292: PREDICTION OF VANCOMYCIN TROUGH CONCENTRATION VIA MACHINE LEARNING IN THE INTENSIVE CARE UNIT

Abstract

Introduction: It is difficult to predict vancomycin trough concentrations in critically ill patients as their pharmacokinetics change with the progression of both organ failure and medical intervention. This study seeks to develop a model to predict initial vancomycin trough concentration using machine learning (ML) and compare the resulting prediction’s accuracy to the population pharmacokinetic (PPK) model. Methods: A single-center retrospective observational study was conducted. Patients who had been admitted to the intensive care unit, received intravenous vancomycin, and had undergone therapeutic drug monitoring between November 2017, and December 2020, were included. Thereafter, ML models were developed with ridge regression, random forest, and LightGBM using 42 features. The deviation between the model’s predicted and measured values with the smallest mean absolute errors was compared with the deviation between the predicted and PPK models. Results: In total, 226 patients were included, and a significant difference was found between the predicted and measured values by the ML and PPK models (ML 1.59 vs PPK 4.19, difference 2.19, 95% confidence interval [1.68, 2.72]). In a subgroup analysis, patients with severe acidosis, patients who receive continuous renal replacement therapy, as well as those who are elderly and obese, showed a higher prediction accuracy using ML. The important features were determined to be renal function (creatinine clearance and serum creatinine), vancomycin dose (dose per hour and total dose), disease severity, and age. Conclusions: This study concludes that ML may be able to predict vancomycin trough concentrations more accurately than the currently used PPK model.