Evaluating AI-assisted surveillance of healthcare-associated infections using a standard European point prevalence study in a regional hospital, the Czech Republic, 2025.

BACKGROUND: Healthcare-associated infections (HAI) remain a major burden in acute care hospitals. Traditional passive surveillance systems underestimate HAI incidence and lack reliability. Daily manual review of health records for HAI detection is not feasible. Artificial intelligence (AI) supported HAI surveillance system HAIDI for daily routine was implemented in the Regional Hospital Liberec in 2023. AIM: To evaluate the methodology and the performance of an AI-assisted surveillance system (HAIDI) by comparing its results with manual point prevalence survey (PPS). METHODS: A three-step study was conducted at a large regional hospital. First, a one-day PPS was performed using ECDC methodology. Second, HAIDI was applied to the same patient cohort, and its outputs were validated by epidemiologists. Then Cohen's kappa and the detection proportion against the combined set of infections identified by either method were calculated. RESULTS: Among 749 patients, manual PPS identified 61 patients with 65 HAI (prevalence 8.1%), while HAIDI identified 62 patients with 65 HAI. Agreement between the two approaches was high (Cohen's kappa κ = 0.8849; observed agreement 0.9826). Manual PPS and HAIDI both missed seven HAI but with different HAI range. The detection proportion against the combined set of infections identified by either method was 90.3% for both manual PPS and HAIDI. CONCLUSION: Evaluation methodology proved useful and AI-assisted surveillance demonstrated high concordance with manual PPS.