An application of classification analysis for skewed class distribution in therapeutic drug monitoring - the case of vancomycin

Abstract

Vancomycin can induce potent adverse side effects if drug concentration is not controlled within a narrow safety range. Therefore, therapeutic drug monitoring (TDM) is followed to adjust dose and help monitor treatment effects. Because TDM are not helpful in patients taking vancomycin for the first time, it's usage has a limitation to ensure medication safety. This study aimed at using decision tree induction to predict outcomes of vancomycin. Research results demonstrate that the asymmetric distribution among classes in the TDM data would result in prediction deviation. An ideal model with good prediction efficacy could be established by adjusting the ratio among outcome classes through "over-sampling for expanding minority data". The prediction model would be helpful in controlling the positive and negative effects of vancomycin treatment, improving care at the patient level and improving costs at the social level. Some interesting decision rules derived from the decision tree were analyzed its clinical meanings. Precious prescription knowledge is thus extracted and accumulated.