Determining steady-state trough range in vancomycin drug dosing using machine learning.

Area of Science:

• Pharmacology
• Clinical Pharmacy
• Machine Learning in Medicine

Background:

• Vancomycin, a critical antibiotic in ICUs, has a narrow therapeutic window and is nephrotoxic.
• Achieving therapeutic vancomycin trough levels is crucial for efficacy and safety.
• Predicting and managing vancomycin dosing is essential for critically ill patients.

Purpose of the Study:

• To develop and compare machine learning models for predicting vancomycin trough levels.
• To assess the risk of sub-therapeutic and supra-therapeutic vancomycin levels.
• To improve the accuracy of vancomycin dosing in intensive care unit (ICU) patients.

Main Methods:

• Utilized a cohort of 5337 vancomycin courses from adult ICU patients (2007-2017).
• Trained and compared various classification models using categorized vancomycin trough levels (sub-therapeutic, therapeutic, supra-therapeutic).
• Evaluated model performance using AUC-ROC, specificity, and sensitivity.

Main Results:

• XGBoost models demonstrated superior performance compared to other machine learning approaches.
• Achieved AUC-ROC of 0.85 and 0.83 for sub- and supra-therapeutic levels, respectively.
• Key predictors included kinetic estimated glomerular filtration rate, vancomycin regimen, comorbidities, BMI, age, sex, and blood pressure.

Conclusions:

• Developed predictive models to identify risks of inappropriate vancomycin trough levels.
• The models aid in optimizing vancomycin dosing strategies for ICU patients.
• Enhanced accuracy in drug dosing can improve patient outcomes in critical care settings.