Leveraging Electronic Health Records for Phenotyping - Translational Informatics: Realizing the Promise of Knowledge-Driven Healthcare - page 62

the EHR. In other cases, a large sample of biospecimens is collected fi rst, but is
not genotyped. Extraction rules for a specifi c phenotype are developed, and this is
run against the EHR for all the subjects in the biobank. Cases and controls are then
selected as a subset based on the matching phenotypes, and then biospecimens from
those specifi c subjects are genotyped. This is a cost-savings approach, because only
a sub-population requires genotyping, which is typically the most costly component
for the research study.
Genome-wide association (GWA) analysis can occur once the genotypes and
phenotypes are created and linked. Analysis is often done with standard statistical
testing, such as analysis of variance, contingency tests, or regression analyses. More
advanced testing is used to account for covariates. More complicated testing is done
when the interest is not just single gene variations, but for interactions among dif-
ferent genes. These multi-locus analyses quickly become computationally diffi cult,
but various methods have been successfully used to fi lter SNPs for analysis. The
results of the analyses are identifi cation of signifi cant associations between gene
markers or combinations of genes and specifi c phenotypes. Like all scientifi c stud-
ies of signifi cance, the result can be further validated by replicated tests, until a
recognized association for a trait is made, and a genetic test can then be developed
to mark an individual's specifi c risk for that condition [ 13 ].

4.3.3 Pharmacogenetics and Personalized Medicine

A specifi c example of applying genotype-phenotype analysis from EHRs is in
discovering medication effi cacy, or pharmacogenetics. Bush and Moore [ 13 ] give
a good example about warfarin, a blood-thinning medication that helps prevent
clots in patients at risk of an embolism. Administering anticoagulation therapy
to patients is a delicate process, where the right dose needs to be determined and
used. If too low a dose is used, the medication will not prevent potentially fatal
clots; if too high a dose is used, the blood can become too thin and the patient
risks dangerous internal and external bleeding. When administering anticoagu-
lation therapy, clinicians must carefully watch the patient's clotting activity, or
prothrombin time, because warfarin has a very narrow therapeutic window. A
GWAS has shown that there is also wide variation due to genetics in a patient's
response to warfarin - in some populations greater than any other known factor
[ 33 ]. Phenotype data for this study were collected from electronic health records,
including warfarin doses for the patients, lab tests indicating clotting activity, and
demographics. Linear regression was then used to analyze the data. The result
of the GWAS is the development of a genetic test for patients to determine their
appropriate safe warfarin doses. This type of test, designed to tailor the care pro-
vided to an individual based on her genotype, is personalized medicine, and rep-
resents an important translation of GWAS to actual patient care. A benefi t of
personalized medicine being developed from data extracted from electronic health
records is that they can be more effective when used with computerized decision