A Systematic Review of AI-Enabled Antimicrobial Stewardship and Therapeutics in Critical Care Medicine

Review article

Abstract

Intensive care units act as epicenters for the selection and dissemination of multidrug-resistant pathogens. High antimicrobial pressure, invasive interventions, and critically ill cohorts drive the selection of resistant strains. A systematic analysis of evidence from January 2020 to December 2025 showed that Gram-negative ESKAPE pathogens, notably carbapenem-resistant Enterobacterales and Acinetobacter baumannii, dominate critical care infections. Mechanistic pathways involve beta-lactamase production, porin mutations such as OprD down-regulation in Pseudomonas aeruginosa, and active efflux systems like the AcrAB-TolC pump. Rapid molecular diagnostic platforms, including polymerase chain reaction assays and matrix-assisted laser desorption ionization-time of flight mass spectrometry, significantly accelerate pathogen identification compared to standard cultures. When integrated with artificial intelligence and machine learning models, clinical decision support tools optimize antibiotic prescriptions, reducing inappropriate administration by twenty to thirty-five percent as measured by days of therapy per one thousand patient-days and strict adherence to institutional antibiograms. Novel drugs, specifically cefiderocol, novel beta-lactam/beta-lactamase inhibitor combinations, and experimental options such as bacteriophages and CRISPR-Cas gene-editing therapies, offer optimistic alternatives against pan-drug-resistant isolates. Effective containment of critical care antimicrobial resistance requires a structured paradigm transition combining real-time genomic surveillance, machine learning risk stratification, and targeted drugs to improve clinical recovery rates while preserving last-resort antimicrobial classes