5 Methods for Association Studies
99
5.3.1 Quantitative Versus Qualitative Traits
There are two primary classes of phenotypes that one might wish to evaluate with
a genetic association study, namely quantitative and qualitative (most often binary
case-control). Quantitative traits generally have higher statistical power to detect
genetic effects, and the interpretation of effects is often more straightforward. For
a genetic variant that influences a quantitative trait, each allele or genotype class
may be interpreted as affecting a unit change in the level of the trait. Alternatively,
one might opt to study subjects at the extremes of a quantitative trait distribution
to maximize power per genotyped individual for detecting associations (Huang and
Lin 2007; Guey et al. 2011).
Many diseases do not have meaningful or well-established quantitative measures.
In such scenarios, individuals are commonly classified as either affected or unaffected, and studies most often implement a case-control design. Frequencies of
genetic variants observed in cases are compared with those observed in controls
in order to evaluate whether an association between genes and disease exists. It
is important to note that for a complex phenotype (e.g., metabolic syndrome) or
one that is diagnosed over a long period (e.g., Alzheimer’s disease), there may be
some measurement error in dichotomizing individuals as cases or controls. Still,
many association studies of binary traits have been extremely successful in detecting
genetic variants correlated with disease (see Chap. 7 on what we have learned from
GWAS).
5.3.2 Subject Selection
The most important facet of subject selection is ensuring that subjects are representative of their source population (Wacholder et al. 1992). For a case-control
study in which cases with a particular disease are compared to unaffected controls,
this means that controls should be individuals who, if diseased, would be cases.
Whenever controls are not selected to represent the source population of the
cases, spurious associations may result. Consider, for example, a scenario in which
controls are selected from a different ancestral population from cases. In such a
circumstance, control subjects might have fundamentally different allele frequencies
in the SNPs of interest relative to cases. As a result, one is likely to find associations
between these SNPs and disease even in the absence of true associations. This
particular bias is called population stratification and can, if unaccounted for,
confound GWAS. We will discuss methods to control for population stratification
later in this chapter.
Cases are commonly recruited from a specific population, hospital, or disease registry. Depending on the study design, controls may either be unrelated
(population-, hospital-, or registry-based) or family members of the cases. Controls
are also commonly matched to cases with respect to ancestry, age, and sex.
Even without rigorous control selection, many GWAS have been successful at
detecting highly replicated variants. Due to the high cost of subject recruitment and
99
5.3.1 Quantitative Versus Qualitative Traits
There are two primary classes of phenotypes that one might wish to evaluate with
a genetic association study, namely quantitative and qualitative (most often binary
case-control). Quantitative traits generally have higher statistical power to detect
genetic effects, and the interpretation of effects is often more straightforward. For
a genetic variant that influences a quantitative trait, each allele or genotype class
may be interpreted as affecting a unit change in the level of the trait. Alternatively,
one might opt to study subjects at the extremes of a quantitative trait distribution
to maximize power per genotyped individual for detecting associations (Huang and
Lin 2007; Guey et al. 2011).
Many diseases do not have meaningful or well-established quantitative measures.
In such scenarios, individuals are commonly classified as either affected or unaffected, and studies most often implement a case-control design. Frequencies of
genetic variants observed in cases are compared with those observed in controls
in order to evaluate whether an association between genes and disease exists. It
is important to note that for a complex phenotype (e.g., metabolic syndrome) or
one that is diagnosed over a long period (e.g., Alzheimer’s disease), there may be
some measurement error in dichotomizing individuals as cases or controls. Still,
many association studies of binary traits have been extremely successful in detecting
genetic variants correlated with disease (see Chap. 7 on what we have learned from
GWAS).
5.3.2 Subject Selection
The most important facet of subject selection is ensuring that subjects are representative of their source population (Wacholder et al. 1992). For a case-control
study in which cases with a particular disease are compared to unaffected controls,
this means that controls should be individuals who, if diseased, would be cases.
Whenever controls are not selected to represent the source population of the
cases, spurious associations may result. Consider, for example, a scenario in which
controls are selected from a different ancestral population from cases. In such a
circumstance, control subjects might have fundamentally different allele frequencies
in the SNPs of interest relative to cases. As a result, one is likely to find associations
between these SNPs and disease even in the absence of true associations. This
particular bias is called population stratification and can, if unaccounted for,
confound GWAS. We will discuss methods to control for population stratification
later in this chapter.
Cases are commonly recruited from a specific population, hospital, or disease registry. Depending on the study design, controls may either be unrelated
(population-, hospital-, or registry-based) or family members of the cases. Controls
are also commonly matched to cases with respect to ancestry, age, and sex.
Even without rigorous control selection, many GWAS have been successful at
detecting highly replicated variants. Due to the high cost of subject recruitment and