From the *Division of Epidemiology and Biostatistics, University of Illinois at Chicago; and the †National Institute of Environmental Health Sciences, Research Triangle Park, North Carolina.
Submitted 10 October 2003; final version accepted 19 November 2004.
This research was supported by the National Institute of Environmental Health Sciences, by a grant from the National Cancer Institute to the University of North Carolina at Chapel Hill (#CA09330-20) and by a grant from the National Cancer Institute to the University of Illinois at Chicago (#CA57699-06).
Correspondence: Garth H. Rauscher, School of Public Health, Division of Epidemiology and Biostatistics, University of Illinois at Chicago, SPHPI (M/C 923), 1603 W. Taylor, Chicago, IL 60612. E-mail: firstname.lastname@example.org.
Background: A family history of cancer is a marker of familially shared genetic and environmental risk factors for cancer but is subject to inaccurate and biased reporting.
Methods: In a case-control study of adult acute leukemia, we examined the accuracy of cancer reports in deceased first-degree relatives using death certificates and the National Death Index (NDI). We submitted information on 1058 deceased relatives to the NDI or civil vital statistics registries. Social security numbers were not available and identifying information was incomplete.
Results: A death certificate was located for 70% of records sent to state registries but for only 32% of records submitted to the NDI. Death certificates confirmed 95% of relatives reported to be cancer-free and 83% of those reported to have any form of cancer. Confirmation of cancer of specific sites ranged from 50% for breast cancer and 54% for leukemia to 86% for prostate cancer.
Conclusions: State searches may be preferable to the NDI when identifying information is incomplete.
Family history of cancer is often used as a marker of familially shared genetic and environmental risk factors for cancer, but may be subject to inaccurate and biased reporting. Between 1986 and 1991, we conducted a case-control study of adult acute leukemia in the United States and Canada.1 Patients who were themselves interviewed were considerably less likely to report a family history of cancer than patients for whom the interview was conducted with a proxy respondent.2 To determine if these differences were the result of differences in the accuracy of patient and proxy reports, we examined the accuracy of family history of cancer reporting by case-control status and interview type. Many of the relatives’ deaths predated National Death Index records (1979) and identifying information was often incomplete. Because relatives were from diverse locations, were often deceased, and had diagnoses spanning many years, it was not feasible to validate relatives’ cancer histories through medical records.
Details regarding the epidemiologic study have been presented elsewhere.1,2 Incident cases of adult acute leukemia (n = 811; ages 15–79 years) were contacted by telephone within a few days of registration. Interviews conducted with the participants themselves were completed a median of 8 days after registration. Proxies participated in 34% of patient interviews (21% without and 13% with the patient also being interviewed); these interviews were completed a median of 34 days after registration. Spouses participated in slightly over half of proxy interviews, and first-degree relatives participated in most of the remainder. Proxies participated in 13% of control interviews, for the most part without the control (11%). More than two thirds (70%) of control proxies were spouses.
As part of a telephone-administered interview, cases and controls (or their proxies) were asked to report for their mother, their father, each sibling, and each child any history of: leukemia; any other “cancer, tumor or growth such as lymphoma or lung cancer” and the specific sites involved; and aplastic anemia or any other blood disorder. When a case or control reported a relative to be deceased, a question was then asked about the cause of the relative's death. We used both cancer/blood disorder reports and cause-of-death reports together to define each relative's history of cancer.
For deceased relatives (2485 deaths, 10,819 relatives), the interviewer requested (regardless of cancer history) the decedent's name, date of birth, date of death, and city and state of death. We identified 1058 decedent records that appeared to have sufficient personal identifying information to warrant a search for cause-of-death information using either state registries or the National Death Index (NDI).
National Death Index Search Strategy
We submitted records for the 489 persons who died in 1979 or later to the NDI for computerized matching of death certificate data.3 We did not collect social security number, and for 192 decedents, we also lacked birth month and were required to submit 12 duplicate records (1 for each possible month) on these individuals. The resulting final submission file contained 2601 records for the 489 decedents.
NDI returned a matching file that appended variables to each original record for which a potential match was found, at times creating multiple records for multiple potential matches. We defined a true match as an identical match on first name, last name, state of death, and birth years that matched to within 5 years and death years that matched to within 2 years. As a result of confidentiality requirements, NDI could not release the names of the decedents in their files. Because we could not assess the closeness of phonetic matches on last name, we elected to discard these potential matches.
State Search Strategy
We had previously requested copies of death certificates from state registries for the purpose of validating cancer histories in a small subset of deceased relatives. We added the information obtained from these death certificates to the data from NDI searches. In addition, we requested death certificates for unsearched relatives deceased before 1979 and death certificates from the states when the NDI search did not yield a “true” match.
We defined verification rate as the percentage of site-specific cancer reports that were listed with the same site on the death certificate or NDI file. We included any mention in the death certificate or NDI file of cancer as the cause of death, underlying cause of death, or other significant condition.
Match rates were 32% for NDI and 70% for state searches (Table 1). For reports examined with both search strategies, state searches yielded higher match rates than NDI searches. State searches were able to locate death certificates for 95% of subjects for whom a match was also found through the NDI. State searches were also able to locate death certificates for 55% of subjects for whom a record could not be located through the NDI. In contrast, NDI searches were able to locate only 63% of the state matches and only an additional 10% of state searches without a match.
Regardless of search strategy, matching was improved when information was available on middle or maiden name, birth day or month, or death month (results not shown). Paradoxically, city of death, used only in state searches, proved to decrease the likelihood of a match, possibly because misreported cities were used by state vital records clerks to rule out true matches.
Match rates were better for younger cases (<50 compared with 50+ years) and for male relatives. In both strategies, match rates were similar for deaths reported by cases or proxies, except for data from control proxies in state searches (58% vs. 71–74%).
Ninety-five percent of negative cancer reports in deceased relatives were verified from cause-of-death information on death certificates (Table 2) (94% for state matches and 95% for NDI matches). Reported positive cancer histories (any site) were verified for 83% of deceased relatives (84% for state matches and 79% for NDI matches). Verification of site-specific cancer reports ranged from 50% for breast cancer and 54% for leukemia to 86% for prostate cancer. Restricting our definition of positive cancer reports to those reported as a cause of death by the respondent only marginally improved verification rates (results not shown).
Prior validation studies have found that a large percentage (85+%) of self-reported family histories of any cancer, breast, colorectal, and lung cancer can be confirmed through medical records.4–12 These studies, however, may not generalize well to reports of family history in case-control studies of rare and rapidly progressing cancers such as adult acute leukemia. In the present study, in which patients and their relatives may live anywhere in the United States, and relatives may have been diagnosed decades earlier, validation through medical records was not possible. The only realistic option was to attempt to validate cancer histories for the 22% of relatives who were deceased.
Validation results for deceased relatives may not generalize well to all relatives living or deceased. Furthermore, lack of confirmation through the death certificate may overestimate the amount of overreporting of relatives’ cancers by study participants. A true cancer history may never appear on the death certificate as a contributing cause of death, especially if the cancer was not lethal.13,14 This may explain why concordance between reports and death certificates was lower for breast cancer than for lung cancer, which is much more lethal. Similarly, cancers that tend to be diagnosed at older ages may be more likely to be listed on the death certificate, which may explain why prostate cancer reports were more frequently confirmed by death certificate than breast cancer reports. Lack of concordance on cancer history, therefore, may reflect these limits of death certificates for capturing some cancers.
Match rates in the context of this study were considerably better when going through individual state and metropolitan vital statistics registries, compared with the NDI Because the NDI is not generally permitted to release decedent names, we chose to discard their phonetic matches on last name. This further reduced our match rates. State searches had more flexibility and were often able to find a match when identifying information was incomplete. It is possible that in some cases we received incorrect matches through either search strategy; such errors would tend to raise our match rates and lower the verification rates.
Despite the strong likelihood of misclassification, self-reported family history of specific cancers has been linked to risk of many types of cancers in both case-control and cohort studies. Kadan-Lottick et al.15 were able to classify a greater percentage of reports of cancer in family members by making small changes in their family history questionnaire. Although validation studies would help to rule out recall bias as an explanation for links between family history and disease, estimating the validity of patient and control reports of their family cancer histories may not be possible in population-based studies that cover a wide geographic area. Efforts may be better spent improving the quality of questionnaires.
We thank Glenn Heartwell, David Shore, the staff at WESTAT, Inc. and CODA, Inc., and the patients and CALGB facilities that participated in this study.
1. Sandler DP, Shore DL, Anderson JR, et al. Cigarette smoking and risk of acute leukemia: associations with morphology and cytogenetic abnormalities in bone marrow. J Natl Cancer Inst. 1993;85:1994–2003.
2. Rauscher GH, Sandler DP, Poole C, et al. Family history of cancer and incidence of acute leukemia in adults. Am J Epidemiol. 2002;156:517–526.
4. Demers PA, Vaughan TL, Checkoway H, et al. Cancer identification using a tumor registry versus death certificates in occupational cohort studies in the United States. Am J Epidemiol. 1992;136:1232–1240.
5. Engel LW, Strauchen JA, Chiazze L Jr, et al. Accuracy of death certification in an autopsied population with specific attention to malignant neoplasms and vascular diseases. Am J Epidemiol. 1980;111:99–112.
6. Aitken J, Bain C, Ward M, et al. How accurate is self-reported family history of colorectal cancer? Am J Epidemiol. 1995;141:863–871.
7. Anton-Culver H, Kurosaki T, Taylor TH, et al. Validation of family history of breast cancer and identification of the BRCA1 and other syndromes using a population-based cancer registry. Genet Epidemiol. 1996;13:193–205.
8. Bondy ML, Strom SS, Colopy MW, et al. Accuracy of family history of cancer obtained through interviews with relatives of patients with childhood sarcoma. J Clin Epidemiol. 1994;47:89–96.
9. Douglas FS, O'Dair LC, Robinson M, et al. The accuracy of diagnoses as reported in families with cancer: a retrospective study. J Med Genet. 1999;36:309–312.
10. Floderus B, Barlow L, Mack TM. Recall bias in subjective reports of familial cancer. Epidemiology. 1990;1:318–321.
11. Garber JE. Validation of family history of breast cancer and identification of the BRCA1 and other syndromes using a population-based cancer registry. J Women's Health. 1997;6:349–351.
12. Kerber RA, Slattery ML. Comparison of self-reported and database-linked family history of cancer data in a case-control study. Am J Epidemiol. 1997;146:244–248.
13. Sijmons RH, Boonstra AE, Reefhuis J, et al. Accuracy of family history of cancer: clinical genetic implications. Eur J Hum Genet. 2000;8:181–186.
14. Theis B, Boyd N, Lockwood G, et al. Accuracy of family cancer history in breast cancer patients. Eur J Cancer Prev. 1994;3:321–327.
15. Kadan-Lottick NS, Friedman DL, Mertens AC, et al. Self-reported family history of cancer: the utility of probing questions. Epidemiology. 2003;14:737–740.
© 2005 Lippincott Williams & Wilkins, Inc.