AI tools might be able to diagnose healthcare-associated infections, but human oversight is vital to ensuring patient safety, according to a study published March 13 in the American Journal of Infection Control.
Artificial intelligence has largely been implemented for behind-the-scenes work in healthcare, but many are probing whether the AI can aid clinicians with diagnoses and treatment plans.
Researchers at Saint Louis University and the University of Louisville (Ky.) School of Medicine analyzed how well two AI-powered tools could identify HAIs. They tested ChatGPT's HAI Assist and Mixtral 8x7B, an open-source large language model, against six fictional patient scenarios.
Each AI tool received information on a patient's age, symptoms, date of admission, and dates when central lines or catheters were inserted or removed. HAI Assist and Mixtral 8x7B were then asked whether the descriptions matched a central line-associated bloodstream infection or a catheter-associated urinary tract infection.
Both tools accurately diagnosed every HAI, but when information was missing — such as a description lacking when a catheter was inserted — the AI tool failed.
"Abbreviations, lack of specificity, use of special characters and dates reported in numeric format instead of with the month spelled out all led to inconsistent responses," according to a March 14 news release from the Association for Professionals in Infection Control and Epidemiology.
From 2021 to 2022, the average standard infection ratios for CLABSI increased by 60% and CAUTI increased 19%, according to The Leapfrog Group.
"Our results are the first to demonstrate the power of AI-assisted HAI surveillance in the healthcare setting, but they also underscore the need for human oversight of this technology," Timothy Wiemken, PhD, a professor at Saint Louis University and lead author, said in the release. "With the rapid evolution of the role of AI in medicine, our proof-of-concept study validates the need for continued development of AI tools with real-world patient data to support infection preventionists."