5 Methods for Association Studies
111
due to population stratification. An alternative adjustment method can be used that
requires LD score regression, which quantifies the association between LD and test
statistics within a GWAS using a reference panel, in order to calculate a correction
factor for genomic control in GWAS analysis (Bulik-Sullivan et al. 2015).
Methods that account for cryptic relatedness specifically tend to be more complex
and are largely beyond the scope of this chapter. Software packages such as KING
(Manichaikul et al. 2010) can be implemented to identify closely related individuals,
who can subsequently be removed from the analytical population. There also exist
approaches to deal with more distant relatedness as part of data analysis (Price et al.
2010).
5.5.7.2 Addressing Other Confounding
Genetic association studies are distinct from traditional epidemiological studies in
that behavioral and environmental factors are unlikely to confound the associations
of interest. Despite this, to draw clinically relevant conclusions, it is critical to
properly account for patient-level covariates that may confound the associations
under investigation. For example, associations may be confounded by sex whenever
allele frequencies differ between the sexes (i.e., for sex-linked traits) (Clayton 2009).
Other associations may be confounded by age whenever tag SNPs are in LD with
both longevity SNPs and causal SNPs for a phenotype that most commonly occurs in
late-life. If one does not adjust for the necessary covariates, one might find spurious
associations due to sampling artifacts or bias in the study design. It is important
to note that covariate adjustment may reduce statistical power because it requires
additional degrees of freedom.
5.5.7.3 Improving Precision
Covariates may also be included in tests of association in an attempt to improve
the precision of estimates. If a behavioral or environmental factor is associated with
a quantitative phenotype under study independently of the genes of interest, then
its inclusion is often beneficial. The covariate can explain some of the variability
in the outcome, thereby reducing noise and increasing power (Mefford and Witte
2012). For binary traits, the story is more complicated. Inclusion of a covariate
associated only with the outcome may actually reduce power for case-control
association studies (Pirinen et al. 2012). There do exist methods, however, that
leverage information about covariates to increase power in association studies of
binary traits (Zaitlen et al. 2012).
5.5.8 Multiple Testing
Genetic association studies generally test hundreds of thousands of associations
and may also examine multiple phenotypes and/or the results from various genetic
models and covariate adjustments. The enormous number of resulting hypothesis
tests must be adjusted for multiple comparisons, lest a large number of false-positive
associations be detected. One approach to management of this issue of multiple
111
due to population stratification. An alternative adjustment method can be used that
requires LD score regression, which quantifies the association between LD and test
statistics within a GWAS using a reference panel, in order to calculate a correction
factor for genomic control in GWAS analysis (Bulik-Sullivan et al. 2015).
Methods that account for cryptic relatedness specifically tend to be more complex
and are largely beyond the scope of this chapter. Software packages such as KING
(Manichaikul et al. 2010) can be implemented to identify closely related individuals,
who can subsequently be removed from the analytical population. There also exist
approaches to deal with more distant relatedness as part of data analysis (Price et al.
2010).
5.5.7.2 Addressing Other Confounding
Genetic association studies are distinct from traditional epidemiological studies in
that behavioral and environmental factors are unlikely to confound the associations
of interest. Despite this, to draw clinically relevant conclusions, it is critical to
properly account for patient-level covariates that may confound the associations
under investigation. For example, associations may be confounded by sex whenever
allele frequencies differ between the sexes (i.e., for sex-linked traits) (Clayton 2009).
Other associations may be confounded by age whenever tag SNPs are in LD with
both longevity SNPs and causal SNPs for a phenotype that most commonly occurs in
late-life. If one does not adjust for the necessary covariates, one might find spurious
associations due to sampling artifacts or bias in the study design. It is important
to note that covariate adjustment may reduce statistical power because it requires
additional degrees of freedom.
5.5.7.3 Improving Precision
Covariates may also be included in tests of association in an attempt to improve
the precision of estimates. If a behavioral or environmental factor is associated with
a quantitative phenotype under study independently of the genes of interest, then
its inclusion is often beneficial. The covariate can explain some of the variability
in the outcome, thereby reducing noise and increasing power (Mefford and Witte
2012). For binary traits, the story is more complicated. Inclusion of a covariate
associated only with the outcome may actually reduce power for case-control
association studies (Pirinen et al. 2012). There do exist methods, however, that
leverage information about covariates to increase power in association studies of
binary traits (Zaitlen et al. 2012).
5.5.8 Multiple Testing
Genetic association studies generally test hundreds of thousands of associations
and may also examine multiple phenotypes and/or the results from various genetic
models and covariate adjustments. The enormous number of resulting hypothesis
tests must be adjusted for multiple comparisons, lest a large number of false-positive
associations be detected. One approach to management of this issue of multiple