Cancer Risk in Relatives of Patients with Common Colorectal Cancer

Methods

The study used a matched casecontrol design [34] to compare the occurrence of cancer in relatives between patients with common colorectal cancer (case patients) and patients without cancer (control patients). Case and control patients were selected from one surgeon's practice and were matched by gender, age (within 5 years and usually within 2 years), and country of birth or upbringing.

Patients

Case patients were selected from a series of more than 2400 patients with colorectal cancer treated between 1952 and 1985. All 576 patients alive at the time of the survey (June 1985 to May 1986) and residing in Australia were reviewed for eligibility. Detailed clinical, surgical, and pathologic information was available for every patient [35, 36]. Exclusion criteria were 1) colorectal cancer associated with ulcerative colitis, Crohn disease, familial adenomatous polyposis, or recognized HNPCC [37]; and 2) referral to the surgeon because he previously had treated a relative or spouse for colorectal cancer or adenoma. The minimum criteria for diagnosis of the HNPCC were three affected first-degree relatives with at least two relatives having colorectal cancer diagnosed before 50 years of age, or multiple primary cancers, or colorectal cancer plus another syndrome cancer [38]. Control patients were selected from patients who had surgery for nonmalignant disease during the same period and had no known history of neoplastic disease.

Data Collection

Questionnaires were mailed concurrently to case patients and their matched control patients. Accompanying letters made case and control patients equally aware that the study was about family history of colorectal cancer. The questionnaires covered personal information; information about first-degree relatives (parents, siblings, and children surviving to the age of 12 years), including year of birth, occurrence of colorectal cancer, age at diagnosis, occurrence of other types of cancer, age at death, and cause(s) of death when relevant; and information about second-degree relatives and spouses. Each case patient and control was contacted by telephone to review responses and to assess the degree of certainty about listed diagnoses. Care was taken to question case and control patients to the same degree. In many instances, responses were reviewed during a subsequent visit. Close relatives were encouraged to help collect information.

All questionnaires were masked to prevent identification of their casecontrol status and were then assessed by an independent gastroenterologist to classify each relative as evaluable or nonevaluable according to whether information about his or her medical history was available up to the time of death or, in the case of living relatives, up to the time of completion of the questionnaire. Half-brothers and half-sisters were excluded. Many subjects were contacted again at this stage.

Verification of diagnoses was sought for all reported cases of bowel cancer, other abdominal cancer, and abdominal surgery of unspecified type using hospital or physician records, Australian state cancer registries, and death registries. Colorectal cancers in relatives were classified as definite when confirmation of the diagnosis was obtained from a medical source; as probable when relatives were able to provide convincing information about the diagnosis (such as name of the surgeon and hospital as well as details about the surgical procedure) but where medical confirmation could not be obtained; and as possible when relatives were stated to have had colorectal cancer but when the evidence was less convincing.

Final Selection of Patients for the Study

Of the 576 patients with colorectal cancer alive at the time of the survey, 20 were excluded because cancer was associated with ulcerative colitis, Crohn disease, or familial adenomatous polyposis and 22 were excluded because of failure to complete questionnaires (15 had terminal illness, 5 refused, and 2 could not be contacted). Five potential control patients failed to complete questionnaires, and suitable replacements were recruited. Eleven case patients (and their respective control patients) were classified as ineligible and were excluded from the study; two proved to be members of families with recognized HNPCC, and the other nine had been referred to the surgeon because he previously had treated a relative or spouse for colorectal cancer. The research protocol was approved by the Ethics Review Advisory Committee at Alfred Hospital, Melbourne, Australia.

Statistical Analysis

Case-Control Analysis

The data were first analyzed as a casecontrol study, with exposure defined by the presence of disease among (subsets of) relatives. The proportion of case patients with affected relatives was compared with the proportion of control patients, with the odds ratio and standard error calculated by standard methods using logistic regression [39].

Risk to Relatives of a Case Patient Compared with Relatives of a Control Patient

A cohort analysis of disease in relatives, which takes into account the matching of casecontrol pairs and the number of relatives and allows for relatives to be dependent in their colorectal cancer status (as is the case if true familial aggregation exists for colorectal cancer), was also done [40].

For each case patient, i, let there be n_i relatives of a predetermined type (for example, mother, parent, or sibling), of whom x_i also have colorectal cancer. Let the matched control patient, i, have m_i such relatives of whom y_i have colorectal cancer. Let t_i = x_i + y_i and N_i = n_i + m_i and the true odds ratio, OR_i, for cancer in a relative of a case patient compared with cancer in a relative of a control patient be defined as (E[x_i]E[m_i y_i])/(E[y_i]E[n_i x_i]). Suppose a true common odds ratio exists; for example, OR_i = OR for all i. The Mantel-Haenszel estimator is OR_MH = R_i/S_i, where R_i = x_i(m_i y_i)/N_i and S_i = y_i(n_i x_i)/N_i. It is asymptotically normal, even if nonindependence exists within sets of relatives. A consistent and robust test statistic for H_o: OR = 1 is

EQUATION OMITTED

which has an approximate chi-square distribution with one degree of freedom, independent of the underlying true distribution for the {x_i} and {y_i}. Because log(OR_MH) has more nearly a normal distribution than OR_MH, an approximate test-based standard error, SE, for log(OR_MH) was calculated by assuming (log[OR_MH]/SE)² = T. An approximate asymmetric 95% CI for OR_MH is (e^log[ORMH sup] 1.96SE, e^log[ORMH sup] + 1.96SE).

The same analysis was applied to subsets of casecontrol pairs selected according to a predetermined characteristic of the case patients (for example, age of onset less than 45 years).

Analysis of Cumulative Incidence among Relatives of (Subsets of) Case and Control Patients

Parents, siblings, and children were pooled. A Kaplan-Meier survival curve, S(t), was calculated for time free of colorectal cancer, t, based on age at diagnosis in relatives [41]. (Age at death was treated as a censoring time.) Cumulative incidence to age, a, was then defined as 1 S(a) and was plotted for 10-year increments in age. Comparison of survival, and therefore of cumulative incidence, between groups was evaluated by the two-tailed generalized Wilcoxon test according to Gehan [42].

Results

Characteristics of Case and Control Patients

The study group comprised 523 cases (250 men, 273 women; mean age, 71.2 years [SD, 10.3 years]), and 523 gender-matched control patients (mean age, 69.9 years [SD, 10.2 years]). Of the case patients, 426 (81%) were born in Australia. Thirteen (2.5%) case patients had been treated for synchronous cancers of the large bowel and 30 (5.7%) for metachronous cancers. The primary lesions were in the proximal colon (proximal to the splenic flexure) in 83 case patients, in the distal colon in 148, and in the rectum or at the rectosigmoid junction in 338. Colorectal cancer was diagnosed before age 45 years in 82 (16%) case patients, between ages 45 and 54 years in 147 (28%) patients, between ages 55 and 64 years in 173 (33%), and at 65 years in 121 (23%). A diagnosis of colorectal cancer was made at least 5 years before the survey in 436 (83%) case patients.

Characteristics of Relatives

The case and control patients reported a total of 7530 first-degree relatives, of whom all but 37 were evaluable. This left 3857 relatives of case patients (mean, 7.4) and 3636 relatives of control patients (mean, 7.0) available for study. Case and control patients did not differ in number or mean age of relatives for each category of relationship (Table 1). The cancer and control groups had 508 and 507 evaluable spouses, respectively. In contrast to first-degree relatives and spouses, information obtained about second-degree relatives was often deficient, mainly due to death of grandparents before probands were born or to the impact of migration on communication within families. Only 62% of grandparents of case patients and 66% of control patients were evaluable, and therefore the data on second-degree relatives was not subjected to detailed analysis.

Colorectal Cancer in Relatives

Of the total of 122 first-degree relatives with definite colorectal cancer (84 relatives of case patients, 38 relatives of control patients), cancer verification was based on pathology reports in 40 instances, on medical reports of clinical details and operative findings in 31, on cancer registry information in 31, and on death certificates in 20. These included 12 instances in which first-degree relatives were reported to have had abdominal surgery for some other reason; further inquiry indicated a true diagnosis of colorectal cancer. The misleading information came from 11 case patients and one control patient. Although it was reported that a first-degree relative had had colorectal cancer in 49 other instances, inquiry indicated another diagnosis (false-positive rate = 32%; SE = 4%). Correct diagnoses included rectal prolapse, complicated diverticular disease, Crohn disease, leiomyosarcoma involving the colon, small bowel tumors, and gastric cancer. This misleading information came from 24 case and 25 control patients.

Twenty-three more first-degree relatives had probable colorectal cancer (19 relatives of case patients, 4 relatives of control patients), and 9 more had possible colorectal cancer (4 relatives of case patients, 5 relatives of control patients). All cancers reported in relatives were symptomatic; no cancer was detected by screening. Subsequent analyses were confined to definite and probable colorectal cancer.

A diagnosis of definite or probable colorectal cancer was accepted for 24 fathers, 25 mothers, 28 brothers, 19 sisters, 3 sons, and 4 daughters of case patients, and for 11 fathers, 10 mothers, 11 brothers, 9 sisters, and 1 son of control patients. For spouses, 7 (1.4%) had colorectal cancer in each group.

Case-Control Analysis

Of the 523 case patients, 85 (16%) had one or more first-degree relatives with definite or probable colorectal cancer, compared with 39 (7.5%) control patients (odds ratio, 2.1; CI, 1.4 to 3.0; P < 0.001). Of these, 67 and 36, respectively, had only 1 affected relative (odds ratio, 1.8; CI, 1.2 to 2.7; P < 0.01), and 18 and 3, respectively, had 2 affected relatives (odds ratio, 5.7; CI, 1.7 to 19.3; P < 0.01). Therefore, the risk for colorectal cancer increased with the number of affected relatives (P < 0.001 for trend), with two affected relatives being associated with a greater risk than only one (P = 0.05).

Review of the 18 families of case patients with 2 affected relatives failed to show diagnostic features attributable to HNPCC or any known history of colorectal cancer among second-degree relatives or earlier generations. In two of these families, both parents of the case patient had had colorectal cancer.

Matched Analysis of Cancer Risk in Parents and Siblings

Table 2 shows that the odds ratio for colorectal cancer was significant (P 0.05) for fathers, for mothers, for parents, for brothers, for siblings, and for parents and siblings combined, but it was not significant for sisters. No significant difference existed between the odds ratio for fathers compared with mothers, for brothers compared with sisters, or for parents compared with siblings. The estimate for parents and siblings of 2.1 was the same as that given by the casecontrol analysis for colorectal cancer associated with one or more affected first-degree relatives.

Table 3 shows that the odds ratio for colorectal cancer in parents or siblings of case patients with age of onset before 45 years was 3.7 (CI, 1.5 to 9.1; P < 0.001). This was marginally greater than for case patients with age of onset 45 years (P = 0.2). No detectable difference in odds ratio existed according to gender, site of cancer, or type of cancer (single versus multiple). A significant odds ratio (P 0.05) was evident for single and for multiple cancers, and with each site of cancer, except proximal colon for which only 17 informative casecontrol pairs existed.

For case or control patients who had at least one affected relative, the odds ratio for an additional affected relative was 4.7 (CI, 1.5 to 14.5; P 0.001).

Cumulative Incidence of Cancer in First-Degree Relatives

Table 4 shows the cumulative incidence of colorectal cancer to ages 40, 50, 60, 70, and 80 years among first-degree relatives of control patients, of case patients, and of case patients according to whether the age at diagnosis was < 55 years or 55 years. Eighteen relatives of case patients but only one relative of control patients had colorectal cancer before age 50 (P = 0.01). Risk to relatives was greater for case patients than for control patients (P 0.001) and was greater for case patients diagnosed at < 55 years compared with case patients diagnosed at 55 years (P < 0.001). Risk to relatives of case patients diagnosed at < 45 years was greater than for case patients diagnosed at 55 years (P < 0.001) and was marginally greater than for case patients diagnosed between the ages of 45 and 54 years (P = 0.07) (Figure 1). No evidence existed that cumulative incidence depended on the gender of case patients, site of cancer, or type of cancer.

Other Cancer in Relatives

Information was obtained on 28 other categories of cancer. No significant difference (P > 0.05) existed in overall occurrence of noncolorectal malignancy between the two groups of relatives; 313 of the 3857 relatives (8.1%) of case patients had cancer at sites other than the large bowel compared with 277 of the 3636 relatives (7.6%) of control patients. For individual types of cancer in relatives, the only significant difference (P < 0.05) occurred with cancer of the prostate, which was more common for case patients compared with control patients (odds ratio, 2.8; CI, 1.2 to 6.7; P < 0.02) and was independent of the increased amount of colorectal cancer in relatives.

Discussion

Our study provides evidence for a modest (twofold) increase in risk for colorectal cancer in first-degree relatives of patients with common colorectal cancer, a result that is similar to that reported in several other studies [17-19, 24-27]. The odds ratio for colorectal cancer was 1.8 for patients having one affected relative and was 5.7 for patients having two affected relatives, results that support the view that the risk is higher in the latter situation [15]. These magnitudes of risk are consistent with one or more strong familial risk factors [43].

In addition, we found a greater cumulative risk for colorectal cancer in relatives of case patients diagnosed before age 55 years, with a trend toward an even higher risk associated with diagnoses before age 45 years (Figure 1). Support for this association also comes from the high odds ratios for colorectal cancer in relatives of case patients diagnosed before age 45 years. Lovett [19] has suggested that relatives of patients diagnosed before age 40 years might have a greater risk of death from colorectal cancer, although she studied only 209 families of case patients, with no control patients. Patients who had an early age of cancer diagnosis or who had multiple relatives with colorectal cancer could have clinical expressions of the same risk factor(s).

We were unable to detect differences in familial risk associated with gender, site, or type (single or multiple) of cancer. The results for site of cancer support the observations of Sondergaard and colleagues [27], who found no relation between site of cancer and risk for parents.

Our study had several strong points, including access to detailed clinical and pathologic records on all cases, evaluation of questionnaires without knowledge of their casecontrol status, and the unbiased intensity of the effort taken to verify diagnoses. Almost all first-degree relatives and spouses were evaluable, and comprehensive information was available about their medical histories. With reported diagnoses, the false-positive rate was 32%, placing doubt on the reliability of other studies when information was obtained at a single interview and when no attempt was made to verify diagnoses. The relatives of our case and control patients were well matched in number and age structure for each category of relationship. Furthermore, the size of the study enabled reasonably precise estimates of risk to be made. None of the diagnoses of colorectal cancer in relatives were established as a result of the screening because, up to the time of the survey, surgeons in Melbourne were only advocating screening of relatives for familial adenomatous polyposis.

Close attention was given to nullifying potential biases. We did not attempt to obtain family medical histories for deceased patients because information about a patient's first-degree relative is more reliable when obtained from the patient than from a surviving spouse or child [44]. We therefore cannot exclude the possibility of some bias if the strength of familiality depends on survival time. The comparable method of approach to case and control patients as well as the verification of diagnoses using medical records and death certificates were designed to minimize the effects of recall bias.

We focused on common colorectal cancer [45], often referred to as sporadic colorectal cancer. Classification of patients as having either common colorectal cancer or HNPCC is an important issue because no distinctive phenotypic features of HNPCC exist [37]. In the families of case patients with two affected first-degree relatives, the anatomic distribution, ages at diagnosis, and types of colorectal cancer were all consistent with classification as common colorectal cancer. The absence of any history of colorectal cancer in more distant relatives provided further support for our conclusion.

Our study provides more comprehensive information than earlier studies about cancer risk in relatives of patients with common colorectal cancer. The first studies on the subject examined death certificates of deceased first-degree relatives and concluded that the risk for cancer was increased between 3.2 and 3.5 times [17-19]. Woolf [17] included control patients and Macklin [18] and Lovett [19] compared observed numbers of cancer deaths with expected frequency for the population. Studies reported during the last decade have investigated occurrence of cancer in living as well as in deceased relatives [20-27], but most of the studies had some unsatisfactory design components. Problems included small sample size [20], collection of data about relatives at a single interview without giving patients an opportunity to consult with other family members [23-26], different methods of approach to case and control patients [25], failure to verify stated diagnoses in relatives [24, 26], and failure to show that relatives of case and control patients were comparable in number and age structure [20-26]. One careful study confined observations to case patients diagnosed under 60 years of age [27], and this and another study limited examination of cancer risk to parents [21, 27].

As reported previously [23, 27], the overall occurrence of noncolorectal malignancies in relatives did not differ between case and control patients. With individual cancers, an excess of prostatic cancer occurred in relatives of case patients. No excess of elderly men existed among relatives of case patients. Considering that 28 different categories of malignancy were examined, the excess of prostatic cancer could have arisen by chance, but it could be explained by risk factors, including diet, common to the two cancers.

The earlier onset of colorectal cancer provides a basis for recommending that relatives of patients should enter screening programs at age 40 years [16], a decade earlier than for average-risk persons [46]. The absolute risk for cancer increased progressively with advancing age and at a greater rate than for relatives of control patients. This result contrasts with that found with HNPCC and familial breast cancer, in which excess risk is greatest in middle age [47, 48]. Strategies for secondary prevention of colorectal cancer should take the progressive increase in risk with advancing age into account by providing surveillance over several decades of life. Recommendations about screening should be based on absolute risk as well as on the accuracy, efficacy, and acceptability of the screening method(s).

Presented in part to the General Scientific Meeting, Royal Australasian College of Surgeons, Brisbane, May 1988 and to the Annual Meeting of the Gastroenterological Society of Australia, Sydney, 9-11 May 1988. Published in abstract form in Aust NZ Med J. 1988; 18:417.

Abbreviation

HNPCC: hereditary nonpolyposis colorectal cancer