Leveraging Electronic Health Records for Phenotyping - Translational Informatics: Realizing the Promise of Knowledge-Driven Healthcare - page 65

from multiple clinical sites, and addressed issues of merging data from different
EHRs to increase both the breadth and depth of the data available. For example, the
Scalable Architecture for Federated Translational Inquiries Network (SAFTINet)
project in Colorado created a distributed research network across multiple sites of
care in multiple states, allowing the creation of large cohorts of patients with EHR
data. They also focused on diverse and underserved populations, and initially stud-
ied heart and blood vessel conditions, as well as breathing conditions [ 40 ]. The
SUrveillance, PREvention, and ManagEment of Diabetes Mellitus (SUPREME-DM)
project at the Kaiser Permanente Institute for Health Research leverages EHR data
to create a longitudinal registry of diabetes care for 1.3 million patients across 11
geographically-disparate integrated delivery systems. They also studied other con-
ditions related to the population, such as gestational diabetes, obesity, and heart
conditions. Like SAFTINet, they used a distributed data network [ 41 ].
The Washington Heights/Inwood Informatics Infrastructure for Comparative
Effectiveness Research (WICER) also leveraged EHR data to create patient pheno-
types. Similar to MURDOCK but different than SAFTINet and SUPREME-DM,
WICER used EHR data as one source among many for a specifi c population. The
focus was an underserved, immigrant population in New York City. WICER linked
data from EHRs of multiple providers (inpatient, ambulatory, home care) to create
a comprehensive and longitudinal view of the clinical data for a patient. For a subset
of the population, WICER also surveyed individuals at various locations on clinical
measures, social attitudes and networks, and health behaviors and beliefs. They also
collected biospecimens for some of the population. For patients with biospecimens,
EHR data across sites and survey data, WICER created one of the most comprehen-
sive datasets for a research population. By collecting data from multiple sources,
WICER investigators are also able to study differences in data quality and com-
pleteness across different sources [ 42 ].
While eMERGE, i2b2 and other studies have advanced the understanding of EHR
data for genotype-phenotype research, these AHRQ-funded studies have been criti-
cal in learning important lessons about using EHRs for cohort studies in a popula-
tion. They have advanced understanding of data display and navigation, data security,
primary data collection, distributed queries, research sustainability, and data quality
[ 18 , 22 , 40 , 42 - 44 ]. They have also each built an infrastructure that is now being
extended for more analyses. These infrastructures demonstrate both the viability and
potential of leveraging EHR data to defi ne subjects for clinical research studies.

4.3.6 Challenges and Future Directions in Using EHR Data

for Research

Along with these successes, however, are challenges. Secondary use of EHR data is
more available and inexpensive than primary research collection, but since it is col-
lected for a different purpose its quality for research is different. For example, EHR
data is most complete on patients who have visits to health providers using the