Automated Data Acquisition for Heart Failure: Performance Measures and Treatment
Jennifer H. Garvin PhD MBA
VA Salt Lake City Health Care System, Salt Lake City, UT
Salt Lake City, UT
Funding Period: April 2010 - September 2013

BACKGROUND/RATIONALE:
There is an association of heart failure (HF) with high mortality and poor quality of life. Chronic Heart failure (CHF) is the primary reason for discharge for veterans treated within the VA health care system. It is the most common Medicare diagnosis-related group and more Medicare dollars are spent for diagnosis and treatment of CHF than for any other diagnosis. In 2005, the estimated total direct and indirect cost of HF in the U.S. was $27.9 billion. Between 1997 and 2004, the total hospital costs associated with these stays increased from $6,206 million to $8,281 million. Further, in a 2003 study acute exacerbations accounted for 55% of potentially preventable hospitalizations.

OBJECTIVE(S):
The objectives of the research are to study the use of information extraction (IE) techniques to obtain performance metrics for the treatment of inpatient CHF within the veteran population, to develop automated methods to determine guideline-concordant care and to evaluate the potential use of the automated system within the VA.

METHODS:
The study cohort is comprised of 1390 inpatients diagnosed with CHF who were discharged from 8 VA medical centers within Region 1. These patients abstracted for the External Peer Review Program (EPRP) and discharged in fiscal year 2008. The key concepts in the study are based on the External Peer Review Program (EPRP) required elements. These consist of the diagnostic and treatment dimensions, respectively, of left ventricular systolic function (LVSF) assessment, and angiotensin-converting enzyme (ACE) inhibitor, or angiotensin receptor blocker (ARB) therapy for patients with CHF who have left ventricular systolic dysfunction (LVSD) for those patients without contraindications (treatment). In this study we assemble data elements for the CHF performance indicators using structured data from VISTA and information extraction techniques using patient records. We will construct and test automated algorithms to generate data elements for CHF performance indicators. Concordance will be measured at the individual patient level between an automated classifier using algorithms and information extraction and a manually-derived classifier based on EPRP data collection. We will examine correlation at the facility level for percent adherence when calculated using the automated classifier versus EPRP-based review. We will enhance the acceptability and utility of the automation of CHF performance indicators through a stakeholder engagement process using the Promoting Action on Research on Implementation in Health Services (PARIHS) implementation framework in order to provide a draft implementation plan within the VA.

FINDINGS/RESULTS:
The project started on April 1, 2010. We have developed a machine-learning based EF module which when used with echocardiogram reports has above 90% accuracy at capturing ejection fraction value and the binary classification of whether the patient's ejection fraction is less than 40%. We have also determined the sections of documents within the overall document set where a high prevalence of the concepts is found across all documents. The following sections where specific concepts are found: EF in assessment and current history, ACEI/ARB, in assessment and medications, LVSD in echocardiogram results and assessment, and reasons why the patient is not on the medications in assessment.

IMPACT:
This project addresses critical gaps in effective use of electronic information to support clinical care and health services research, as well as the efficiency of measuring performance and quality in the clinical domain of chronic heart failure. To this end, we have begun testing the use of an NLP system to improve efficient extraction of data for automated performance measurement for veterans with congestive heart failure. We have also worked with OQP and clinical experts to determine the clinical business rules and data flows associated with the information our team will provide to their office. And we are engaging with implementation science investigators from the VA Center for Implementation Practice and Research Support (CIPRS) to determine an appropriate draft implementation framework and to design an implementation science-based stakeholder engagement process. We have begun stakeholder engagement to develop a draft implementation plan.

PUBLICATIONS:

Journal Articles

1. Garvin JH, DuVall SL, South BR, Bray BE, Bolton D, Heavirland J, Pickard S, Heidenreich P, Shen S, Weir C, Samore M, Goldstein MK. Automated extraction of ejection fraction for quality measurement using regular expressions in Unstructured Information Management Architecture (UIMA) for heart failure. Journal of the American Medical Informatics Association : JAMIA. 2012 Sep 1; 19:(5):859-66.

Conference Presentations

1. Garvin JH. Detecting Mentions and Values of Left Ventricular Ejection Fraction in Echocardiogram Reports. Poster session presented at: VA HSR&D / QUERI National Meeting; 2012 Jul 16; National Harbor, MD.
2. Garvin JH. Comparing Methods for Left Ventricular Ejection Fraction Clinical Information Extraction. Paper presented at: American Medical Informatics Association Annual Symposium; 2012 Mar 19; San Francisco, CA.

DRA: Health Systems, Cardiovascular Disease
DRE: Research Infrastructure, Diagnosis
Keywords: Cardiovasc’r disease, Clinical Performance Measures, Healthcare Algorithms, Information Management, Natural Language Processing, Quality assessment, Quality Indicators, Research method
MeSH Terms: none