Background: Clinical practice guidelines recommend the use of invasive care to treat acute coronary syndrome (ACS) in most patients with chronic kidney disease (CKD). Preliminary unadjusted results from our pilot work indicate that such care is commonly not provided to Veterans with CKD. However, these preliminary findings did not account for the appropriateness of invasive care. To address this gap, we submitted a VA HSR&D Merit Review proposal last year to examine whether invasive care was implemented less frequently in Veterans with CKD accounting for the appropriateness of invasive care. We proposed assessing the appropriateness of invasive care using Global Registry of Acute Coronary Events (GRACE) scores derived from VA administrative data and using natural language processing (NLP) to extract component variables contained as free text in the VA Corporate Data Warehouse (CDW). A key issue raised in the review of our proposal was the need for proof of concept testing of GRACE scores calculated using this approach. Accordingly, we propose to conduct a pilot study to develop a method to calculate GRACE scores in this manner and assess their accuracy. Significance/Impact: By developing a method to calculate GRACE scores using VA administrative data and NLP, our proposal addresses the VA priorities to advance research methods (i.e., Data/Measurement Science) that cut across conditions and/or care settings and to transform VA data into a national resource. Furthermore, as our long-term goal is to ensure the equitable provision of evidence-based care for ACS in Veterans with CKD, this pilot is the first step in a series of studies that will address multiple VA HSR&D priority areas, including management of complex chronic diseases, access to care, health equity, and quality and safety of heath care. Innovation: Our use of NLP to extract free text from VA administrative data and incorporate those data elements into the calculation of GRACE scores is highly innovative and has broad relevance to health services research. Specific Aim 1: To develop a precise method for retrospective GRACE score calculation in Veterans with and without CKD who were hospitalized with ACS using standard administrative data extraction from the VA CDW combined with rule-based and machine-learning NLP techniques for extraction of free text from the CDW. Specific Aim 2: To assess the accuracy of GRACE scores calculated using administrative and free text data from the CDW by comparing to reference standard scores derived from manual chart reviews. Methodology: We will use a national cohort of Veterans with and without CKD who were hospitalized within VA with ACS between 1/1/2013 and 12/31/17. We will calculate GRACE scores for each patient’s ACS hospitalization by extracting 6 of the 8 variables of the score from the CDW and using NLP to extract the remaining 2 free text variables from text integration utility notes in CDW files. For 300 randomly selected patients, we will perform manual chart reviews of their ACS hospitalization to calculate GRACE scores based on data contained in the medical record. We will assess the level of agreement in the identification of each of the 8 individual component variables between administrative/NLP data extraction and manual chart review for these 300 patients. The findings from these analyses will inform modifications in the administrative/NLP data extraction process. We will then recalculate GRACE scores for our cohort and compare the recalculated administrative/NLP derived GRACE scores with manually derived GRACE scores for the 300 patients to confirm their accuracy. Next Steps/Implementation: We will incorporate the findings from this pilot study into the resubmission of our Merit Review proposal that will seek to: (1) confirm the under-utilization of invasive care in Veterans with ACS and CKD adjusting for the appropriateness of such care with administrative/NLP derived GRACE scores; (2) evaluate associated clinical outcomes and; (3) elucidate the factors underlying this practice.
NIH Reporter Project Information
None at this time.
Technology Development and Assessment, TRL - Applied/Translational
Cardiovascular Disease, Decision Support
None at this time.