Researchers develop new algorithm for predicting CIED infections
Researchers have developed an algorithm that can predict when infections may occur following cardiovascular implantable electronic device (CIED) procedures, sharing their findings in JAMA Network Open.
Healthcare-associated infections (HAIs) are a massive issue, leading to more than 99,000 patient deaths annually in the United States alone, and recent research suggests they are on the rise. Providers are fully aware of this challenge, and HAI surveillance is common, but detecting HAIs is no simple task.
The study’s authors built their algorithm from the ground up, using EMR data from Veterans Health Administration (VA) for CIED procedures from October 2016 to September 2017. CIED infections were defined as device pocket infections, lead infections or endocarditis, and any infection present at the time of the procedure was excluded.
To develop the algorithm, the team started with commonly documented red flags that suggest an infection—a high temperature or antibiotics being prescribed for more than three days, for example—and then considered “other factors that might affect the presence of infection-related flags, including comorbidities, emergent procedure and death within 90 days.” Cases flagged by the algorithm were then assessed by manual reviewers, slowly helping the team perfect its model. The algorithm was then validated using data from more than 9,000 cases.
In that validation set, the algorithm achieved an overall sensitivity, specificity and positive predictive validity of 94.4%, 48.8% and 41.4%, respectively.
“Infection prevention programs outside of inpatient care settings are limited even though most surgical and nonsurgical procedures occur in outpatient settings,” wrote lead author Hillary J. Mull, PhD, department of surgery at Boston University School of Medicine, and colleagues. “Thus, there is a broad opportunity to reduce HAIs by focusing more attention on invasive outpatient procedural care. We developed a novel method using a combination of structured data and keyword searches of unstructured clinical text data to flag potential infections in cardiac device procedures. This new method demonstrated strong predictive value for measuring true infections and can be used as a model system for expanding HAI surveillance activities to currently uncovered areas with limited dedicated resources.”
The full study is available here.
Michael has more than 19 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.