Predicting appropriateness of antibiotic treatment among ICU patients with hospital acquired infection

Abstract Antimicrobial resistance is a growing threat to global health, leading to ineffective treatment of infection and increasing treatment failure, mortality, and healthcare costs. Inappropriate antibiotic therapy is often administered in the Intensive Care Unit (ICU) due to the urgency of treatment, but can lead to poor patient outcomes. In this study, we developed a machine learning model that predicts the appropriateness of antibiotic treatments for ICU inpatients with ICU-acquired blood infection. We analyzed data from electronic medical records (EMRs), including demographics, administered drugs, previous microbiological cultures, invasive procedures, lab measurements and vital signs. Since EMRs have high rates of missing values and since our cohort is relatively small and imbalanced, we introduced novel computational methods to address these issues. The final model achieved an AUROC of 82.8% and an AUPR of 60.6% on the training set and an AUROC score of 77.3% and an AUPR score of 40.4% on the validation set. Our study shows the potential of machine learning models for inappropriate antibiotic treatment prediction.