Optimizing Initial Vancomycin Dosing in Hospitalized Patients Using Machine Learning Approach for Enhanced...

Area of Science:

• Pharmacokinetics and Pharmacodynamics
• Machine Learning in Medicine
• Antibiotic Dosing Optimization

Background:

• Standard vancomycin dosing often leads to suboptimal outcomes due to inter-patient variability.
• Therapeutic Drug Monitoring (TDM) adjustments are reactive, highlighting the need for predictive, personalized initial dosing.

Purpose of the Study:

• To develop and evaluate a machine learning (ML)-based algorithm for predicting optimal initial vancomycin doses.
• To ensure initial vancomycin doses fall within the therapeutic range, defined by the 24-hour area under the curve to minimum inhibitory concentration.

Main Methods:

• A retrospective cohort of 415 patients receiving intravenous vancomycin was used for training and testing (7:3 ratio).
• Four ML models (gradient boosting, random forest, SVM, XGB) were evaluated to develop the OPTIVAN algorithm.
• The OPTIVAN algorithm was validated using an external cohort (n=268) and integrated into a web-based clinical support tool.

Main Results:

• The Support Vector Machine (SVM) model, incorporated into OPTIVAN, showed the best predictive performance with an AUROC of 0.832 (training) and 0.720 (validation).
• Key covariates for the SVM model included age, BMI, glucose, BUN, eGFR, hematocrit, and dose/weight.
• Consistent performance was observed across various patient subgroups, including different renal functions, sexes, and BMIs.

Conclusions:

• The OPTIVAN algorithm offers a significant advancement in personalized initial vancomycin dosing, overcoming limitations of current TDM.
• This ML-driven approach can reduce subsequent dosage adjustments and improve vancomycin treatment efficacy.
• A user-friendly, web-based application of the OPTIVAN algorithm is available for practical clinical use.