Internal Validation of a Machine Learning-Based CDSS for Antimicrobial Stewardship.

Area of Science:

• Biotechnology
• Artificial Intelligence
• Infectious Diseases

Background:

• Antimicrobial resistance (AMR) necessitates effective antimicrobial stewardship programs (ASPs).
• Limited resources often impede ASP implementation.
• A machine learning (ML)-driven clinical decision support system (CDSS) was developed to optimize antimicrobial prescribing.

Purpose of the Study:

• To validate a novel ML-driven CDSS for guiding antimicrobial prescribing.
• To assess the system's accuracy in distinguishing trained from novel data.
• To evaluate the clinical consistency of the system's recommendations.

Main Methods:

• Prospective observational studies and retrospective analysis were used for validation.
• The ML system's performance was tested on distinguishing trained versus novel data using BioFire molecular panel inputs and internal lab results.
• Independent clinicians reviewed the system's recommendations against standard clinical guidelines.

Main Results:

• The ML system achieved 100% accuracy in identifying novel data points.
• It correctly identified all trained and novel complex datasets.
• Clinical recommendations showed no major discrepancies, with minor differences in only 100 out of 644 reports.

Conclusions:

• The novel ML system demonstrates high accuracy in data differentiation.
• Its recommendations align with established clinical guidelines.
• The system shows significant potential for enhancing CDSS and ASPs to combat AMR.