Jaegu Kang 1, Ki Young Huh 1
1 Seoul National University College of Medicine and Hospital (Seoul, Republic of Korea)
Introduction/Objectives: Ventricular assist devices (VADs) are implantable mechanical pumps used in advanced heart failure as bridge-to-transplantation or destination therapy. Despite widespread use of vancomycin for perioperative infection control after VAD implantation surgery, model-informed therapeutic drug monitoring (TDM) remains challenging due to altered pharmacokinetics and marked physiological variability after surgery. VAD-implanted patients exhibit elevated vancomycin volume of distribution (1.25-1.5-fold typical values). However, unlike ECMO, the VAD circuit volume is insufficient to explain this expansion through circuit extension or drug adsorption alone. Despite the profound hemodynamic alterations caused by continuous flow VAD support, their impact on vancomycin pharmacokinetics remains poorly characterized. This study explored whether hemodynamic parameters could serve as covariates in the vancomycin PK model for VAD-implanted patients.
Methods: Patients undergoing VAD implantation at Seoul National University Hospital (SNUH) between 2015 and 2024 were included. Pseudonymized clinical data, including vancomycin dosing records and serum concentrations were extracted from the clinical data warehouse. Population PK models for adult and pediatric patients were built separately using NONMEM version 7.6. One- and two-compartment models with a first-order elimination were explored as base models. Interindividual variability was modeled exponentially, and residual error was assessed using proportional, additive, and combined models. In covariate selection step, time-varying covariates including body weight, renal function biomarkers (serum creatinine and cystatin C, continuous renal replacement therapy (CRRT)), and hemodynamic parameters (systolic/diastolic blood pressure, mean arterial pressure and pulse pressure) were analyzed for each PK parameter. Covariate selection employed stepwise forward inclusion (p<0.05) and backward elimination (p<0.01). Model evaluation included standard goodness-of-fit plots, prediction-corrected visual predictive checks, and bootstrap analysis for parameter precision assessment. Results: A total of 260 vancomycin serum concentrations (adult: 87) from 44 VAD-implanted patients (adult: 23) were included in the analysis. The final models were one-compartment models with combined residual error for both adult and pediatric patients. In the adult model, clearance was influenced by serum creatinine (exponential model), while body weight (linear model) was a significant covariate for central volume. In the pediatric model, clearance was influenced by serum creatinine and body weight (both exponential), and central volume was modeled as a linear function of body weight. Typical values of central volume were 1.75 L/kg for adults, and 1.94 L/kg for pediatric patients. However, no hemodynamic parameters were included as covariates for clearance or central volume in either the adult or pediatric model. Conclusions: Vancomycin pharmacokinetics in VAD-implanted patients were best described by a one-compartment model incorporating only body weight and serum creatinine as covariates. VAD-implanted patients showed larger vancomycin central volume than those of typical patients without VAD. However, time-varying hemodynamic parameters were insufficient to physiologically describe the expanded central volume. Further model exploration using a larger dataset including peak concentrations or additional covariates such as VAD parameters or edema markers is warranted.
Reference: PAGE 34 (2026) Abstr 11856 [www.page-meeting.org/?abstract=11856]
Poster: Drug/Disease Modelling - Infection