Artificial intelligence for analysing antibiotic use: an exploratory study in a tertiary care paediatric hospital in Italy

BACKGROUND: Monitoring antibiotic consumption and resistance is central to antimicrobial stewardship (AMS). However, this is particularly challenging in paediatrics due to weight-based dosing and heterogeneous patient populations. OBJECTIVES: To explore the potential role of artificial intelligence (AI) in supporting the evaluation of antibiotic consumption and antimicrobial resistance within a paediatric tertiary-care paediatric hospital in Italy. METHODS: A retrospective analysis (2020-2024) was conducted at the IRCCS Istituto Giannina Gaslini, Genoa. Data included bloodstream infections (BSIs) caused by Staphylococcus aureus and Enterobacterales, intravenous formulations of antibiotic dispensing (expressed as defined daily doses [DDD], stratified according to the World Health Organization AWaRe classification), and both static and adaptive drug resistance indices (DRIs). Spearman's correlation was applied to assess associations. AI (Google Gemini v2.5 Flash) was used for data organization and preliminary analyses; and all outputs were verified by the authors. RESULTS: Use of Access antibiotics increased progressively, exceeding 60% of prescriptions in 2024, whereas Watch antibiotic use decreased to approximately 30%. Reserve antibiotic use fluctuated but showed an overall upward trend. Despite increasing hospital activity, normalized antibiotic consumption (DDD/1000 discharges or patient-days) remained stable or declined. For S. aureus BSIs, methicillin-resistant S.aureus (MRSA) rates and DRIs both declined. Enterobacterales displayed mixed trends, with some resistance indicators increasing, although overall resistance decreased. Most correlations were not statistically significant. CONCLUSIONS: AI-assisted analyses identified stewardship-relevant trends, but expert oversight remained essential. Broader multicentre paediatric studies incorporating patient-level outcomes are warranted to confirm the feasibility and validity of AI-assisted antimicrobial surveillance.