Transcript Lessons learned in developing a national registry for

Lessons learned in developing a national
registry for community-led Patient Centered
Outcomes Research
October 29, 2012
APHA Conference 2012
San Francisco, CA
Reesa Laws, BS, Thu Quach, PhD, Rosy Chang Weir, PhD, Erin Kaleba, MPH, Chris
Grasso, MPH, Stephan Van Rompaey, PhD, Jon Puro, MPH-HA, Joe Carroll, MD,
Suzanne Gillespie, MS
Background
• The Community Health Applied Research Network
(CHARN) is a federally funded research network comprised
of 18 community health centers (CHCs) organized into four
research nodes (each including an academic partner),
and a data coordinating center (DCC).
• Goal of establishing a community-led network for patientcentered outcomes research (PCOR).
• One key initiative is to develop a robust central CHARN
Data Registry.
Community Health Centers
• CHC’s were created to provide health and social
services access points in poor and medically
underserved communities and to promote
community empowerment.
• They are considered the “safety net” for many of
the underserved population.
• 7000 CHCs nationwide in all 50 U.S. states and
territories, serving approximately 20 million.
The CHARN registry
population
• V1 of the registry was defined as all patients who had at
least one primary care encounter that occurred on or
after January 1, 2008 and prior to December 31, 2010.
• Detailed data was shared for those patients that had at
least one diagnosis, medication or laboratory test from
one of the seven CHARN diseases of interest, which will
be discussed in a later slide.
• Number of patients across the CHARN nodes:
o
o
o
o
AAPCHO: 89,889
Alliance: 280,171
Fenway: 4,907 (only HIV patients were included)
OCHIN: 156,848
CHARN Data Registry Update
V1 Patient Characteristics tables
Proportion of Patients by Node
60.0%
50.0%
40.0%
30.0%
20.0%
10.0%
0.0%
AAPCHO
Alliance
Fenway
OCHIN
CHARN Data Registry Update
V1 Patient Characteristics tables
Age
Gender
30.0%
60.0%
25.0%
50.0%
20.0%
40.0%
15.0%
30.0%
10.0%
20.0%
5.0%
10.0%
0.0%
Less than
18
18-25
26-39
40-64
65-79
80 and
older
0.0%
Male
Female
CHARN Data Registry Update
V1 Patient Characteristics tables
Race
35.0%
30.0%
25.0%
20.0%
15.0%
10.0%
5.0%
0.0%
White
Black/African
American
Am. Indian/Alaska
Native
Asian/NHOPI
Multi-racial
Other
Objective
• To establish a centralized data registry extracted
from electronic health record (EHR) data systems at
CHCs to:
o Better describe our vulnerable, diverse safety-net
populations traditionally underrepresented in research,
and
o Establish a multi-site, multidisciplinary collaborative
infrastructure to advance PCOR.
o Address scientific questions that can be easily answered by
a large-scale combined community health center registry.
Methods
• As a key initial step, we developed multi-level Data
Use Agreements (DUA) between all participating
CHCs and their representative node, and between
the nodes and the data coordinating center
(DCC).
• Simultaneously, a multidisciplinary team of
community clinicians, researchers, and data
programmers defined data elements needed to
support future PCOR.
Data Use Agreements
• The multi-step DUA process added complexities
and time to the development and approval
process, however, this process was crucial for
building trust in using individual CHC’s data at a
national level.
• Different strategies were reviewed for streamlining
the DUA approval proces.s
o Individual CHC’s with the DCC: It was too laborious and
time intensive.
o Creation of a node specific or CHARN specific IRB: Most
CHC’s wanted to maintain their data sharing authority at
their individual organizations. However, one node has a
central IRB established and at least one other node is
moving towards a central IRB process for their node.
Registry Design
• The CHARN Steering Committee (SC),
which is the leadership body of the
Network, defined the high level goal for the
CHARN Registry. The focus for version 1
(V1) of the registry was to compile data on
seven specific disease cohorts:
o
o
o
o
o
o
Diabetes
HIV
Hepatitis B and Hepatitis C
Cardiovascular Disease
Hypertension
Dyslipidemia
Registry Design
• The Data Sharing and Registry Subcommittee
(DSRS) was tasked with the development of a data
schema for the organization and extraction of the
CHC level data for the construction of the CHARN
registry.
o A standardized data dictionary (DD) was created to define
requested data elements.
o A data submissions process was developed to specify
procedures for compiling, querying and transmitting the
data.
• Microsoft SQL Server was chosen as the database
to store data at the CHC, node and DCC as it’s a
commonly used and robust tool.
Registry Design
• Limited data sets were created with patient
identifiers removed, as defined by the HIPAA
privacy rules. Dates of service are included in the
registry.
• As a result, we are not able to de-duplicate patients
across nodes; however, we rely on the nodes to deduplicate the patients within their individual nodes.
• Metadata will be captured on the procedures used
at the node for this process.
Registry Implementation
• The DCC created node-specific SQL script that the
nodes used to create empty registry tables.
• Data from the individual CHC’s could then be
loaded into the node level registry.
• Standardized data queries were then run at the
node level before transferring the data to the DCC.
Registry Implementation
• Two levels of data queries were run at the node
level to ensure data integrity. Level one included:
o Confirming that all data conformed to the defined SQL
server field data types.
o All records loaded into the tables conformed to the
primary key constraints.
• Level 1 checks had to be passed in order to load
the data into the tables.
Registry Implementation
• Level two checks consisted of the following:
o Data format (field level data conformity)
o Required fields (no missing data in required fields)
o Foreign key (data values exist in other tables where an explicit
relationship exists)
o Valid code (values confirm to a list of pre-defined codes)
o Valid range (values confirm to a pre-defined range)
o Orphan records (every record in a “non-patient” table links to
a record in the “patient” table)
• After completion of any needed data cleaning at
the nodes, the data were uploaded to a secure
website at the DCC for aggregation across the
nodes and additional data queries.
Data Complexities
• Lack of a standardized data classification system for labs
and medications.
• Missing data.
• Current workgroups were limited to the seven disease
cohorts due to the decision to create a registry vs.
pulling all data to create a warehouse (e.g., untreated
hypertension was an area of interest that could not be
ascertained with the current version).
• Not all CHC’s had an EHR system in place.
• Linking encounters (visits) to medication orders and lab
orders.
• Multiple data sources at the CHC level made it time
intensive to compile all required data.
Results
• Nodes have executed DUA’s with their CHC’s and with
the DCC.
• Each participating CHC has approved this project
through its institutional review board (and local research
committee when required).
• Data for version 1 have been submitted by all nodes to
the DCC, and subsequently aggregate reports are being
returned to the nodes for QA and to inform research
study proposals.
• V2 of the registry is being developed to capture all
patient data across more years (a data warehouse).
• Study of diabetes using version 1 registry data is currently
underway. Other version 1 studies are pending.
Conclusions
• It is feasible to create a centralized data
registry among multiple CHC partners, with
different types of EHRs, and varying levels of
experience and research topics.
• For this type of unprecedented research
endeavor, it is essential to allow for significant
time for:
o approval processes
o discussions which ultimately foster collaboration and trust among
new partners
o addressing technical issues in an era of suboptimal data
standardization