Chronic Heart failure (CHF) is a highly prevalent discharge diagnosis for Veterans treated within the VA health care system. Greater than 90% of Veterans with CHF are on guideline-recommended medications at inpatient discharge, yet not all Veterans are at doses aligned with guideline-directed medical therapy. Patients with CHF have better symptom management and improved cardiovascular outcomes when treated in accordance with the guidelines. Preliminary analyses of VISN 12 data, undertaken during preparation of the proposal, indicated that only one half of patients with CHF were on a beta blocker, and among those patients, doses of beta blockers prescribed were only half of those recommended by the CHF management guidelines. We proposed to address this gap in guideline-directed medical therapy by developing a clinical reminder, using specific patient data and actionable tools at the point of care.
The main objective of this research was to provide an accurate and effective communication aid to improve beta blocker titration to assist with the delivery of guideline-recommended care. We aimed to provide key clinical data along with recommended beta blocker dose to the PACT team at the time of outpatient visit. The specific aim was to develop a communication aid using an automated information extraction system that identifies CHF candidates for beta blocker titration. This aim was accomplished by (Step 1) establishing the accuracy of existing algorithms in the VISN 12 data set and capturing beta blocker data and (Step 2) identifying optimal means of delivering the communication aid to the PACTs.
We used natural language (NLP) and Information Extraction techniques as well as Human Factors analysis to create decision support for providers delivered at the right time in the clinical workflow through the VA electronic health record. For Aim 1, Step 1, we used documents from patients in a retrospective cohort to extend the extraction abilities of our information extraction system, Congestive Heart Failure Extraction Framework (CHIEF), to documents in VISN 12. Our goal was to accurately obtain ejection fraction as well as reasons why the patient should not be on a beta blocker and to examine the accuracy of the NLP for each patient by comparing NLP results to a reference standard. We also aimed to obtain patient-specific data such as heart rate, blood pressure, and beta blocker doses from existing electronic health record data. We partnered with clinicians to develop needed decision support and to identify the delivery mechanism of this support using Cognitive Task Analysis (CTA) and Usability Assessment (UA).
Aim 1 Step 1 Findings: We completed the extension of the existing CHIEF NLP system and found the following performance metrics: for ejection fraction extraction, recall of 96.7%; precision of 97%; and F1-measure of 96.85. For extraction of medications we found: recall of 97.9%; precision of 99%; and F1- measure of 98.5%. In terms of reasons why the providers did not prescribe beta blocker medications we found: recall of 86.2%; precision of 84.7%; and F1-measure of 85.5%. These results indicate good performance with respect to algorithmically identifying the relevant data needed for the decision support tool using NLP. The concepts we extracted are comparable to concepts found in the Unified Medical Language System (UMLS) and biomedical ontologies. While a mapping of the concepts we extracted to external knowledge bases is beyond the scope of funded proposal, mapping of concepts with subsequent output from CHIEF could be accomplished with additional funding.
Aim 1 Step 2 Findings: The CTA interviews indicated that a sizable proportion of PCPs were not familiar with the recommended doses or specific types of Beta Blockers for CHF, and some were uncomfortable with managing treatment in even fairly simple cases. Challenges included time pressures, difficulty accessing clinical data in the EHR, and ambiguity about coordination with other services.
To address the issues of, time pressures, difficulty accessing clinical data in the EHR, and ambiguity about coordination with other services we developed decision support, in the form of a clinical reminder, to identify patients at risk of under-treatment, and for titration if appropriate. Our design for the clinical reminder addresses potential knowledge and attitude gaps by presenting a short summary of the guideline, including the benefits of efficacious beta blocker medications, as well as guideline-directed target doses. The clinical reminder facilitates efficient access to clinical data by providing personalized patient-specific information, some of which is otherwise time-consuming to find (i.e., ejection fraction), as well as up-to-date medication, vital sign, and other clinical information from the Veterans Health Information Systems and Technology Architecture (VistA) files. It could conceivably improve clinician self-efficacy by providing access to guidance on Beta Blocker titration procedures as well as options to utilize other services such as cardiology and pharmacy consults.
Two rounds of UA evaluations with providers were conducted (4 were undertaken remotely with 9 completed on-site with VISN-12 providers). The preliminary results confirm that our clinical reminder provides patient-specific clinical information that providers find useful and needed when considering a CHF patient for Beta Blocker titration. The patient-specific clinical information provided includes blood pressure, weight trends, and important contra-indications. The feedback indicates that the prototype facilitates the assessment for Beta Blocker titration, and guideline-informed care.
We developed decision support in this two year Phase 1 research to improve beta blocker titration according to guideline-recommended care in our national VA network of hospitals, thereby working toward reducing Veteran readmissions and improving Veteran outcomes.
- Smith MW, Brown C, Virani SS, Weir CR, Petersen LA, Kelly N, Akeroyd J, Garvin JH. Incorporating Guideline Adherence and Practice Implementation Issues into the Design of Decision Support for Beta-Blocker Titration for Heart Failure. Applied clinical informatics. 2018 Jun 27; 9(2):478-489.
- Meystre SM, Kim Y, Heavirland J, Williams J, Bray BE, Garvin J. Heart Failure Medications Detection and Prescription Status Classification in Clinical Narrative Documents. Studies in health technology and informatics. 2015 Jan 1; 216:609-13.
- Hysong SJ, Woodard L, Garvin JH, Murawsky J, Petersen LA. Publishing protocols for partnered research. Journal of general internal medicine. 2014 Dec 1; 29 Suppl 4:820-4.
- Garvin JH, Redd A, Bolton D, Graham P, Roche D, Groeneveld P, Leecaster M, Shen S, Weiner MG. Exploration of ICD-9-CM coding of chronic disease within the Elixhauser Comorbidity Measure in patients with chronic heart failure. Perspectives in health information management / AHIMA, American Health Information Management Association. 2013 Oct 1; 10:1b.
- Smith MW, Morison AM, Brown C, Garvin J. Requirements for a proposed distributed attention system for supporting awareness of omissions in healthcare. Poster session presented at: Mexican Conference on Human-Computer Interaction; 2014 Nov 2; Oaxaca, Mexico.