© 1999 by Oxford University Press
Journal of the National Cancer Institute Monographs, No. 26, 49-54,
1999
© 1999 Oxford University Press
II. GENE CHARACTERIZATION PANEL |
Detection of Interaction Involving Identified Genes: Available Study Designs
Affiliations of authors: A. M. Goldstein, Genetic Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, MD; N. Andrieu, Unité de Recherche en Epidémiologie des Cancers, Institut de la Santé et de la Recherche Médicale U351, Institut Gustave-Roussy, 94805 Villejuif, France.
Correspondence to: Alisa M. Goldstein, Ph.D., National Institutes of Health, Executive Plaza South, MSC 7236, 6120 Executive Blvd., Bethesda, MD 20892-7236 (e-mail: (ag26o{at}nih.gov).
 |
ABSTRACT |
|---|
 Top Abstract IntroductionGene-Environment InteractionGene-Gene InteractionConclusionsReferences |
|---|
Advances in molecular genetic techniques have led to an increased ability to examine gene-environment interactions. Studies to detect gene-environment interactions are motivated by different situations, including 1) most identified cancer genes having associated lifetime risks less than 100% (i.e., incomplete penetrance), 2) hereditary factors that control the metabolism of carcinogens that may modulate risk of disease as hypothesized in pharmacogenetics, and 3) inconsistent associations across studies between a cancer and a suspected risk factor. The above situations and others have led to increased study of interaction between genetic and environmental factors. Less studied so far, but with increased potential for the future, is interaction between identified genes. Gene-gene interaction studies would also be motivated by the situations described above. Approaches to detect gene-environment and gene-gene interactions are reviewed. Available risk estimates, required types of subjects, and feasibility of the proposed study designs are discussed; efficiency and power for interaction assessment are summarized where available. In general, most designs allow for estimating risk associated with a genetic factor, environmental factor, and interaction effect. Although power and efficiency for detecting interactions have been assessed for specific situations in some of the methods, further investigations are needed to define the efficiency spectra of each design.
|
INTRODUCTION |
|---|
|
TopAbstractIntroductionGene-Environment InteractionGene-Gene InteractionConclusionsReferences |
|---|
Before looking at different study designs, we need to define interaction, since interaction is a model-dependent concept. We define the parameters for modeling the exposure-disease relation. Let P(g) = frequency of the genetic factor g in the population and P(e) = frequency of the environmental factor e in the population.
Assume OReg is the odds ratio for exposure e and genetic status g, with e = 1 denoting exposed and e = 0 denoting nonexposed and with g = 1 denoting the presence of the genetic factor and g = 0 denoting the absence of the factor. By definition, OR00 = 1.
A fundamental component of the exposure-disease relation is the specification of the joint OR, OR11, under the null and alternative hypotheses. Suppose I00, I10, I01, and I11 are the disease incidence rates (e.g., per person per year) for nonexposed without the genetic factor (e = 0 and g = 0), for exposed without the genetic factor (e = 1 and g = 0), for nonexposed with the genetic factor (e = 0 and g = 1), and for exposed with the genetic factor (e = 1 and g = 1), respectively. We wish to define I11 in terms of the other rates. One characterization of the joint association is additive, where I11 is the sum of the background disease rate and the excess rates for the exposure and for the genetic factor, i.e., I11 = I00 + (I10 - I00) + (I01 - I00) = I10 + I01 - I00.
Dividing by I00, this expression can be rewritten in terms of relative risks, as I11/I00 = I10 /I00 + I01/I00 - 1. For rare diseases, this expression is approximated by ORs, namely, OR11 = OR10 + OR01 - 1.
An alternate characterization for the joint association is multiplicative, where I11 is the product of the risks for the individual factors, i.e., I11 = I10 x I01/I00. Again, dividing each side of the equation by I00 and assuming rare diseases, the multiplicative association is approximated by OR11 = OR10 x OR01. The choice of the joint association will depend on many factors, including the overall goals of the study.
Departures from multiplicativity can be assessed by the interaction effect (1), which is the joint OR for the exposure and the genotype divided by the product of the ORs for the effect of exposure alone and of the genotype alone
 |
Departures from an additive model of risk may also be assessed by dividing the joint ORs for the exposure and the genotype by the sum of the ORs for the effect of exposure alone and of the genotype alone minus 1
 |
An interaction effect of more than one indicates a greater than multiplicative/additive effect between the exposure and the genotype, while an interaction effect of less than one indicates a less than multiplicative/additive effect.
|
GENE-ENVIRONMENT INTERACTION |
|---|
|
TopAbstractIntroductionGene-Environment InteractionGene-Gene InteractionConclusionsReferences |
|---|
Case-Only Study
In case-only studies, the association between an environmental exposure and a genotype is examined among case subjects only (affected subjects with a given disease). Case subjects are selected using epidemiologic principles of case selection as for any case-control study, with the exposure effect assessed among only case subjects. Case subjects without the susceptibility genotype form the control group and the nonexposed case subjects (with respect to the environmental exposure) serve as the referent group. ORs and confidence intervals are obtained using standard crude analyses or multivariate models to adjust for other covariates (2-4). To yield valid estimates of interaction between a gene and environmental exposure, the study design requires that the exposure and genetic factor occur independently and that the disease is rare. Indeed, ORs relating exposure and the genotype among case subjects only, ORcase-only, have been shown to be a function of ORint estimated from a standard case-control study and OR among control subjects relating the exposure and the genotype (2,4). Under the assumption of independence of exposure and genotype, the latter OR is equal to 1. If the assumption of independence is valid and the disease is rare, the OR (ORcase-only) is the interaction effect (4) as measured in a traditional case-control study under a multiplicative model [for more details, see (3)].
Case-only studies offer better precision for estimating gene-environment interactions than those based on both case and control subjects (subjects not affected by the disease under study) (i.e., there are smaller standard errors due to elimination of control group variability) (2). The power for detecting gene-environment interactions in case-only studies is comparable to the power for assessing a main effect in a classical case-control study (5).
The main disadvantages of the case-only design are that the main effects of the genetic and environmental factors cannot be estimated and that the genetic and environmental factors must occur independently. In addition, even if many biologically plausible gene-environmental interaction models should cause departures from multiplicative effects (6), case-only studies cannot detect gene-environmental interaction models with departures from additivity. For these reasons, Umbach and Weinberg (7) recently proposed an alternative method that keeps the advantages of the case-only design while simultaneously allowing estimation of the main effects. The proposed incomplete-data-case-control design collects both genotype and environmental exposure data from the case subjects but only environmental exposure or only genotype data from the control subjects. The estimation of main effect(s) is possible only if the genotype and environmental exposures are independent and the studied disease is rare. Umbach and Weinberg (7) proposed a maximum likelihood method based on log-linear models, which allows imposition of the independence assumption, whereas usual logistic regression does not permit such an imposition. The authors showed that their method may need fewer than half as many individuals as methods that do not impose the gene-environment independence assumption to reach the same power for detecting gene-environment interaction. This approach allows estimation of interaction that is a departure from either multiplicativity or additivity in the relative risk. However, in case-only and incomplete-data- case-control designs, the required assumption of independence between a gene and environmental factor may be violated when the exposure and the gene both vary with an unmeasured variable. To ensure the validity of the critical independence assumption, a random sample of control subjects with both genetic and exposure data should be collected. This additional requirement may lead to a questionable gain in efficiency for this design compared with a full case-control design.
Case-Control Study Design Using Unrelated Control Subjects
The case-control study with population-based, hospital-based, or other unrelated control subjects is the most commonly used design in studies of gene-environment interaction. By use of unexposed subjects with no susceptibility genotype as the referent group, ORs for all other groups can be estimated under either multiplicative or additive models. Adjustment for potential confounding variables may be accomplished by using stratification or multivariate approaches.
The power of this traditional case-control study design for assessing interaction was first studied examining interaction between unspecified risk factors (1,8,9) and then focusing on gene-environment interaction (10-12). Among the parameters needed to determine the power to detect a gene-environment interaction, the frequencies of the environmental exposure and the genetic factor appeared to be the most important in determining the number of case subjects required. In fact, results from multiple studies (10-12) suggest that case-control designs may be used to detect gene-environment interaction only when the environmental factor and genetic factor are common. Indeed, certain combinations of ORs and frequencies of the genetic factor and the environmental factor make study sizes prohibitive.
The main advantage of this design is that the main effects of the environmental exposure and genetic susceptibility as well as their interactive effect may be estimated. The main disadvantages are that this design may not be appropriate for the study of gene-environment interaction involving rare genes or uncommon environmental exposures (assuming reasonable values of the interaction effect). In addition, population stratification or genetic admixture might adversely influence this design. The extent of the problem produced by population stratification is discussed elsewhere (13).
Two-Stage Case-Control Study: Countermatching and Balanced Design
To detect a gene-environment interaction when one of the factors under study is rare, one approach would be to conduct a two-stage or multistage design that increases the numbers of case and control subjects with the rare factor of interest without prohibitively increasing the number of measurements to perform (14).
For two-stage study designs, samples of case and control subjects are drawn from the population at risk. After classification according to an exposure of interest, subsamples of case and control subjects are selected for purposes of covariate assessment. Approaches include balancing the numbers of exposed and unexposed case and control subjects as well as countermatching in which control subjects are countermatched rather than matched to case subjects. The goal of both approaches is to improve the statistical efficiency of exposure risk estimates compared with the classical random case-control study (15,16).
Countermatching designs. Countermatching has been recently proposed by Langholz and Clayton (16) as a method of sampling controls from a cohort for nested case-control studies. In gene-environment interaction assessment, one scenario for countermatching would be to use a surrogate of the disease gene such as family history, which is generally available on the entire cohort/population at risk. In contrast, a measure of the gene itself would likely be too costly to obtain for the whole cohort. Assuming the disease is rare, all case subjects from the cohort/population at risk would be sampled. Each case subject's risk set would be stratified by the surrogate of the gene, i.e., family history of disease, and control subjects for that risk set would be selected from the strata other than the case subject's stratum. The goal of countermatching is to maximize the number of discordant case-control pairs from which information comes in a matched case-control study. The gain in efficiency will depend on how predictive family history of disease is for the gene of interest. Alternatively, one could countermatch on a rare environmental exposure or on both the environmental factor and the genetic one. A partial likelihood has been developed to estimate exposure effects in countermatching (17) using weighting that takes into account the probability that subjects were selected from specific strata. This method has been shown to increase the efficiency of main effect estimation by approximately 25% compared with classical random sampling (18). In gene-environment interaction assessment, countermatching on both the gene and environmental factor has been shown to be more efficient than a standard nested case-control study using three control subjects per case as well as designs that countermatched on either the environmental factor or the gene. The relative efficiencies were influenced mainly by the frequencies of the true risk factors and the sensitivity and specificity of the factors' surrogates (Andrieu N, Goldstein AM, Thomas DC, Langholz B: manuscript submitted for publication). Studies to examine additional scenarios and evaluate cost, efficiency, and feasibility need to be conducted to determine the utility of these approaches for interaction assessment.
Balanced designs. A more intuitive approach to studying a rare factor would be to oversample for the rare covariate of interest. If the gene or environmental factor under study is rare, rather than choose a subset at random, case and control subjects may be sampled nonrandomly, that is oversampled for the rare factor (called exposure). Thus, for example, stage II of a two-stage design would consist of selecting all case subjects (if the disease is rare) exposed to the rare factor and sampling an equal number in each of the three other categories. Then the other factor (called covariate) involved in the gene-environment interaction is measured in this subsample. The oversampling is taken into account in the analysis to obtain unbiased estimates of the exposure effect and interaction effect in which the exposure is involved (15,19). Cain and Breslow (19) investigated the efficiency of a balanced design compared with a random sampling case-control design, particularly for estimating exposure-covariate interaction. Given a rare exposure (prevalence = 0.05; OR = 2.0) and a covariate of moderate frequency (0.3), with a range of ORs and correlations between the exposure and covariate as measured by the ORs in the control group, the results showed that the balanced design was always much more efficient than a random sampling design for estimating the exposure-covariate interaction in terms of standard errors for both a rare and a common disease. The efficiency decreased as the correlation between the exposure and covariate increased but still remained substantial. For example, when the degree of confounding between the exposure and covariate was set at 5.0, the relative efficiency of the balanced design was greater than three for a common disease and equal to two for a rare disease.
In the two-stage designs, the rare exposure is assumed to be known for the entire population at stage I. Thus, if a rare genetic or environmental factor requires expensive investigation to be measured, measurement may not be affordable at the first stage. One could then propose a three-stage study with two successive sampling stages. Stage I would consist of selecting all case subjects (if the disease is rare) exposed to an inexpensive surrogate variable for the factor under study and sampling equal numbers of unexposed case subjects and exposed and unexposed control subjects. The factor assessment could then be performed on the stage II subsample. Stage III would consist of selecting all case subjects exposed to the rare factor and sampling an equal number of unexposed case subjects and exposed and unexposed control subjects. The environmental or genetic factor of interest for the gene-environment interaction evaluation would be collected in this stage III subsample. Such an approach needs to be statistically evaluated before being considered. At present, identifying good surrogates for the factor(s) of interest and the costs associated with measuring a genetic or environmental factor on large numbers of subjects may be the major determinants for deciding whether or not to conduct multistage balanced and countermatched designs.
Case-Control Study Design Using Related Control Subjects
To our knowledge, a case-control study design using related control subjects has not yet been applied in studies of gene-environment interaction. In this approach, each case subject is matched to one or more unaffected relative(s) and a standard conditional logistic regression is performed to assess gene-environment interaction. This method does not result in biased estimates and does yield consistent estimates, even when there is a correlation in risk factors under study between relatives (20,21). However, bias may occur if there is a correlation, within matched case-control pairs, between measured and unmeasured risk factors (22,23).
Several variations of this study design have been proposed. The use of unaffected siblings has been proposed to study the role of environmental factors and specific gene loci and to provide evidence of gene-environment interaction (24). Case subjects are affected siblings of the proband (case subjects who led to family selection), and control subjects are randomly chosen from unaffected siblings of the proband. The genetic factor (e.g., exposure) for each case and control subject is defined as the number of alleles shared identical by descent (ibd) with the proband at the studied locus. Disease risks can then be estimated for the three groups: share 0, share 1, and share 2. Under the null hypothesis of no linkage between the disease susceptibility locus and the genetic factor locus, disease risks are expected to be identical in the three groups. ORs associated with the genetic factor can be obtained, for example, by comparing the share-2 and share-1 allele groups with the baseline risk group, share-0 alleles. Covariates including environmental exposures may be investigated for main effects and for evidence of interaction with the studied locus by using stratification or multivariate analyses. An alternative matched design has also been proposed for matching on family. One form of this design requires families with at least one affected and one unaffected sibling in addition to the proband limiting the number of available families. An alternative form uses all family members from eligible families in a conditional logistic regression approach. Further methodologic studies are needed to examine issues related to design, conduct, and analysis of these types of studies. In addition, power, efficiency, and feasibility for gene-environment interaction detection using these approaches have not yet been thoroughly investigated.
Another design is based on the use of parental genotypes (25,26). One method of analysis (transmission disequilibrium test [TDT] method) consists of comparing the genotype of each case subject with the genotype of a fictitious control subject composed of the nontransmitted alleles from each parent. The method was first proposed for genetic markers-disease association studies. It has been extended by stratifying case subjects according to the presence or absence of an environmental exposure. The ORs associated with the genetic factor can be obtained for nonexposed and exposed case subjects. Differences in the estimates across strata may reflect gene-environment interaction (27). One disadvantage of this study design is that it does not allow for assessing the main effect of the environmental exposure. Also, differences in estimates across strata are not restricted to resulting from only gene-environment interaction. Differentiating between the various reasons for the differences may be difficult.
The advantage common to these case-related control study designs is the elimination of control subjects whose genetic backgrounds differ systematically from those of case subjects by using relatives as the control subjects. In addition, relatives are easily identified and may be more willing to participate in a research study. Some of these approaches permit estimation of the main effects of the environmental exposures and genetic factors as well as their interaction effects, whereas other approaches only allow estimation of the interaction effect. There may also be a gain in efficiency for detecting gene-environment interaction involving a rare gene compared with a traditional case-control study approach because of oversampling on the genetic factor by using related control subjects. Witte et al. (28) evaluated the efficiency of using either sibling or cousin control subjects for assessing the main effect of the genetic factor and the gene-environment interaction effect. Although relative control subjects were less efficient than population-based control subjects for detecting the genetic factor main effect, sibling control subjects were the most efficient control group for detecting interaction. The gain in efficiency, however, decreased as the frequency of the genetic factor increased (28). One disadvantage of these case-related control designs is the potential for overmatching, particularly on environmental exposures of interest leading to a loss in efficiency. The extent of this loss will depend on the exposures of interest and their correlation among relatives. Thus, there is a potential paradox—efficiency may be improved by oversampling; it may also be reduced by oversampling. The relative efficiency will depend on a number of parameters, including the environmental exposure and genetic factor concordance between relatives. Another possible disadvantage is the cost of finding and ascertaining the relative control subjects. In addition, for designs requiring known ibd status (e.g., comparisons of affected and unaffected siblings), parents must be genotyped otherwise ibd status will not be known and identity by state will need to be used. Also, there may be additional ethical issues associated with using related control subjects versus unrelated control subjects.
Family Study Design
Another proposed study design is a family study with no separate control group. Families are recruited through one or more diseased members, and data on disease status and covariates are collected on all family members. The proposed analysis is segregation analysis or combined segregation and linkage analysis (without or with a known candidate gene), the goals of which are to assess gene-environment interaction. Recently, several approaches have been developed to allow direct assessment of gene-environment interaction (29-33). The study designs allow for estimating genetic, environmental, and gene-environment interaction effects using somewhat different approaches. The regressive models (31) were constructed by specifying a relationship between each person's phenotype for a studied trait and a set of explanatory variables, including the person's major genotype, the phenotype of relatives to take into account residual family dependencies of unspecified origin (genetic or environmental), and measured covariates. Studied traits were able to be either quantitative or qualitative. The major gene is assumed to be in Hardy-Weinberg equilibrium following the rules of mendelian transmission through generations. Abel and Bonney (34) introduced survival analysis concepts into the regressive models to take into account variable ages at trait diagnosis. The censored survival model (33) is an extension to the Cox proportional hazards model and includes measured covariates and unobserved covariates for the genetic component adjusting for variable age at diagnosis for the trait. Two unobserved latent variables are considered: a diallelic major gene and a polygene. The difference between the two approaches in terms of explanatory variables is in residual family likeness for the studied trait that can be taken into account by a polygene in Gauderman and Thomas' censored survival model and by family dependencies in the original model of Bonney. As such, an extra interaction between the environment factor(s) and the polygene may be taken into account in the censored survival approach.
To date, these approaches have been rarely used to detect gene-environment interaction in human diseases. The major limitations are the need to collect individual data on the trait of interest and environmental covariate data on all family members in a sample size sufficient to have power to detect an interaction. Few investigators have been able to collect such data. In addition, the effects of differential participation of family members related to survival, availability, and motivation needs to be assessed. Although limited examinations of power and efficiency for the censored survival model have been conducted (33,35,36), further methodologic studies are needed to evaluate a larger spectrum of models by varying the mode of inheritance, allele frequencies, and relative risks associated with the environmental factor(s), major gene, and interaction effect. Similarly, further methodologic studies are needed to evaluate regressive model efficiency in gene-environment interaction detection for different interaction patterns.
Partly because of some of the difficulties in obtaining genetic and environmental factor data on multiple family members, modifications of the above approaches have been proposed. These variations include family-based cohort designs (e.g., the kin-cohort design) (37), family-based case-control association studies (38), and case-control family study designs (39). These approaches are discussed further elsewhere in the Workshop.
|
GENE-GENE INTERACTION |
|---|
|
TopAbstractIntroductionGene-Environment InteractionGene-Gene InteractionConclusionsReferences |
|---|
All of the study designs previously described, i.e., case-only study, incomplete-data-case-control study, case-control study using unrelated or related control subjects, and multistage case-control study designs, are adaptable for gene-gene interaction assessment. Indeed, the approach is essentially identical and one may replace the environmental factor by a second genetic factor. In family study designs, a second genetic factor would be introduced in segregation and joint segregation and linkage analyses as a covariate similar to gene-environment assessment. In the case where one or both of the two genetic factors are not yet identified, two-locus segregation, linkage, and linkage/association methods may be used. These methods include the MASC method (40), the COMDS method (41), parametric (42), and nonparametric (43) two-locus linkage analyses. This topic will not be further discussed here. Examination of two-locus segregation, linkage, and linkage/association methods are presented by the Gene Discovery Panel (44).
|
CONCLUSIONS |
|---|
|
TopAbstractIntroductionGene-Environment InteractionGene-Gene InteractionConclusionsReferences |
|---|
Table 1
presents the study designs reviewed here
and the references for the designs. Most designs currently available
for examining interactions allow for estimating risk associated with a
genetic factor, environmental exposure factor, and interaction effect.
Case-only studies and TDT-like approaches, however, allow for assessing
interaction effects only. Since case-only studies can efficiently
estimate gene-environment interactions, though, one might consider
using this approach if resources are sufficient to only genotype cases,
if one is interested in studying the effect of gene-environment
interactions on tumor characteristics, if only tumor specimens are
available, or if there are institutional review board concerns about
linking genetic and environmental exposure data in control subjects. On
the other hand, if the exposure risk is not already well known, which
may be the case when the genetic factor involved in the interaction is
common, the usefulness of case-only study designs may be questionable
and the incomplete-data-case-control design developed by Umbach and
Weinberg (7) might be recommended as an alternative.
Assessment of gene-environment interaction effects without estimating
and understanding the main effects of the genetic and environmental
factors would be of little use for public health or individual risk
assessment. However, there is potential for increasing knowledge about
biologic mechanisms underlying gene-environment interaction. The
critical assumption for these studies, the independence of the genetic
and environmental factors, requires linked genetic and environmental
factor data in at least some control subjects (7).
|
Efficiency and power for gene-environment and gene-gene interaction assessment have been rarely studied for the various designs presented, and further investigations are needed to define the efficiency spectra of each approach in interaction assessment. Most available designs seem to be inefficient for detecting interaction of a rare factor(s). Multistage balanced or countermatched studies could be an alternative approach if the rare event is easily and inexpensively measured. As such, at present, it is likely too costly to use this approach to study rare genetic factors. One cautionary note should be added. As the number of known genes increases, the ability to assess gene-environment interaction will also increase. This multiple testing may lead to false evidence for interaction; thus, care must be exercised when evidence for a new gene-environment interaction is first discovered. As is true for new observations in general, confirmation of a new gene-environment interaction as well as a new main effect should be sought.
To date, no single design appears universally applicable for assessing gene-environment or gene-gene interactions. The most appropriate approach(es) will depend on the disease, environmental exposure, genetic factor(s) and the interaction effect, their risk values and frequencies, and how much data are available on each of these components. Further assessment of existing designs and development of new approaches are needed to improve the ability to detect gene-environment and gene-gene interactions in complex diseases.
|
REFERENCES |
|---|
|
TopAbstractIntroductionGene-Environment InteractionGene-Gene InteractionConclusionsReferences |
|---|
1
Smith PG, Day NE. The design of case-control studies: the
influence of confounding and interaction effects. Int J Epidemiol 1984;13:356-65.
2 Piegorsch WW, Weinberg CR, Taylor JA. Non-hierarchical logistic models and case-only designs for assessing susceptibility in population-based case-control studies. Stat Med 1994;13:153-62.[ISI][Medline]
3 Begg CB, Zhang ZF. Statistical analysis of molecular epidemiology studies employing case-series. Cancer Epidemiol Biomarkers Prev 1994;3:173-5.[Abstract]
4
Khoury MJ, Flanders WD. Nontraditional epidemiologic
approaches in the analysis of gene-environment interaction: case-control studies with no controls. Am J Epidemiol 1996;144:207-13.
5 Breslow NE, Day NE. The analysis of case-control studies. In: Davis W, editor. Statistical methods in cancer research. Lyon (France): IARC; 1980.
6 Khoury MJ, Adams MJ, Flanders WD. An epidemiologic approach to ecogenetics. Am J Hum Genet 1988;42:89-95.[ISI][Medline]
7 Umbach D, Weinberg C. Designing and analysing case-control studies to exploit independence of genotype and exposure. Stat Med 1997;16:1731-43.[CrossRef][ISI][Medline]
8 Greenland S. Test for interaction in epidemiologic studies: a review and a study of power. Stat Med 1983;2:243-51.[Medline]
9 Thomas DC, Greenland S. The efficiency of matching in case-control studies of risk factor interactions. J Chronic Dis 1985;38:569-74.[CrossRef][ISI][Medline]
10
Hwang SJ, Beaty TH, Liang KY, Coresh J, Khoury MJ.
Minimum sample size estimation to detect gene-environment interaction in case-control designs. Am J Epidemiol 1994;140:1029-37.
11 Goldstein AM, Falk RT, Korczak JF, Lubin JH. Detecting gene-environment interactions using a case-control design. Genet Epidemiol 1997;14:1085-9.[CrossRef][ISI][Medline]
12
Foppa I, Spiegelman D. Power and sample size calculations for
case-control studies of gene-environment interactions with a polytomous exposure variable. Am J Epidemiol 1997;146:596-604.
13 Caporaso N, Rothman N, Wacholder S, Case-control studies of common alleles and environmental factors. Monogr Natl Cancer Inst1999 ;26:25-30.
14 Siegmund KD, Whittemore AS, Thomas DC. Multistage sampling for disease family registries. Monogr Natl Cancer Inst 1999;26:43-8.
15 Breslow NE. Two phase case-control studies. In: Armitage P, Colton T, editors. Encyclopedia of biostatistics. New York (NY): John Wiley & Sons; 1998;1:532-40.
16 Langholz B, Clayton D. Sampling strategies in nested case-control studies. Environ Health Perspect 1994;102(suppl 8):47-51.
17
Langholz B, Borgan O. Counter-matching: a stratified nested
case-control sampling method. Biometrika 1995;82:69-79.
18 Steenland K, Deddens JA. Increased precision using countermatching in nested case-control studies. Epidemiology 1997;8:238-42.[CrossRef][ISI][Medline]
19
Cain KC, Breslow NE. Logistic regression analysis and efficient
design for two-stage studies. Am J Epidemiol 1988;128:1198-206.
20
Goldstein AM, Hodge SE, Haile RW. Selection bias in
case-control studies using relatives as the controls. Int J Epidemiol 1989;18:985-9.
21 Robins J, Pike M. The validity of case-control studies with non random selection of controls. Epidemiology 1990;1:273-84.[Medline]
22
Gladen BC. Matched-pair case-control studies when risk factors
are correlated within the pairs. Int J Epidemiol 1996;25:420-5.
23
Andrieu N, Goldstein AM. Use of relatives of cases as controls
to identify risk factors when an interaction between environmental and genetic factors exists. Int J Epidemiol 1996;25:649-57.
24 Khoury MJ, Flanders WD, Lipton RB, Dorman JS. The affected sib-pair method in the context of an epidemiologic study design. Genet Epidemiol 1991;8:277-82.[CrossRef][ISI][Medline]
25 Ewens WJ, Spielman RJ. The transmission/disequilibrium test: history, subdivision and admixture. Am J Hum Genet 1995;57:455-64.[ISI][Medline]
26
Khoury MJ, James LM. Population and familial relative risks of
disease associated with environmental factors in the presence of gene-environment interaction. Am J Epidemiol 1992;137:1241-50.
27 Maestri NE, Beaty TH, Hetmanski J, Smith EA, McIntosh I, Wyszynski DF, et al. Application of transmission disequilibrium tests to nonsyndromic oral clefts: including candidate genes and environmental exposures in the models. Am J Med Genet 1997;73:337-44.[CrossRef][ISI][Medline]
28
Witte JS, Gauderman WJ, Thomas DC. Asymptotic bias and
efficiency in case-control studies of candidate genes and gene-environment interactions: basic
family designs. Am J Epidemiol 1999;149:693-705.
29 Moll PP, Sing CF, Lussier-Cacan S, Davignon J. An application of a model for a genotype-dependent relationship between a concomitant (age) and a quantitative trait (LDL cholesterol) in pedigree data. Genet Epidemiol 1984;1:301-14.[CrossRef][Medline]
30 Konigsberg LW, Blangero J, Kammerer CM, Mott GE. Mixed model segregation analysis of LDL-C concentration with genotype-covariate interaction. Genet Epidemiol 1991;8:69-80[CrossRef][ISI][Medline]
31 Bonney GE. Regression logistic models for familial disease and other binary traits. Biometrics 1986;42:611-25[CrossRef][ISI][Medline]
32 Demenais F, Lathrop M. REGRESS: a computer program including the regressive approach into the LINKAGE programs. Genet Epidemiol1994 ;11:291.
33 Gauderman WJ, Thomas DC. Censored survival models for genetic epidemiology: a gibbs sampling approach.Genet Epidemiol 1994;11:171-88.[CrossRef][ISI][Medline]
34 Abel L, Bonney G. A time-dependent logistic hazard function for modeling variable age of onset in analysis of familial diseases. Genet Epidemiol 1990;7:391-407.[CrossRef][ISI][Medline]
35 Gauderman WJ, Morrison JL, Carpenter CL, Thomas DC. Analysis of gene-smoking interaction in lung cancer. Genet Epidemiol 1997;14:199-214.[CrossRef][ISI][Medline]
36 Gauderman WJ, Faucett CL. Detecting gene-environment interactions in joint segregation and linkage analysis. Am J Hum Genet 1997;61:1189-99.[CrossRef][ISI][Medline]
37 Gail MH, Pee D, Carroll R. Kin-cohort designs for gene characterization. Monogr Natl Cancer Inst 1999;26:55-60.
38 Gauderman WJ, Witte JS, Thomas DC. Family-based association studies. Monogr Natl Cancer Inst 1999;26:31-7.
39 Zhao LP, Aragaki C, Hsu L, Potter J, Elston R, Malone KE, et al. Integrated designs for gene discovery and characterization. Monogr Natl Cancer Inst 1999;26:71-80.
40 Dizier MH. Babron MC. Clerget-Darpoux F. Interactive effect of two candidate genes in a disease: extension of the marker-association-segregation chi(2) method. Am J Human Genet 1994;55:1042-9.[ISI][Medline]
41 Morton NE, Shields DC, Collins A. Genetic epidemiology of complex phenotypes. Ann Hum Genet 1991;55:301-14.[ISI][Medline]
42 Schork NJ, Boehnke M, Terwilliger JD, Ott J. Two-trait-locus linkage analysis: a powerful strategy for mapping complex genetic traits. Am J Hum Genet 1993;53:1127-36.[ISI][Medline]
43 Knapp M, Seuchter SA, Baur MP. Two-locus disease models with two marker loci: the power of affected-sib-pair tests. Am J Hum Genet 1994;55:1030-41.[ISI][Medline]
44 Schaid DJ, Buetow K, Weeks DE, Wijsman E, Guo SW, Ott J, et al. Discovery of cancer susceptibility genes: study designs, analytic approaches, and trends in technology. Monogr Natl Cancer Inst 1999;26:1-16.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J. Lorenzo Bermejo and K. Hemminki Gene-environment studies: any advantage over environmental studies? Carcinogenesis, July 1, 2007; 28(7): 1526 - 1532. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||