Review Article
Measures of familial aggregation depend on definition of family history: meta-analysis for colorectal cancer

https://doi.org/10.1016/j.jclinepi.2005.07.018Get rights and content

Abstract

Objective

Familial aggregation, a primary theme in genetic epidemiology, can be estimated from family studies based on an index person. The excess risk due to the presence of affected family members can be classified according to whether disease in the relatives is considered a risk factor for the index person (type I relative risk) or whether the disease status of the index person is considered a risk factor for the relatives (type II relative risk).

Study Design and Setting

A meta-analysis of published colorectal cancer studies reporting a measure of familial association was performed and application of multilevel linear regression to model age-specific relative risks presented.

Results

The pooled type I relative risk of colorectal cancer given any affected first-degree relative (based on 20 studies) was 2.26 (95% confidence interval CI = 1.86, 2.73) and decreased with the age of the consultand. The pooled type II estimate (based on seven studies) was 2.81 (95% CI = 2.05, 3.85).

Conclusion

Type I relative risks are useful in clinical counseling settings when a consultand wants to know his/her disease risk given his or her family history. Type II relative risks can be used to quantify the risk of disease to relatives of an affected individual and then identify subjects eligible for screening.

Introduction

Familial aggregation of a disease exists when the disease occurs at a higher frequency in the relatives of an affected person than in the general population [1]. Familial aggregation is a primary theme of genetic epidemiology, a discipline that uses family-based designs to assess the roles of genes and environment in the etiology of disease in the population. Determining how risk of disease is dependent on family history is necessary to develop referral guidelines to services, such as screening and genetic counseling. For complex disorders, such as the common cancers, studies of familial aggregation may contribute to disease taxonomy in that they help identifying characteristics that distinguish hereditary from nonfamilial forms and thereby recognizing causal factors.

The sampling scheme for family studies typically consists of starting with an individual (called the index person or proband) and collecting information about his or her relatives. Note that the index person need not be affected. For example, the case–control–family design [2] consists of sampling cases and controls on a population basis as the index persons, and then sampling their relatives. Reports of cancers, both for the index persons and their relatives, are ideally confirmed by reviewing pathology reports, medical records, and death certificates. Cancer registries can be used to identify affected index persons (case probands) and to verify any reports of cancer in the relatives.

As clarified by Susser and Susser in their seminal article published in 1989 in the American Journal of Epidemiology [1], two different estimates of excess risk are possible when dealing with family data, through (i) treating the presence of disease in relatives as a risk factor and evaluating whether it confers an excess risk by comparing its prevalence in affected versus unaffected index persons (here termed type I relative risk) and (ii) evaluating whether the relatives of an affected index person have an excess risk of disease in comparison to an appropriately defined population (here termed type II relative risk) (Table 1). Type I and type II relative risks are both expressions of the probability that an individual (termed the consultand in genetic counseling terminology) will develop the disease, given the distribution of disease or diseases among his or her relatives. Estimation can be made according to the disease in question, the characteristic of the consultand (e.g., age; sex) and his or her family characteristics (e.g., definition of the disease considered to be risk factors, subsets of relatives whose diseases are evaluated, the age and sex distribution of the relatives, and the age at onset of disease in relatives). Notably, the only situation in which the type I and the type II estimates would be equivalent is when the consultand has only one relative for consideration; for example, (i) risk of prostate cancer in a man with just one brother given his disease status (type I) and risk to the only brother of an affected man (type II); (ii) risk of breast cancer in a woman given that her mother got breast cancer (type I) and risk to the daughters of an affected woman (type II). In practice, the distinction between type I and type II estimates has been ignored [3], [4], [5], [6] and often misunderstood [7], [8]

Traditional epidemiological study designs provide a useful framework for evaluating familial aggregation of disease, provided that some concepts are modified or extended for the collection and analysis of family data. In classical epidemiology, in which unrelated individuals are studied, the most commonly used methods to address associations between diseases and exposures are the case–control design and the cohort design. In family studies, in which the association between a disease and its distribution in pedigrees is studied, the distinction between case–control and cohort designs is not clear, mainly because family history is not an attribute of any single individual, and also because the causal criterion of temporal sequence between exposure and disease does not hold [1]. A direct measure of the relative risk may be obtained by analyzing the data from a cohort study perspective, in which the cohort is composed of initially unaffected individuals, classified as exposed or not exposed according to the disease status of their relatives, followed up prospectively over time. The ratio of the incidence of disease in exposed subjects to that in the unexposed is an estimate of type I relative risk. The type II relative risk is estimated by following up (usually retrospectively) the cohort of relatives for whom the exposure status is defined by the disease status of the index person. This study design is sometimes termed the kin–cohort design. In studies that recruit solely families of affected individuals (i.e., the index person is a case proband), the incidence of disease in the cohort of the probands' relatives is compared with the incidence in the general population (external comparison). In this situation, the relative risk can be measured in the form of a standardized incidence ratio. Conversely, when relatives of both affected and unaffected index persons (i.e., case probands and control probands) are studied, the incidence of disease in relatives of the case proband can be compared with the incidence in those of the control proband (an internal comparison). The same data can also be analyzed from a case–control perspective, in which the relative risk is estimated as an odds ratio of the exposure between affected and unaffected subjects, with the exposure being the presence of disease in relatives, as well as an odds ratio of exposure between affected and unaffected relatives of case probands and control probands (i.e., by a case–control study nested in the cohort of relatives). Under the rare disease assumption, the corresponding odds ratios are a close approximation of type I and type II relative risks.

The type I relative risk is a measure of the excess risk of disease for an individual, given the distribution of disease among his or her relatives (family history). The type II relative risk is a measure of the excess risk of disease for an individual given the disease status of a selected member of his or her family. In estimating the type I relative risk, the sampling of subjects is independent of their exposure status (which is the disease status of the relatives). In estimating the type II relative risk, the sampling procedure is not independent of the exposure (which in this case is the disease status of the index person). A potential problem related to the type II approach is that the cohort is of relatives, and therefore is not of independent individuals whenever it is composed of more than one member of the same family (e.g., cohort of siblings). As a consequence, any standard error estimate based on the assumption of independence within exposure sets will be an underestimate, by an amount dependent on the strength of the familial aggregation and the size and structure of the families within the cohort, and the analysis should be modified accordingly to take the clustering into account.

Colorectal cancer (CRC) is one of the most commonly diagnosed cancers and one of the most important causes of cancer mortality [9], [10], [11]. Given that ∼1 in 20 men and women are diagnosed with CRC in their lifetime, a family history of CRC is also common.

The estimated proportion of CRC that is caused by a dominantly inherited gene ranges from 5% to 13% [12], [13]. Known CRC susceptibility genes include the APC gene associated with familial adenomatous polyposis (FAP) and attenuated familial adenomatous polyposis (AFAP) [14], [15], [16], [17] and the mismatch repair genes, in particular MSH2 and MLH1 (formerly hMSH2 and hMLH1), that are implicated in a large proportion of the hereditary nonpolyposis colorectal cancer (HNPCC) syndrome [18], [19], [20], [21].

The mechanisms leading to CRC are complex and likely involve not only the currently known major genes, but many other genetic and environmental risk factors shared by family members. Evidence for this comes from studies that have shown substantial residual unexplained familial aggregation after excluding families suspected of carrying mutations in mismatch repair genes and in the APC gene [22], [23], [24], [25], [26], [27], [28], [29], [30], [31]. A population-based study conducted on families ascertained via an early-onset case of CRC showed that, after removing families found to be carrying mutations in MSH2 and MLH1 genes, first-degree relatives of CRC cases were still at increased risk of the disease [32].

The first report of familial aggregation of CRC in the general population was made in 1960, when Macklin [33] published the results of a study of 145 patients with CRC, with a greater than expected number of deaths attributable to CRC in first- and second-degree relatives. Many studies of the association between CRC and family history have been published since then; a review and meta-analysis of the literature from 1966 to 1999 appeared in 2001 [34].

Here, we review the publications reporting a measure of familial aggregation of CRC, classify the estimates based on the study design and analytical approach, and estimate the overall type I and type II increased risk by combining the findings of all studies. A method based on multilevel linear regression is also applied to model relative risks as a function of ages.

Section snippets

Search methods

Studies reporting a measure of familial aggregation of CRC based on incidence data were identified through a Medline search employing the following algorithm: (“colorectal neoplasm” or “colon neoplasm” or “rectal neoplasm”) and (“case–control” or “cohort”) and “famil*”, where the wildcard asterisk means that any word starting with “famil” is selected. Additional articles were ascertained by searching the references cited in publications. Studies were limited to those published in English from

Results

Thirty-three articles reporting a measure of familial aggregation of CRC were identified [22], [23], [24], [26], [27], [28], [29], [30], [31], [42], [43], [44], [45], [46], [47], [48], [49], [50], [51], [52], [53], [54], [55], [56], [57], [58], [59], [60], [61], [62], [63], [64], [65]. The main characteristics of the studies are presented in Table 2. Publication dates ranged over three decades: 6 articles were published in the 1980s, with the earliest published in 1982 [22], [28], [47], [48],

Discussion

We have focused on the interpretation of estimates of familial aggregation of disease provided by epidemiological studies. Because of the growing interest in identifying subjects at genetic risk of developing cancer, those measures with a clear clinical meaning have a greater importance for counseling purposes. Type I estimates refer to situations where an individual (consultand) presents at a genetic counseling clinic and wants to know his or her risk of developing a disease given his or her

Acknowledgments

This study was supported by The Cancer Council Victoria and funded by NHMRC program grant ECHIDNAS (Epidemiology of Chronic Disease, Health Interventions and DNA studies) (209057); NHMRC capacity-building grant PLATYPUSES [Platform for young public health researchers to upgrade their scientific training experience and independent status] (251533); NIH Australasian Colorectal Cancer Family Study. (1U01CA97735-01). The contributions of L.B. and G.S. were supported in part by fellowships from the

References (72)

  • P.D. Pharoah et al.

    Family history and the risk of breast cancer: a systematic review and meta-analysis

    Int J Cancer

    (1997)
  • L.E. Johns et al.

    Colorectal cancer mortality in first-degree relatives of early-onset colorectal cancer cases

    Dis Colon Rectum

    (2002)
  • D.W. Bruner et al.

    Relative risk of prostate cancer for men with affected relatives: systematic review and meta-analysis

    Int J Cancer

    (2003)
  • L.E. Johns et al.

    A systematic review and meta-analysis of familial prostate cancer risk

    BJU Int

    (2003)
  • M.J. Khoury et al.

    Bias in using family history as a risk factor in case–control studies of disease

    Epidemiology

    (1995)
  • J.L. Hopper

    Different, not biased

    Epidemiology

    (1996)
  • D.M. Parkin et al.

    Estimates of the worldwide incidence of 25 major cancers in 1990

    Int J Cancer

    (1999)
  • P. Pisani et al.

    Estimates of the worldwide mortality from 25 cancers in 1990

    Int J Cancer

    (1999)
  • D.M. Parkin et al.

    Estimating the world cancer burden: Globocan 2000

    Int J Cancer

    (2001)
  • L.A. Aaltonen et al.

    Replication errors in benign and malignant tumors from hereditary nonpolyposis colorectal cancer patients

    Cancer Res

    (1994)
  • D.T. Bishop et al.

    The genetics of colorectal cancer

    Cancer Surv

    (1990)
  • H. Bussey

    Familial polyposis coli: family studies, histopathology, differential diagnosis, and results of treatment

    (1975)
  • S. Bulow

    Familial polyposis coli

    Dan Med Bull

    (1987)
  • C. Boland

    Hereditary nonpolyposis colorectal cancer

  • H.T. Lynch et al.

    Hereditary nonpolyposis colorectal cancer (Lynch syndrome): an updated review

    Cancer

    (1996)
  • L. Bonelli et al.

    Family history of colorectal cancer as a risk factor for benign and malignant tumours of the large bowel: a case–control study

    Int J Cancer

    (1988)
  • M.C. Boutron et al.

    Family history of colorectal tumours and implications for the adenoma-carcinoma sequence: a case control study

    Gut

    (1995)
  • A.N. Freedman et al.

    Familial and nutritional risk factors for p53 overexpression in colorectal cancer

    Cancer Epidemiol Biomarkers Prev

    (1996)
  • N.R. Hall et al.

    Genetic susceptibility to colorectal cancer in patients under 45 years of age

    Br J Surg

    (1994)
  • R.A. Kerber et al.

    Risk of colon cancer associated with a family history of cancer or colorectal polyps: the diet, activity, and reproduction in colon cancer study

    Int J Cancer

    (1998)
  • K. Kotake et al.

    Relation of family history of cancer and environmental factors to the risk of colorectal cancer: a case–control study

    Jpn J Clin Oncol

    (1995)
  • G.A. Kune et al.

    The role of heredity in the etiology of large bowel cancer: data from the Melbourne Colorectal Cancer Study

    World J Surg

    (1989)
  • M.P. Longnecker

    A case–control study of alcoholic beverage consumption in relation to risk of cancer of the right colon and rectum in men

    Cancer Causes Control

    (1990)
  • D.J. St John et al.

    Cancer risk in relatives of patients with common colorectal cancer

    Ann Intern Med

    (1993)
  • B.M. Stephenson et al.

    Frequency of familial colorectal cancer

    Br J Surg

    (1991)
  • Cited by (79)

    • Approach to Familial Predisposition to Colorectal Cancer

      2022, Gastroenterology Clinics of North America
      Citation Excerpt :

      Individuals with a family history of CRC are at increased risk for developing CRC; the magnitude of risk depends on the number of relatives with CRC, whether the relatives are first-degree relatives (FDR) versus second-degree relatives, and the age at which their relatives were diagnosed with CRC. Studies have reported an estimated 2-fold increase in risk of CRC in individuals with 1 or more FDR with CRC compared with individuals without a family history.1–5 Furthermore, the age of CRC diagnosis in a relative affects the risk of CRC.

    • Familial colorectal cancer

      2022, Best Practice and Research: Clinical Gastroenterology
      Citation Excerpt :

      Thus, the pooled RR among cohort studies was 1.67 (95% CI 1.52–1.82) and the pooled RR among case-control studies was 2.22 (95% CI 2.00–2.48) in the presence of at least one FDR with CRC [10]. When evaluating the risk of having two or more FDRs, the RR estimates have also decreased among meta-analyses over the years, remaining close to 3 [6–10] (Table 1). Regarding the age of the affected relative, individuals with ≥1 FDR with CRC at an age younger than 60 years have been reported to have a doubled risk of developing CRC [10], and a tripled risk when the affected relative was diagnosed before 50 years of age [7,9,10].

    • Prevalence of a First-Degree Relative With Colorectal Cancer and Uptake of Screening Among Persons 40 to 54 Years Old

      2020, Clinical Gastroenterology and Hepatology
      Citation Excerpt :

      In line with those earlier reports, data from the RaPS study showed that far fewer people who should undergo CRC screening have done so, even though higher rates of a history of a colonoscopy were obtained than originally anticipated. This is worrying because it has been shown consistently in numerous publications that a FH of CRC increases the risk for CRC,4,28,29 especially for people with an affected FDR diagnosed at younger ages,4 and younger people altogether30 (our data suggest that both groups overlap widely). Although CRC is a disease that is associated especially with increasing age, there is strong evidence that FH is offsetting the decreased risk in younger ages.30

    View all citing articles on Scopus
    View full text