Prevention is all you need: generative artificial intelligence for infection prevention and healthcare epidemiology.

Abstract

Purpose of review: Generative artificial intelligence, particularly large language models, has emerged as a promising tool for processing the vast amounts of unstructured clinical data generated in healthcare. Infection prevention and control relies heavily on manual review of clinical text for surveillance, reporting, and risk assessment, making it a natural application area. We examine the current evidence for these technologies in infection prevention workflows.

Recent findings: Large language models have been evaluated for healthcare-associated infection surveillance, including central line-associated bloodstream infections, surgical site infections, and catheter-associated urinary tract infections, with pooled sensitivities exceeding 90% across studies. These tools have also been applied to diagnostic stewardship, risk assessment for multidrug-resistant organism exposure, public health surveillance for avian influenza, and central line necessity auditing. Across applications, models performed best when used to augment rather than replace expert review. Common limitations included reduced specificity, sensitivity to prompt framing, and dependence on the completeness of clinical data provided to the model.

Summary: Generative artificial intelligence applications show the greatest promise when aligned with their core capability: natural language processing of unstructured clinical text. While current evidence supports their use as screening and decision-support tools with human oversight, further validation across diverse settings and integration within electronic health records are needed before widespread adoption.