Direct prediction for carbapenemase-producing and colistin-resistant Klebsiella pneumoniae isolates from routine MALDI-TOF mass spectrum using machine learning

The emergence of carbapenem-nonsusceptible K. pneumoniae (CnSKP) leads a serious threat to patient survival and colistin resistance makes the treatment of CnSKP more difficultly. To make treatment strategy properly and quickly, we aimed to develop a rapid prediction method for CnSKP and colistin-resistant K. pneumoniae (ColRKP) based on the spectra of routine matrix-assisted laser desorption/ionization-time-of-flight mass spectrometry (MALDI–TOF MS). The machine learning (ML) model for differentiating CnSKP and carbapenem-susceptible K. pneumoniae (CSKP) showed accuracy of 0.8869 and AUC of 0.9551; the model for ColRKP and colistin-intermediate K. pneumoniae (ColIKP) showed accuracy of 0.8361 and the AUC of 0.8447.