Artificial Intelligence model to predict resistances in Gram-negative bloodstream infections

Abstract

Artificial intelligence (AI) models are promising tools for predicting antimicrobial susceptibility in gram-negative bloodstream infections (GN-BSI). Single-center study on hospitalized patients with GN-BSI, over 7-year period, aimed to predict resistance to fluoroquinolones (FQ-R), third generation cephalosporins (3GC-R), beta-lactam/beta-lactamase inhibitors (BL/BLI-R) and carbapenems (C-R) was performed. Analyses were carried out within a machine learning framework, developed using the scikit-learn Python package. Overall, 2552 patients were included. Enterobacterales accounted for 85.5% of isolates, with E. coli, Klebsiella spp, and Proteus spp being most common. Distribution of resistance was FQ-R 48.6%, 3GC-R 40.1%, BL/BLI-R 29.9%, and C-R 16.9%. Models' validation showed good performance predicting antibiotic resistance for all four resistance classes, with the best performance for C-R (AUC-ROC 0.921 ± 0.013). The developed pipeline has been made available ( https://github.com/EttoreRocchi/ResPredAI ), along with documentation for running the same workflow on a different dataset, to account for local epidemiology and clinical features.

Fig. 1. Mean values of the most relevant coefficients of the logistic regressions, over the 10 iterations of the outer cross-validation, for the four antibiotic resistances.

Each panel shows the mean values of the 10 largest coefficients (in module, the positive ones, related to resistance, represented in red, the negative ones in green) of the logistic regressions, over the 10 iterations of the outer cross-validation. There is one panel for each of the four antibiotic resistances. The blue error bar of each coefficient represents its standard deviation value over the 10 cross-validation iterations.

Fig. 2. Mean confusion matrices and performance metrics over the outer cross-validation iterations for different types of antibiotic resistance.

Each panel shows the mean row-normalized confusion matrices over the 10 iterations of the outer cross-validation. The mean value of the weighted F1-score, Matthews Correlation coefficient and Area Under Receiver Operating Characteristic Curve (AUROC) are reported for each antibiotic resistance; the associated uncertainty is the standard deviation of the metrics over the above mentioned 10 iterations.