Development of a predictive model for eosinophilia in patients treated with ampicillin/sulbactam or...

We aim to identify risk factors for antibiotic-induced eosinophilia in hospitalized patients receiving penicillin/beta-lactamase inhibitor therapy and to develop a machine learning-enhanced predictive risk-scoring model.

A retrospective cohort study was conducted involving 490 hospitalized patients treated with intravenous ampicillin/sulbactam or piperacillin/tazobactam. After applying eligibility criteria, 378 patients were included in the final analysis. Multivariate logistic regression was used to identify independent risk factors for eosinophilia, which were incorporated into a clinical risk-scoring system. To improve predictive accuracy, machine learning models including elastic net regression were developed and evaluated using five-fold cross-validation. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC).

Among the 378 patients, 27 (7.1%) developed eosinophilia. Nine independent risk factors were identified: aminoglycosides, arrhythmias, chlorpheniramine maleate, glycopeptides, levetiracetam, polystyrene sulfonate, rheumatoid arthritis, stroke and ursodeoxycholic acid. The risk score ranged from 0 to 8, corresponding to predicted eosinophilia probabilities of 3.3% to 93.8% (AUROC = 0.787). Among the machine learning approaches tested, elastic net regression achieved the highest predictive performance.

This study proposes a clinically applicable risk-scoring model for predicting eosinophilia in patients treated with penicillin/beta-lactamase inhibitors. Incorporating machine learning improved model accuracy, enabling more precise identification of high-risk patients. The model supports proactive risk assessment and may inform tailored monitoring and intervention strategies in clinical settings.