Underreporting of Family History of Colon Cancer: Correlates and Implications¹

CRC³ affects roughly 5% of Americans at some time in their lives and large bowel cancer represents 15% of all cancers (1). There is growing evidence that a family history of CRC is associated with increased risk for CRC (2, 3, 4, 5, 6, 7). Recent molecular studies have led to the identification of genes associated with hereditary nonpolyposis colon cancer (8, 9) and to the development of genetic susceptibility testing for hereditary nonpolyposis colon cancer.

Most advances in cancer genetics depend on obtaining accurate information on family history. Cancer risk counseling, management of high-risk patients, and decisions about the appropriateness of predictive genetic testing also require information about familial cancer history. Because some people may not report, or may be unaware of, cancer in their families, it is important to understand whether and why people may underreport their family histories.

The accuracy of self-reported family history information has been examined for many diseases including cancers (10, 11, 12, 13, 14, 15). Studies have found family history of melanoma to be often overreported (14, 16) and breast cancer family history to be very accurate (11, 12, 13). However, in the only two studies to date designed to detect underreporting, about one-quarter of those who had blood relatives with cancer documented in medical records did not report a family history (12, 17).

Few studies have reported on the accuracy of reports of family history of colon cancer or have included very small numbers of cases of CRC (15, 18, 19, 20). In this study, we explore the extent of underreporting of CRC family history and its predictors (demographics, cancer characteristics, knowledge, family communication) and correlates (cancer worry, perceived risk). We hypothesized that accurate reporting of CRC family history (awareness) would be predicted by certain demographic characteristics; by cancers diagnosed at an earlier age and in a more distant stage; by greater knowledge about CRC; and by greater family communication and social support. We further hypothesized that awareness of a FDR with colon cancer would be associated with higher levels of cancer worry and higher perceived risk.

Data reported here are from two sources: (a) a population-based case-control study of the genetic epidemiology of CRC (Ref. 7; the epidemiological survey); and (b) the “family survey of attitudes toward genetic testing for colon cancer” (the psychosocial survey). Relatives of case probands and control index persons of Caucasian, Japanese, and Hawaiian ethnicity who were originally identified for, and agreed to participate in, the epidemiological survey were invited to complete the psychosocial survey.

Case probands consisted of men and women, ages 60 and younger, diagnosed with invasive adenocarcinoma of the large bowel between 1987 and 1996 who were ascertained through the HTR and whose diagnoses were histologically confirmed. Cases were asked to identify parents, siblings, and adult children (over age 18) living in Hawaii. These family members were asked to complete a mailed epidemiological survey (85% response rate). Because the survey was part of a case-control study, the invitation did not state whether or not they were included because of a family history of colon cancer.

The psychosocial survey was a self-administered mail survey of CRC-free relatives who completed the epidemiological survey. These analyses included all of the siblings and adult children of cases. All of them had a verified FDR with colon cancer and provided reports of FDR cancers (among parents and siblings). The total sample includes 426 people from 160 families (77% response rate).

Data collected from both the epidemiological and psychosocial surveys and the HTR data, were used in these analyses. Survey data included background information, reported family history of cancer, knowledge about cancer, family support and communication, cancer worry, and risk perception. Information about cancer characteristics of the probands was ascertained from the HTR. The theoretical foundation for the psychosocial constructs was the precaution adoption model (21), which is concerned with how individuals perceive threatening situations or hazards, and what factors influence them to engage in protective action or adaptive coping behaviors. According to the precaution adoption model, major determinants of beliefs about one’s disease risk include personal experience with a threat (e.g., a family member being diagnosed with a disease), risk perception, and information about a risk (21, 22).

Background factors included relationship to the proband (sibling or adult child), ethnicity, sex, age, education, and religion. Cancer characteristics for the proband’s CRC included: the proband’s age at diagnosis with CRC and the stage of the cancer at diagnosis.

Reported family history of cancer, or “awareness” was determined from reports about whether a parent and/or sibling had been diagnosed with cancer of the colon or other cancers (breast, ovary, lung, prostate, or other). All of the responses to “other” were examined and were recoded into the colon cancer category if they specified rectal, bowel, large bowel, intestinal, or colorectal. Because all of the respondents were known to have at least one confirmed FDR with CRC, a binary colon cancer awareness variable was created. Persons reporting no parent or sibling diagnosed with CRC were considered not aware, and those reporting one or more parent or sibling with CRC were classified as aware.

Knowledge about CRC was assessed with an 11-item measure of general knowledge of colon cancer (α = 0.59) and a 10-item measure of knowledge of heredity and cancer (α = 0.78; Ref. 23). Communication/social support was measured by asking how often respondents would talk to family members about serious problems, such as a major illness, and how many family members they could talk with about such problems. The cancer-related worry measure assessed the frequency of concerns and intrusive thoughts about getting colon cancer and was measured with a four-item scale (α = 0.77; adapted from Ref. 24). Risk perception was measured by asking how high the respondents believed their chances of getting colon cancer or polyps were, compared with other people their age (two items, α = 0.92).

After computing descriptive statistics for all of the variables, bivariate analyses were performed to examine the associations between independent variables and awareness and between awareness and cancer worry and risk perception, using t tests and contingency table analysis (χ² tests). Variables that were found to be significant or marginally significant (P < 0.10) were included in subsequent logistic regression analyses.

Because there were multiple observations within families, we used generalized estimating equations (25, 26), to obtain consistent estimates of regression coefficients and standard errors. Adjusted for background factors, the family intraclass correlation for awareness that a FDR has colon cancer has a value of 0.42.

Multiple logistic regression was used to assess the contributions of demographic variables, cancer characteristics of probands, and knowledge and communication variables to the binary measure of awareness of CRC. Independent variables that were nominal with more than two classes, such as ethnic group, were treated by creating a set of dummy variables for the classes, with one class (the comparison group) being the intercept (27); a multivariate Wald test was used to test for differences in the mean of the dependent variable across the different classes. Multivariate analyses proceeded in hierarchical fashion, testing blocks of covariates and independent variables in the following sequence: (a) demographic factors; (b) familial cancer characteristics; and (c) the combined model (both demographics and familial cancer characteristics). A final model was fit by adding the block of variables assessing knowledge and communication.

To examine the contribution of background factors, cancer characteristics, and awareness to concern about CRC, regression models were fit with cancer worry and risk perception as dependent variables. Finally, both bivariate and multivariate analyses were repeated excluding relatives of probands with in situ cancers (n = 30).

Respondents were siblings (n = 274, 64.3%) and adult children (n = 152, 35.7%) of 160 CRC probands living in Hawaii and were of Japanese (78.9%), Hawaiian/Part-Hawaiian (11.7%), or Caucasian (9.4%) ethnicity. The sample was 49.3% male, with a mean age of 50 years (range, 19–84 years). About one-third had no more than a high school education and 29.8% were college graduates. One-fourth indicated Protestant religious affiliation, 31.7% were Buddhist, and the remainder were Catholic (12.4%), other (13.6%), or no religion (17.4%). Almost all of the respondents (98.1%) were born in the United States. Although all of the respondents were known by us to have a sibling or parent with colon cancer, 25.4% indicated that they had no FDRs with colon cancer.

We compared the 25.4% of respondents who indicated that they did not have a parent or sibling with colon cancer (unaware) with those who were aware (Table 1). Awareness was not significantly related to relationship to the proband (sibling or child), ethnicity, religion, or age of the proband at diagnosis. Males and older persons tended to be unaware more often. Educational level and cancer stage were strongly associated with awareness in a stepwise manner. Those with higher knowledge scores, greater family communication, more cancer worry, and higher perceived risk were significantly more likely to report a FDR with colon cancer.

The major factor associated with awareness of a relative’s cancer is the invasiveness of the cancer. The odds of being aware increased 7-fold if the cancer was localized rather than in situ and 20 times greater if the cancer was regional or distant. The other variable that reached statistical significance in the original multivariate models was respondent gender, with males being one-half as likely as females (odds ratio = 0.56; P = 0.015) to correctly report a FDR with colon cancer. Adding the block of variables reflecting knowledge and communication significantly improved the fit of the model (Wald χ² = 16.40 with 4 degrees of freedom; P < 0.005), although only the partial odds ratios for the variables sex, stage, and “knowledge of colon cancer” were significant, after adjusting for the other variables (Table 2).

Exponential regression models were fit with cancer worry—as the dependent variable—and demographics, cancer characteristics, and awareness as independent variables. The strongest association with cancer worry was being aware that a relative had colon cancer. Awareness of an affected FDR multiplied the index by exp(0.1247) = 1.133, or an average increase of about 13% in the cancer worry index.

Multiple linear regression was also performed with risk perception as the dependent variable and demographics, cancer characteristics, and awareness as independent variables. In these regression models, only two variables were significant: (a) being a college graduate; and (b)awareness of a relative having colon cancer. The effect of having a college education increased perceived risk by at least as much as knowing that a first degree relative has colon cancer.

We repeated both bivariate and multivariate analyses excluding relatives of probands with in situ cancers (n = 30). The results of these analyses were similar to those including the entire sample of relatives, with only minor changes in the levels of statistical significance and no change in the direction of observed associations.

Although our study was not primarily designed to investigate awareness of familial colon cancer, the surprising finding that more than one-fourth of the respondents known to have a sibling or parent with CRC reported having no FDRs with CRC warranted further examination. Awareness was highly correlated within families. Caucasians, males, less educated persons, Catholics, and older respondents tended to be unaware more often. Those who were aware of a FDR’s colon cancer were more knowledgeable about CRC and about heredity and reported more family and friend social support, more cancer worry, and higher perceived risk. When the relative’s cancer was diagnosed at an earlier stage, a sibling or child was significantly less likely to be aware of it. Interestingly, ethnicity, age, religion, and education were no longer significant predictors after controlling for knowledge and stage-at-diagnosis.

The bivariate associations that we found between awareness and gender, age, and education are consistent with findings of several other recent studies (10, 15, 18, 19). However, none of these comparable reports reported multivariate analyses; therefore, it is difficult to rule out alternative explanations such as disproportionate age/sex/education distributions.

Underreporting of family history of CRC may be due to confusion or lack of awareness of a relative’s cancer, lack of communication within families, or reluctance to report family history on a mailed survey. Research to date suggests that there may be large differences across cancer sites in the accuracy of reporting family history (10, 13, 18).

Some people may choose not to disclose a family history of cancer on a mailed questionnaire (19). If this is the case, such underreporting might be reduced in clinical settings where the family history is taken in writing and followed by an interview with a family physician, oncologist, or genetic counselor. Direct interviews with relatives and the use of diagnostic screening criteria have been shown to improve the accuracy of reporting family history in research on neurological and psychiatric conditions (28, 29, 30).

Even skillful interviewing, however, may be insufficient to address this problem. Both negative and positive family history should be verified through review of medical records or linkage to cancer registries. Still, it may not be possible to verify all of the negative family histories, if family members are geographically dispersed, estranged, or no longer alive.

In genetic epidemiology research, reporting bias in family history may not affect study results if under- or overreporting is equal across case and control groups (14, 18). However, in clinical medicine, failure to report family cancer history may lead to missed opportunities for surveillance and ultimately, to avoidable mortality (20). Thus, clear and accurate provider-patient communication and family communication are of paramount importance. Health professionals should stress the need for open family communication to CRC patients and emphasize that this information is important for other family members to make health decisions.

The accuracy of family health history information is central to advances in research and to the provision of clinical services, counseling, and patient education (31). As new studies continue to reveal the significance of family history as a significant predictor of susceptibility mutations for breast, ovarian (32), colon, and prostate cancers (33), this issue takes on increasing importance.

We acknowledge the contributions of Dr. Haiou Yang, Dr. Caryn Lerman, Rosemary Casey, Dr. Lue Ping Zhao, Maj Earle, Dr. Jana Hall, Susan Donlon, and Gwen Ramelb. We also thank personnel at the Hawaii Tumor Registry, Kaiser-Permanente Medical Center, Kuakini Medical Center, Straub Clinic and Hospital, Queen’s Medical Center, and St. Francis Medical Center for their assistance.