Leveraging Electronic Health Records for Phenotyping - Translational Informatics: Realizing the Promise of Knowledge-Driven Healthcare - page 66

EHR. If a patient is not being treated, the information is not collected. If the pro-
vider does not use the EHR, the information is not collected. If the information is
not directly relevant to the care being provided, the information may not be col-
lected by the provider. In contrast, with research-based case report forms the spe-
cifi c data elements of interest are defi ned prospectively, and the data are collected
for each subject.
The real effect of these biases is still unknown. Researchers acknowledge that
data quality could undermine the ability to use EHR data as a surrogate for primary
research collection. Previous studies using billing data have had noted quality issues
[ 2 ]. Currently, researchers with EHR data for phenotypes are investigating methods
for assessing quality [ 40 , 45 ]. Best practices have been defi ned for researchers to
validate the accuracy of phenotypes that are extracted from EHRs [ 13 ]. And studies
have shown that the data, while imperfect, are at least similar in specifi c cases to
self-report data.[ 26 ] At the same time, a study of differences in phenotype defi ni-
tions for diabetes showed signifi cant variation in populations depending on the
phenotype defi nition used [ 25 ]. Data quality continues to be an area of concern,
though it has yet to invalidate the approach.
One issue of data quality from EHRs that is expected to decrease over time is the
issue of data completeness. With government incentives for EHR adoption under
the Meaningful Use program, institutions have increased their use of EHRs. More
signifi cantly, the Meaningful Use criteria have specifi ed certain data types that must
be collected above threshold levels [ 46 ]. For the data elements that are part of the
Meaningful Use regulations, this will undoubtedly increase the consistency of their
creation, and will likely increase the consistency of phenotypes defi ned for extract-
ing that information from the EHRs for research.

4.4 Implications for Stakeholders

As was introduced in Chap. 2 , a variety of stakeholders can and will benefi t from the
ability to leverage data sources, such as EHRs, in order to enable patient- and
population-level phenotyping. Critical examples of these benefi ts stratifi ed by
stakeholder type include the following:

Evidence and Policy Generators

Research at the patient and/or population levels requires the derivation and
analysis of complex and discrete phenotypes . Such phenotyping is even more
critical when attempting to link clinical presentations of health or disease with bio-
molecular markers, such as the activities introduced in Chap. 3 . While the adop-
tion of healthcare IT platforms, such as EHRs, provide a basis for such phenotype
generation, the act of computerizing patient records does not represent a complete
solution to such information needs. Thus, the application of phenotyping princi-
ples such as those described in this chapter are central to generating data critical to
research endeavors that will ultimately result in new, actionable knowledge.