Direct carbapenemase typing from disc diffusion antibiograms with MALCA (MAchine Learning CArbapenemase)

Area of Science:

• Clinical microbiology
• Machine learning in diagnostics
• Antimicrobial resistance

Background:

• Carbapenemase-producing Enterobacterales (CPE) pose a significant threat due to limited treatment options.
• Accurate and rapid identification of CPE and their specific carbapenemase type is crucial for effective patient management.

Purpose of the Study:

• To develop and validate MALCA, a machine-learning classifier for direct CPE detection and carbapenemase typing.
• To assess MALCA's performance against existing screening algorithms.

Main Methods:

• Development of a stepwise random-forest pipeline using disc diffusion antibiogram data from 11,992 clinical isolates.
• Creation of two classifiers, MALCA-22 and MALCA-8, based on different antibiotic panels.
• External validation on 8,514 isolates.

Main Results:

• MALCA classifiers achieved >96% sensitivity and specificity for CPE detection in external validation.
• High accuracy for identifying prevalent carbapenemases (OXA-48-like, NDM, KPC) with >97% sensitivity and >98% specificity.
• MALCA outperformed European and French CPE screening algorithms.

Conclusions:

• MALCA is a rapid, cost-effective diagnostic tool utilizing existing antibiogram data.
• Enables earlier targeted therapy and diagnostic guidance for CPE infections without extra resources.
• Facilitates improved clinical decision-making in the face of antimicrobial resistance.