known and the identified relative was alive. This latter relationship, as well as the lower retrieval rate among second- and third-degree relatives, may reflect the record age, as there was a significant relationship in the likelihood of record retrieval with the record age.

These data suggest it may be more difficult to obtain detailed family history information and medical records from populations in which the probands are older and of “lower risk.” A more effective screening family history tool, instituted at the time of first clinical contact, may reduce the time of the detailed family history interview process and identify families of greatest research interest. These data also suggest direct contact with distant relatives may aid in cancer reporting and medical record retrieval. It would be useful to conduct a similar study in a younger group of participants with similar “low risk.”

Ten cancers, identified on record review in this study, were initially not reported by the proband. In 1 family, three unreported cancers led to the increase in their clinical cancer risk assessment. As many as 35 nonreported cancers have been identified on record review from a series of 288 families (6, 10). These cancers were most often second primaries unknown to the proband. Second primary cancers are critical, as they represent a key clinical diagnostic feature in the identification of increased familial cancer risk. To address the issue of underreporting, investigators have performed random checks of relatives’ medical records, with a rate of 3–6% underreporting rate noted (1, 5). For this study, noncancer procedure records were requested as a check of cancer reporting, but only 2 cancers were identified.

Several factors have been evaluated for their effect on cancer reporting accuracy (1, 3, 4, 10, 11). The relationship of the identified relative to the proband is one such factor. Probands in our series had a higher reporting accuracy among first-degree relatives than more distant relatives, which is consistent with other reports (4, 10). Of the 10 inaccurate reports among first-degree relatives, 50% were for the probands’ mothers. This may reflect the relatively older age of the study population, as cancer diagnoses were not as openly discussed among previous generations. In this study, we saw no difference in the cancer reporting accuracy between identified male or female relatives or between relatives who were deceased or living. To the best of our knowledge, the effects of gender or living status have not been reported previously.

Ultimately, the process to verify reported family history data are made to assess the family-based cancer risk necessary for cancer genetics investigations. Two probands had an increase in their family’s cancer risk assessment. This increase in risk assessment contrasts with published reports and may reflect our use of a general research population. Of the 213 families attending a high-risk cancer clinic, management was altered in 23 families (11%), and of these, intensified screening was no longer recommended in 15 families (10). With 43% of cancer reports verified with medical records, Sijmons (8) found 5% of families had a change in clinical management, and the majority had a decrease in risk assessment. Additional data are needed among general research populations to determine trends in the clinical risk assessment after medical record collection.

In summary, family-based cancer genetics investigations are time consuming and costly. The inclusion of general populations in these investigations poses unique challenges in the collection and verification of the reported family history. Our study suggests family history data may be more difficult to collect among a general research population, with older probands, then has been reported previously in high-risk populations. The ultimate goal is to gather data for the family-based cancer risk assessment that can be used in conjunction with molecular data. Efforts to improve efficiency of data collection and identify inaccurate reporting are needed with similar aged and younger probands used in cancer genetics investigations.

We thank the women and their families who participated in this study. We also thank the Biostatistics Core in the Siteman Cancer Center for their review of this manuscript.