Application of data mining to predict the dosage of vancomycin as an outcome variable in a teaching hospital population

Data mining is a process used to extract potentially valuable information hidden in large volumes of raw data. The aim of this study was to explore the possibility of using easy to implement and effective supervised learning techniques to predict the dosage of vancomycin.To reach this goal, we considered the prediction of the dosage of vancomycin as a classification problem. We chose the C4.5 decision tree technique for the dosage prediction process and supplied it with a boosting technique to enhance its performance.The potential predictor variables were collected from 833 patients with methicillin-resistant Staphylococcus aureus, or penicillin intolerance who were being treated with vancomycin and undergoing therapeutic drug monitoring (TDM) after attainment of steady state blood concentrations. Attributes tested as potential predictors included age, sex, weight, serum creatinine concentration, dosing interval, and variables from 1-compartment model kinetics such as Kd, Vd, and t(1/2).The results showed that the proposed method can utilize a variety of parameters to predict the dosage of vancomycin in the population used and that it performs well over a range of patient ages and renal function. The method may offer an alternative to existing methods used to support decision-making in clinical practice.