FIG 5.

(a) Schematic illustration of the application of the VREfm model. A timeline of the bacterial culture testing using currently used clinical tests (i.e., traditional approach) and a modified timeline with the VREfm model incorporated (i.e., ML approach) are shown. In the traditional approach, specimens are collected for bacterial culture test. Usually, 1 day is needed for growth of a single colony for species identification (by MALDI-TOF MS). Vancomycin AST for VREfm requires another 1 day. In contrast, in the ML approach, the VREfm model can provide preliminary AST results at the time when the bacterial species is identified by MALDI-TOF MS. In the treatment of VREfm, the ML approach can improve the accuracy of antibiotic use. Meanwhile, the turnaround time of the bacterial culture test can be reduced to 1 day, which is a 50% reduction. (b) Schematic illustration of the study design. The study included several steps, i.e., data collection, data preprocessing, predictor candidate extraction and important predictor selection, and model training, evaluation, and testing. Data were obtained from two tertiary medical centers (Linkou and Kaohsiung branches of CGMH). The data included mass spectra and results of the vancomycin susceptibility testing of E. faecium strains. Data from the CGMH Linkou branch were used for model training and validation, while data from the CGMH Kaohsiung branch served as independent testing data. In the steps of data preprocessing and predictor candidate extraction and important predictor selection, a specific set of crucial predictors was used for model training. k-fold, timewise CV, and external validation were used to confirm the models’ robustness. The VREfm prediction model can detect VREfm accurately at least 1 day earlier than the current method can.