| Japanese Journal of Clinical Oncology | Pages |
Non-traditional Study Designs in the Analysis of Gene-Environment Interactions
Tomotaka Sobue, Cancer Information and Epidemiology Division, National Cancer Center Research Institute, Tokyo, Japan
Study designs targeting high-risk families for particular cancers, such as linkage analysis or segregation analysis, have successfully revealed several rare germline mutations with high penetrance. A notable example is the intense search for breast cancer genes in high-risk families. Using linkage analysis in families with multiply affected members followed by direct sequencing of the gene, investigators have identified a gene on chromosome 17 (BRCA1) (1). Although very important for understanding the mechanism of carcinogenesis and for the genetic counseling of the affected family members, these findings have rather limited impact on society owing to their small attributable risks.
To make the social impact more substantial, genetic susceptibility with high prevalence alleles would be more important, although these are usually responsible only for small relative risks. Studies targeting high-risk families will not be suitable for this purpose. Alternatively, association studies, which are based on a comparison of a large number of affected and unaffected individuals from the unrelated general population, can provide sufficient power to distinguish slight variations in cancer risks. Since association studies use just the same approach as used in conventional epidemiological studies, the boundary between genetic and conventional epidemiology is now becoming less clear.
For rare germline mutations with high penetrance, study design and analysis are straightforward since the mutation at a single gene is the only variable influencing predisposition to cancer. For polymorphism with high prevalence, however, the risk of cancer is the result of a complex combination of polygenic and environmental factors, hence special attention is needed to disentangling them.
First, although the term `interaction' is frequently used as having a broad meaning, it has a much narrower meaning in epidemiological settings and needs to be defined. Interaction is a model-dependent concept. For example, gene-environment interactions exist if the joint effect of genetic factors and the environmental exposure differs from the product of the risks of the individual factors on a multiplicative model and of the sum of the background disease rate and the excess rates for the environmental exposure and for the genetic factors on an additive model. In most previous studies, interaction is defined on according to a multiplicative model, mainly because of its convenient statistical properties. The definition of interaction can be applied to a combination of any risk factors.
In non-genetic epidemiological studies, which analyze the effects of several life-style factors simultaneously, interaction is evaluated basically by stratification. To date, however, most epidemiological studies could not deal with interaction sufficiently because of the limited sample size, relative to the number of risk factor variables to be considered. In the usual analysis, priority is set on the evaluation of the dose-response relationship assuming no interaction. Therefore, interaction has been intensively investigated only for the established risk factors with strong association, such as smoking and alcohol for esophageal cancer.
In genetic epidemiological studies, since genetic and environmental factors can be clearly separated, interaction between them is of primary concern and non-traditional study designs have been developed for this purpose. A case-only study reported by Hamajima et al. (2) in this journal is one example. This case-series design was originally introduced by Begg et al. (3) and Piegorsch et al. (4) for assessing genetic susceptibility and has been reviewed by several investigators (5-7). In this method, investigators use case subjects only to assess the magnitude of the interaction between the exposure and genotype. Therefore, sample size will be less than half and the estimated odds ratios will not suffer from potential biases related to control selection. Moreover, as Hamajima et al. (2) indicated, along with the previous papers by Piegorsch et al. (4) and Yang et al. (8), case-only studies can result in even greater precision in estimating interactions (i.e. smaller standard errors) than can be obtained by traditional case-control studies. On the other hand, a case-only study has several assumptions and limitations as indicated in the paper (2). One major limitation is the inability to measure individual genotype and environment effects. Therefore, application should be limited to the environmental factors for which effect has been well established, such as smoking. Also, since effect due to genotype alone is of great concern when identifying patterns of interaction, it can be used only for screening purposes. In addition, it has to be assumed that environmental factors and genotype are independently distributed, but this assumption can be confirmed only by taking appropriate controls. Recently, there have been several reports indicating that genotype may regulate life-style factors, such as alcohol drinking (9) and smoking (10). In addition, several methodological points should be considered further, such as the influence of errors of environmental exposure measurement, genotype misclassification, confounding adjustment and dose-response evaluation. Other study designs without taking controls, such as case-parental control study or the affected relative-pair method, have also been proposed in the field of genetic epidemiology (5).
Advances in genetic technology, such as DNA chips and whole gene amplification, and the work of Human Genome Project will provide valuable opportunities for the study of gene-environment interaction to become an integrated part of epidemiological research. A comprehensive search for single nucleotide polymorphism (SNPs) on several important genes will further promote the methodological aspects of how to evaluate gene-environment interaction precisely and efficiently.
Similar new study designs that involve only cases as subjects have appeared not only in genetic epidemiology but also in other fields, such as case-crossover studies for time trends and case-specular studies for the effect of an electromagnetic field. As Greenland pointed out (11), however, it is important to recognize that new designs and analysis methods are subject to validity and precision considerations that do not arise in conventional designs. The epidemiological community may need considerable experience with case-distribution (case-only) designs before the strengths and limitations of such designs are fully appreciated.
References
This page is run by Oxford University Press, Gr eat Clarendon Street, Oxford OX2 6DP, as part of the OUP Journals
Comments and feedback: jnl.info{at}oup.co.uk
Last modification: 1 Dec 1999
Copyright© 1999 Foundation for Promotion of Cancer Research.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||