Pharmacokinetic Data Mining for Managing Clinical Use of Vancomycin

Drug-related problems have become a growing concern in clinical medicine. Particularly prevalent are problems surrounding suband over-therapeutic doses of high-alert medications. In this study, we use a model-tree based regression (namely M5) and an SVM technique to build systems that predict the regimen adequacy of vancomycin, a glycopeptide antimicrobial antibiotic effective for Gram-positive bacterial infections that has a narrow therapeutic index and considerable toxicity and adverse effects. We evaluate each system’s predictions of the patient’s peak and trough concentrations resulting from a vancomycin regimen, using a total of 1,099 clinical cases collected from a major tertiary medical center in southern Taiwan. We also examine the use of Bagging for enhancing the prediction accuracy of the respective systems and include in our evaluation a prevailing one-compartment model for performance benchmark purposes. Overall, our analysis results show that both M5 and SVM are significantly more accurate in predicting peak and/or trough concentrations than the one-compartment model, which approximates the current practice at the studied medical center. In addition, M5 appears to benefit considerably from Bagging, which also has a positive effect on SVM but seemingly to a lesser extent. Taken together, our findings suggest promising utilities of supervised learning techniques for improving the clinical use of vancomycin and similar high-alerting drugs, thus complementing clinicians’ current practices.