Leveraging Electronic Health Records for Phenotyping - Translational Informatics: Realizing the Promise of Knowledge-Driven Healthcare - page 59

hereditary link and rare genetic variants [ 11 , 12 ]. But there are also many common
disorders caused by many common genetic variants, with weaker links. New types
of studies other than family-based studies were needed for common genetic vari-
ants. GWAS are population-based studies of common variants with small effects
[ 13 ]. GWAS have also been used successfully to identify variations in patients'
responses to different medications, leading to personalized medicine.
One thing that makes GWAS particularly interesting is that the genotype, once it
has been sequenced, is available for studying multiple associations. A challenge is
to then have a large enough sample to do population-based studies on common dis-
eases [ 14 ]. Various institutions have created biobanks to address this, where bio-
logic samples are collected for use in GWAS studies, many with tens of thousands
of samples [ 15 ]. The last decade has seen a tremendous growth in sequencing abil-
ity, with a corresponding reduction in costs. EHRs can then be used as a source of
phenotypic information for genotype-phenotype studies like GWAS.
The emergence of GWAS has changed the potential of using EHRs for research
data for many reasons. First, the large number of subjects required for GWAS has
increased the overall benefi t of using EHRs. When smaller numbers of subjects
were needed, alternatives to phenotype extraction from EHRs could be an accept-
able option. Second, successes in GWAS have been both signifi cant and rapid. Since
the fi rst GWAS study in 2005, thousands of GWAS studies have examined hundreds
of traits and diseases [ 16 ]. The results and interest in GWAS, and the subsequent
need for phenotypic data to expand it, has been explosive. Third, successful GWAS
studies have included development of an infrastructure that can be more easily lev-
eraged for successive studies. The creation of biorepositories is comparatively more
diffi cult than extraction methods from electronic health records. Once the biore-
pository is created, however, its marginal cost of use drops rapidly especially for
genotyped samples. The comparative cost-benefi t of using EHR phenotypes makes
it a more worthwhile pursuit.
Initial demonstrations using EHRs for research has only expanded their per-
ceived potential. Beyond GWAS, they are seen as a method to rapidly identify vari-
ables and outcomes of cohorts that can then be applied to similar patients under
treatment [ 17 ]. Comparative effectiveness research has also increased the demand
for using EHR data in research [ 18 ]. Effectiveness research focuses on studying
interventions in the “real world,” or the environment where the interventions would
be most likely to be received. This is in contrast to clinical trials that have actively
limited the environment of the trial to study effi cacy of an intervention, without
confounders. EHRs collect data in the world where care is provided, so their data
are more relevant to effectiveness studies than data collected in case report forms
during effi cacy studies. The learning health system is focused on effectiveness, and
the use of EHR data to support it is fundamental to its vision [ 17 ]. Phenotype extrac-
tion from EHRs has become a critical requirement of both research and health care
transformation.
Understanding the history of leveraging EHRs for research is important to recog-
nize both the potential benefi ts and challenges. It has been pursued for a long time,
with great successes and even greater potential. Some of the most signifi cant early