Electronic Health Record (EHR) Workgroup


Electronic Health Record (EHR) Workgroup Membership


Joseph Erdos
Michael Matheny

Core Members:

Norman Silliker (Yale)
William Roddy (WRNMMC/USU) (HJF, CTR)
J.C. Sarver (DoD)
Denise Hynes (VA)
Margaret Gonsoulin (VA)


Workgroup Co-Chairs – Qualifications and Experience

Joseph Erdos, MD, PhD, was the VACHS Chief Information Officer from 1995 to 2011 and was a member of the VA Chief Information Officers Council. He was instrumental in the early design and implementation of the VistA EHR system into care in 1999 as well as VistA Imaging. A demonstration project on Telemedicine at VA Connecticut led to the development of home telemedicine in the VA. In 2011 Dr. Erdos was tasked with development of the VA’s Corporate Data Warehouse (CDW), a repository of standardized mapped, accessible, EHR data for operations and research. He was promoted to serve as Director of the VA East Coast Business Intelligence Service Line of the Office of Information. He and his co-directors have overseen the conception and implementation of the VA CDW which has extracted and modeled over 50 clinical and administrative domains from approximately 150 VistA platforms resulting in approximately 2 trillion rows of data!  Dr. Erdos has experience in VistA, Cerner, Epic, CHCS, AHLTA and other EHRs and systems used over the years. He daily advises PRIME investigators in the appropriate use of VA data to achieve their goals. If data do not exist in the CDW he provides tools and develops methods to get them.

Michael Matheny, M.S., M.D. is an Associate Professor of Biomedical Informatics with secondary appointments in Medicine and Biostatistics at Vanderbilt University Medical Center.  He is a board certified in Internal Medicine and Clinical Informatics. Dr. Matheny is a nationally recognized investigator in predictive analytics, machine learning, automated medical device surveillance, and NLP.  His work focuses on studying how to best leverage large EHR data for discovery of risk prediction using both structured data and NLP derived data, as well as conducting automated adverse event surveillance in medical devices. He is an Associate Director of VA HSR&D VINCI which helps provision data to the VA research community and provide informatics and health services research tools to assist in studies. He is also deeply involved with the Joint Interagency Funded initiative to provide DoD data for those patients registered in the VA and helping lead the initiative to transform both the VA and DoD data sources into the Observational Medical Outcomes Partnership (OMOP) common data model.  He is the Co-PI of the pScanner PCORI Clinical Data Research Network and is the lead investigator for the VA for this initiative that executes an array of distributed analytic queries for large-scale observational cohort analyses.

Workgroup Goals

A critical barrier to the conduct of pragmatic clinical trials of non-pharmacological approaches to chronic pain management in VA and DoD health systems, or in fact, in virtually all practice settings, are limits in the type and quality of pain-relevant data in EHRs. Data are largely limited to routinely documented ratings of pain severity (otherwise known as Pain as the 5th Vital Sign) and pain-relevant diagnoses, tests and procedures, clinical encounters, and medications. Despite these limitations, important large scale observational studies of chronic pain and even pragmatic studies of pain care in VA have been possible. Using ML and NLP, our team and others have begun to develop strategies for extraction of additionally relevant information from unstructured text notes of providers and other team members. For example, our team is developing methods for extracting information relevant to characterizing dimensions of pain care quality (i.e., provider assessment and treatment plans) including patient/family education and documentation of delivery of or recommendations for non-pharmacological interventions (including complementary and integrative health approaches). Our group is also at the forefront of developing strategies for the automated collection of PROs and integration of these data into the EHR.33 Of particular importance are initiatives in VA and DoD that leverage the NIH PROMIS pain measures and/or the PASTOR/PROMIS and CHOIR registry approaches. Ultimately, the goals of the EHR Work Group, in conjunction with our broader team’s experience and expertise related to the VA and DoD EHR, will be to optimize use of existing EHR data, support integration of patient reported data, and to create new data from unstructured text in EHRs using innovative ML and NLP tools. The Work Group will promote data standardization and interoperability across systems and EHRs and to review, develop, and adapt policies and tools that can be easily shared and adopted in the service of supporting the selected pragmatic trials.

Additional Value-added Expertise

PRIME Center investigators, in particular, have extensive expertise with VistA, the current VA EHR both at the point of care for clinical decision support (CDS) as well as using the CDW for near-real time pragmatic trials.  As just a few examples, PRIME investigators Goulet, Kerns, Brandt, Heapy, Becker and Fried have used VA EHR data for patient sampling and recruitment based on clinical characteristics and medications as well as for outcome assessment. As noted above, our team led by Dr. Kerns developed and employed novel approaches to assessment of pain severity, medication use and multimodal pain care in the context of a pragmatic system-level intervention at VACHS. We have also developed processes and procedures to manage the import of digital patient data (e.g. EEG, EHR, imaging) and activity monitors such as FitBit and patient reported outcome data. As DoD is now migrating to the MHS Genesis (Cerner), our team’s expertise with Cerner will likely prove valuable. The DoD members of our team also bring a wealth of expertise and experience with the current DoD EHR and it is expected that they will acquire the same expertise with the MHS Genesis EHR as it becomes more widely implemented.