USR PUBLISHER

Healthcare-associated infections (HAIs) remain a leading cause of preventable morbidity, mortality, and excess healthcare spending worldwide, and traditional manual surveillance is labor-intensive, inconsistent, and often too slow to support timely intervention. Artificial intelligence (AI) including machine learning (ML), natural language processing (NLP), and hybrid rule-based/statistical systems have emerged as a candidate solution for automating and accelerating HAI detection using data already captured in electronic health records (EHRs). This narrative review synthesizes recent evidence on AI-driven infection surveillance across major HAI categories, including sepsis, surgical site infections (SSIs), device-associated infections, and outbreak-level public health surveillance. Across studies, AI-augmented surveillance systems have demonstrated the ability to match or exceed the sensitivity of manual chart review while substantially reducing reviewer workload, and NLP applied to unstructured clinical notes has repeatedly improved detection accuracy relative to structured data alone. However, performance is highly heterogeneous across institutions, infection types, and algorithms, and several widely deployed proprietary tools, most notably early sepsis early-warning systems have shown poor external validity, low positive predictive value, and a tendency to generate clinically disruptive alert fatigue. Persistent barriers include limited model portability across health systems, opaque ("black box") decision logic, inconsistent reference-standard definitions, and the absence of standardized regulatory or reporting frameworks for surveillance-oriented AI. We conclude that AI-driven infection surveillance can meaningfully augment infection prevention programs when rigorously externally validated, transparently reported, and implemented as a human-in-the-loop workflow rather than a fully autonomous decision system, and we outline priority areas for future research and implementation science.