Christopher Manrodt1, Anne B Curtis2, Dana Soderlund3, Gregg C Fonarow4. 1. Medtronic, plc, Mounds View, MN, United States. Electronic address: Christopher.m.manrodt@medtronic.com. 2. Jacobs School of Medicine and Biomedical Sciences, University at Buffalo, NY, United States. 3. Medtronic, plc, Mounds View, MN, United States. 4. Ahmanson-UCLA Cardiomyopathy Center, Ronald Reagan UCLA Medical Center, Los Angeles, CA, United States.
Abstract
BACKGROUND: Implantable cardioverter-defibrillators (ICDs) have been shown to reduce sudden cardiac death in appropriately selected patients, but they remain underutilized among indicated patients. OBJECTIVE: To develop a new approach to identifying guideline indications among patients implanted with ICDs by creating algorithms that extract data from electronic health records (EHR). METHODS: Published guidelines providing recommendations for ICD use were distilled into categories of diagnoses, measures, procedures, and terminologies. Criteria for each indication category were translated into clinical algorithms using administrative codes, search terms, and other required data. Cardiologists with guideline-development expertise reviewed these algorithms. After developing applications using a subset of data, phenotypes were evaluated against a curated Optum® de-identified EHR dataset, including 94,441 patients with ≥1 procedure codes for ICD implantation or follow-ups from 47 US provider networks. RESULTS: Guideline-concordant indications were identified in 83.7 % of 49,560 patients with new ICD implants. The percentage of ICD patients with guideline-concordant indications ranged from 69.4%-88.1% for patients whose initial EHR records were 0-6 days to >365 days prior to implant, respectively. Many guideline criteria used data which could only be derived from unstructured provider notes and required significant algorithm development. CONCLUSIONS: Defibrillator implant indications were detected in >80 % of patients receiving ICDs using rule-based algorithms in a curated EHR dataset. Computable phenotypes may enable researchers to analyze EHRs in more reproducible ways, by identifying guideline indications in patients with specific therapies such as ICDs, and, by extension, identifying patients who meet indications yet do not yet have indicated therapies.
BACKGROUND: Implantable cardioverter-defibrillators (ICDs) have been shown to reduce sudden cardiac death in appropriately selected patients, but they remain underutilized among indicated patients. OBJECTIVE: To develop a new approach to identifying guideline indications among patients implanted with ICDs by creating algorithms that extract data from electronic health records (EHR). METHODS: Published guidelines providing recommendations for ICD use were distilled into categories of diagnoses, measures, procedures, and terminologies. Criteria for each indication category were translated into clinical algorithms using administrative codes, search terms, and other required data. Cardiologists with guideline-development expertise reviewed these algorithms. After developing applications using a subset of data, phenotypes were evaluated against a curated Optum® de-identified EHR dataset, including 94,441 patients with ≥1 procedure codes for ICD implantation or follow-ups from 47 US provider networks. RESULTS: Guideline-concordant indications were identified in 83.7 % of 49,560 patients with new ICD implants. The percentage of ICDpatients with guideline-concordant indications ranged from 69.4%-88.1% for patients whose initial EHR records were 0-6 days to >365 days prior to implant, respectively. Many guideline criteria used data which could only be derived from unstructured provider notes and required significant algorithm development. CONCLUSIONS: Defibrillator implant indications were detected in >80 % of patients receiving ICDs using rule-based algorithms in a curated EHR dataset. Computable phenotypes may enable researchers to analyze EHRs in more reproducible ways, by identifying guideline indications in patients with specific therapies such as ICDs, and, by extension, identifying patients who meet indications yet do not yet have indicated therapies.