Optimization of dosing strategy for vancomycin using a population pharmacokinetic model combined with machine learning approaches
Minji Kwon (1), Won Gun Kwack (2), Sooyoung Lee (1), Kwang-Youl Kim (3), Bo-Hyung Kim (1,4,5)
(1) Department of Clinical Pharmacology and Therapeutics, Kyung Hee University Hospital, Seoul, Republic of Korea, (2) Division of Pulmonary, Allergy, and Critical Care Medicine, Department of Internal Medicine, Kyung Hee University, Seoul, Republic of Korea, (3) Department of Clinical Pharmacology, Inha University Hospital, Inha University School of Medicine, Incheon, Republic of Korea, (4) East-West Medical Research Institute, Kyung Hee University, Seoul, Republic of Korea, (5) Department of Biomedical Science and Technology, Graduate School, Kyung Hee University, Seoul, Republic of Korea
Objectives: The optimized dose and frequency of dosing for vancomycin are critical to achieve target AUC and trough concentrations at steady state. This is because these target AUC or trough concentrations determine therapeutic efficacy or safety. These targets are determined by pharmacokinetic (PK) parameters; volume of distribution or clearance can be explained by individual kidney function, comorbidity, age, body weight, gender, and so on. Previous studies have mainly focused on PK parameters for the therapeutic drug monitoring (TDM). Such an approach inevitably had limitations in determining the dosage regimen, considering various changes in the clinical environment. This study was planned to overcome these limitations by applying the developed PK model and machine learning (ML) method together.
Methods: Vancomycin PK model was developed using NONMEM. We obtained a reliable population PK model and PK parameters by combining PK parameters obtained from a clinical study on real patients with previous results in the literature. The current clinical study was a PK study conducted in patients admitted to an intensive care unit and receiving vancomycin. PK blood sampling was performed at least 9 times for each patient to determine the concentration of vancomycin and estimate PK parameters. In addition, demographic characteristics (gender, age, body weight, height, body mass index), clinical information related to renal function (Cockcroft-Gault equation, modification of diet in renal disease (MDRD) equation) and other clinical information (liver function test results, concomitant medication, etc.) were collected for the estimation of PK parameters. An approach combining results from literature studies with those from the current study was also required to test or determine important covariates from various clinical information. Based on the determined PK models, various ML approaches were used; Decision Tree (DT), Random Forest (RF), and XGBoost were applied using each R package.
Results: The PK parameters for the population model were estimated using the 1- or 2- compartment model. The ML approaches presented the risk of vancomycin-induced nephrotoxicity as a probability. The major covariates are as follows; demographic characteristics (age, body weight, etc.), creatinine clearance using Cockcroft-Gault equation. These covariates could explain the changes of PK parameters such as volume of distribution or clearance. Certain variables influencing the risk of renal toxicity were specifically subdivided to account for the model.
Conclusions: This study developed a population PK model of vancomycin, which might be similar to previous findings. Additionally, the ML model had multiple factors explaining vancomycin-induced nephrotoxicity or vancomycin clearance. This approach can be further improved with more information, such as on additional covariates and the improvement of the ML model. The results of this study are expected to contribute to improving the dosing strategy of vancomycin or increasing the efficiency of TDM support tools for vancomycin or other antibiotics.
References:
[1] Lee S., et al. Application of Machine Learning Classification to Improve the Performance of Vancomycin Therapeutic Drug Monitoring. Pharmaceutics. 2022;9;14(5):1023. doi: 10.3390/pharmaceutics14051023.
[2] Huang X., et al. Prediction of vancomycin dose on high-dimensional data using machine learning techniques. Expert Rev Clin Pharmacol. 2021;14(6):761-771. doi: 10.1080/17512433.2021.1911642.
[3] Imai S., et al. Using Japanese big data to investigate novel factors and their high-risk combinations that affect vancomycin-induced nephrotoxicity. Br J Clin Pharmacol. 2022;88(7):3241-3255. doi: 10.1111/bcp.15252.
[4] Aljutayli A., et al. An Update on Population Pharmacokinetic Analyses of Vancomycin, Part I: In Adults. Clin Pharmacokinet. 2020;59(6):671-698. doi: 10.1007/s40262-020-00866-2.