Machine learning models for coagulation dysfunction risk in inpatients administered β-lactam antibiotics.

Area of Science:

• Pharmacology
• Medical Informatics
• Data Science

Background:

• β-Lactam antibiotics are widely used but carry a risk of coagulation dysfunction.
• Proactive assessment of this risk is often overlooked.
• Machine learning offers potential for evaluating this risk.

Purpose of the Study:

• Identify risk factors for β-lactam-associated coagulation dysfunction.
• Develop machine learning models to estimate this risk using real-world data.

Main Methods:

• Retrospective study using electronic health records (EHR) from 45,179 adult inpatients (2018-2021).
• Employed five distinct machine learning methods for risk prediction.
• Validated models using 5-fold cross-validation and test sets, assessing performance via AUC.

Main Results:

• Incidence of coagulation disorders varied by antibiotic: cefazolin (2.4%), cefoperazone/sulbactam (5.4%), cefminol (1.5%), amoxicillin/sulbactam (5.5%), piperacillin/tazobactam (4.8%).
• Optimal machine learning models achieved AUCs ranging from 0.768 to 0.919.
• Models demonstrated effectiveness in predicting risk for specific β-lactam antibiotics.

Conclusions:

• Machine learning classifiers are valuable tools for identifying patients at risk of β-lactam-induced coagulation dysfunction.
• High-risk predictions enable timely intervention and improved pharmacovigilance.
• Integrating more administrative and clinical data can enhance model predictive performance.