Retrospective validation study of a machine learning-based software for empirical and organism-targeted antibiotic...

Area of Science:

• Infectious Diseases
• Medical Informatics
• Pharmacology

Background:

• Antibiotic prescription errors are common, leading to inadequate infection coverage.
• Machine learning offers potential solutions for optimizing antibiotic selection.

Purpose of the Study:

• To assess the efficacy of iAST, a machine-learning-based software, in providing accurate antibiotic recommendations.
• To compare iAST's recommendations against physician prescriptions for empirical and organism-targeted therapy.
• To evaluate iAST's impact on antibiotic stewardship.

Main Methods:

• Retrospective validation of iAST using data from 325 patients with acute infections across 12 Spanish hospitals.
• Fine-tuning the iAST model with 27,531 historical antibiograms.
• Comparing the success rates of iAST's top three recommendations against physician-prescribed antibiotics, confirmed by antibiogram results.

Main Results:

• iAST's top three recommendations showed non-inferiority to physician prescriptions for empirical therapy (91.06% success vs. 68.93%) and organism-targeted therapy (97.83% vs. 84.16%).
• iAST demonstrated improved antibiotic stewardship, recommending more access and reserve antibiotics and fewer watch antibiotics.
• Statistical significance (P < 0.001) was observed for all primary endpoint comparisons.

Conclusions:

• iAST accurately predicts antibiotic susceptibility, improving prescription outcomes.
• The software shows significant potential for enhancing antibiotic stewardship practices.
• iAST can be a valuable tool in combating antimicrobial resistance through optimized therapy selection.