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

Summary

Using routine disc diffusion antibiograms, MALCA accurately detects CPE and types carbapenemases with >96% sensitivity and specificity in external validation. This rapid, reagent-free tool outperforms existing screening algorithms and can expedite targeted therapy decisions for severe infections including sepsis.

Key Findings

• Two classifiers (MALCA-22 and MALCA-8) trained on 11,992 isolates and externally validated on 8,514 isolates.
• Both classifiers achieved >96% sensitivity and specificity for CPE detection; >97% sensitivity and >98% specificity for common carbapenemases (OXA-48-like, NDM, KPC).
• Outperformed established European and French CPE screening algorithms using only routine disc diffusion data.

Clinical Implications

Integrating MALCA into laboratory information systems can enable same-day carbapenemase typing from routine antibiograms, guiding early selection of agents (e.g., ceftazidime–avibactam for KPC/OXA-48-like vs. aztreonam–avibactam for NDM) in suspected or confirmed sepsis.

Limitations

• Performance may depend on local antibiotic panels and breakpoints; generalizability to rare carbapenemases is uncertain.
• Clinical impact on time-to-appropriate therapy and outcomes was not prospectively measured.

Future Directions

Prospective implementation trials measuring time-to-appropriate therapy, mortality, and antimicrobial stewardship metrics across diverse settings, including low-resource laboratories.