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

Carbapenemase-producing Enterobacterales (CPE) present limited therapeutic options. Optimal treatment requires identifying the carbapenemase type, often requiring confirmatory testing beyond routine susceptibility results. We develop MALCA, a machine-learning classifier that uses routine disc diffusion antibiogram results to directly detect CPE and identify the carbapenemase type. From 11,992 clinical isolates, we build a stepwise random-forest pipeline and derive two classifiers based on panels of 22 or 8 antibiotics (MALCA-22 and MALCA-8). In an external validation study involving 8514 isolates, both MALCA classifiers achieved sensitivity and specificity >96% for CPE detection, outperforming European and French algorithms developed for CPE screening. For the most prevalent carbapenemases, MALCA achieve sensitivities exceeding 97% and specificities above 98%, particularly for OXA-48-like, NDM, and KPC producers. MALCA is a rapid, and inexpensive diagnostic tool that uses solid antibiogram data to detect and type CPE, enabling earlier targeted therapy and diagnostic guidance without additional reagents or human resources.