Objectives: In January 2022, EUCAST guidelines recommending replacement of the "intermediate" category with a "susceptible, increased exposure" (SFP) category were implemented in our hospital. We aimed to assess the impact of these changes on antibiotic prescriptions for Pseudomonas aeruginosa and Staphylococcus aureus infections.
Methods: This retrospective before-after study included adult inpatients with monobacterial infections between March-August 2021 (BEFORE) and March-August 2022 (AFTER). Antibiotic use and relevance were compared. Meropenem was masked when imipenem was categorized as SFP.
Results: We included 240 antibiotic susceptibility tests (195 patients). Infectious disease consultations increased significantly during implementation (53.0 % vs. 28.9 %, p = 0.0005). Meropenem prescriptions for P. aeruginosa declined (13.8 %-6.2 %), while high-dose regimens for SFP antibiotics likewise decreased (50.0 %-35.4 %). Overall, prescription appropriateness remained high (>92 %).
Conclusion: The introduction of SFP reporting was associated with increased ID consultation and a trend toward reduced broad-spectrum use, highlighting a need for targeted prescriber education.
Artificial intelligence (AI) is set to permeate every facet of infectious disease practice-from prevention and public health surveillance to epidemic management and bedside care. Routine care data (laboratory results, medication orders, progress notes) and research-generated datasets now fuel state-of-the-art machine-learning (ML) pipelines that sharpen diagnosis, prognosis, antimicrobial stewardship, and, by combining both sources, accelerate drug discovery. In diagnostics, deep networks that now flag pneumonia or tuberculosis on chest images are increasingly able to identify-and localize-virtually more infectious processes throughout the body, while simultaneously predicting pathogen identity and antimicrobial resistance from routine microbiology. Prognostic models trained on Electronic Health Records surpass traditional scores in anticipating clinical deterioration or postoperative sepsis, enabling earlier targeted interventions. Predictive analytics can also personalize antimicrobial dosing by fusing real-time drug-monitoring data. Large language models (LLMs) build upon these advances by transforming unstructured clinical narratives into structured phenotypes suitable for predictive modeling, automatically summarizing patient encounters, generating synthetic cohorts for rare conditions, and providing real-time conversational decision support at the patient's bedside. Despite rapid progress, real-world deployment faces hurdles: high computational and licensing costs, vendor-specific implementation constraints, limited cross-site model transferability, and fragmented governance of safety, bias, and cybersecurity risks. Rigorous, lifecycle-based evaluation frameworks-covering external validation, cost-effectiveness analysis, and post-deployment monitoring-are required to ensure safe, equitable, and sustainable AI adoption. This review synthesizes current applications, evidential strengths, and unresolved challenges, and proposes a translational roadmap aligning technical innovation with clinical and regulatory realities.

