ML-based risk assessment tool to rule out empiric use of ESBL-targeted therapy in endemic areas.

BACKGROUND: Antimicrobial stewardship focuses on identifying patients who require ESBL-targeted therapy. Rule-in tools have been extensively researched in areas of low endemicity; however, such tools are inadequate for areas with high rates of ESBL, as almost all patients will be selected.

AIM: To develop a machine learning-based rule-out tool suitable for areas with high levels of resistance.

METHODS: We used gradient boosted decision trees to train and validate a risk prediction model on data from 17,913 (45% ESBL) patients with Escherichia coli and Klebsiella pneumoniae in urine cultures. We evaluated the predictive power of different sets of variables, using Shapley values to evaluate variable contributions.

FINDINGS: Our model successfully identified patients with low risk of ESBL resistance in ESBL-endemic areas (AUC-ROC 0.72). When used to select the 30% of patients with the lowest predicted risk, the model yielded a negative predictive value ≥ 0.74. We also demonstrated that a model with seven input features can perform nearly as well as our full model. This simplified model is freely accessible as a web application.

CONCLUSIONS: Our study demonstrates that a risk calculator for antibiotic resistance can be a viable rule-out strategy to reduce ESBL-targeted therapy usage in ESBL-endemic areas. Robust performance of a model with only limited features makes the clinical use of such a tool feasible. In an era with growing rates of ESBL where some experts have called for empirical use of carbapenems as first-line therapy for all patients in high-ESBL-prevalence areas, our tool provides an important alternative.