Novel Method to Detect Cardiac Device Infections by Integrating Electronic Medical Record Text with Structured Data in the Veterans Affairs Health System. (October 2020)

• Record Type: Journal Article
• Authors: Mull, Hillary; Stolzmann, Kelly; Shin, Marlena; Kalver, Emily; Schweizer, Marin; Branch-Elliman, Westyn
• Is Part Of: Infection control and hospital epidemiology. Volume 41(2020)Supplement 1
• Page Start: s332
• Page End: s332
• Publication Date: 2020-10
• Subjects: Nosocomial infections -- Epidemiology -- Periodicals; Health facilities -- Sanitation -- Periodicals; Hospital buildings -- Sanitation -- Periodicals; Cross Infection -- Periodicals; Epidemiology -- Periodicals; Hospitals -- Periodicals; Infection Control -- Periodicals; 614.44
• DOI: 10.1017/ice.2020.936
• Languages: English
• ISSNs: 0899-823X

Abstract

Abstract : Background: Cardiovascular implantable electronic device (CIED) infections are highly morbid, yet infection control resources dedicated to preventing them are limited. Infection surveillance in outpatient care is also challenging because there are no infection reporting mandates, and monitoring patients after discharge is difficult. Objective: Thus, we sought to develop a replicable electronic infection detection methodology that integrates text mining with structured data to expand surveillance to outpatient settings. Methods: Our methodology was developed to detect 90-day CIED infections. We tested an algorithm to accurately flag only cases with a true CIED-related infection using diagnostic and therapeutic data derived from the Veterans Affairs (VA) electronic medical record (EMR), including administrative data fields (visit and hospital stay dates, diagnoses, procedure codes), structured data fields (laboratory microbiology orders and results, pharmacy orders and dispensed name, quantity and fill dates, vital signs), and text files (clinical notes organized by date and type containing unstructured text). We evenly divided a national dataset of CIED procedures from 2016–2017 to create development and validation samples. We iteratively tested various infection flag types to estimate a model predicting a high likelihood of a true infection, defined using chart review, to test criterion validity. We then applied the model to the validation data and reviewed cases with high and low likelihood of infection to assess performance. Results: The algorithm development sample included 9, 606 CIED procedures in 67 VA hospitals. Iterative testing over 381 chart reviewed cases with 47 infections produced a final model with a C-statistic of 0.95 (Table 1). We applied the model to the 9, 606 CIED procedures in our validation sample and found 100 infections of the 245 cases identified by the model to have a high likelihood of infection We identified no infections among cases the model as having low likelihood. The final model included congestive heart failure and coagulopathy as comorbidities, surgical site infection diagnosis, a blood or cardiac microbiology order, and keyword hits for infection diagnosis and history of infection from clinical notes. Conclusions: Evolution of infection prevention programs to include ambulatory and procedural areas is crucial as healthcare delivery is increasingly provided outside traditional settings. Our method of algorithm development and validation for outpatient healthcare-associated infections using EMR-derived data, including text-note searching, has broad application beyond CIED infections. Furthermore, as integrated healthcare systems employ EMRs in more outpatient settings, this approach to infection surveillance could be replicated in non-VA care. Funding: None Disclosures: None