Exploring population pharmacokinetic models in patients treated with vancomycin during continuous venovenous hemodiafiltration on different anticoagulant modalities

Abstract

Uncertainty remains concerning pharmacokinetics (PK) of antimicrobials in critically ill patients due to the scarcity of data and the variable physiology of the patient cohort. This project incorporates two distinct aims. The first aim was to describe the population PK estimates and the influence of patient covariates on PK of vancomycin following intermittent infusion in critically ill patients receiving continuous venovenous hemodiafiltration (CVVHDF), and on different anticoagulant modalities. Whereas, the second aim was to explore the potential for automatic data extraction from the electronic health record (EHR) for population PK analysis. Vancomycin dose and concentration data (peak, trough) were collected retrospectively from 31 critically ill patients (n= 280 levels). A one-compartment model was used to describe the vancomycin concentration-time profiles, using Pmetrics software. A base structural model was developed and covariates assessed for inclusion in PK models were as follows: Dialysis status was categorised as 1) dialysis with regional citrate anticoagulation (RCA) (60 levels) 2) dialysis not using RCA (non-citrate) (120 levels) or 3) no dialysis (44 levels). Continuous covariates were: cumulative fluid balance (FB), effluent flow rate (EFR), blood flow rate, body weight, albumin concentration (Alb) and age. Considering the high inherent variability in this patient cohort, the structural base models were reasonable using both clearance (CL) with volume of distribution (V) and elimination rate constant (Ke) with V, even without addition of covariates, with observed vs. predicted (o-p) R2 values of 0.51 and 0.81 for population and posterior individual predicted models respectively. The mean ? SD of total vancomycin PK estimates were: CL: 2.16 ? 0.57 L/h, V: 80.65 ? 22.65 L, and Ke: 0.03 ? 0.01 h. The best population predicted models (o-p R2= 0.59) included FB as a continuous covariate applied to the V parameter and Alb applied to the Kei parameter (using the inverse approach). Interestingly, there was a suggested improvement in the population model of the RCA modality group over the non-dialysis cohort, with o-p R2 of 0.86 and 0.28, respectively (o-p R2 non-citrate dialysis= 0.58). The less variable PK estimates in the RCA group model suggests that initial vancomycin dosing in this type of patient might be more straightforward than in the other anticoagulant categories, due to the homogeneity of patients in the citrate group. No single covariate model using all patients provided a notable improvement simultaneously in both population predictions and posterior individual predictions. Inclusion of an intercept in the PK parameter-covariate relationship was necessary to detect significant covariate relationships with PK parameters. Furthermore, investigated automated data extraction method was potentially valuable for use in PK analysis, after further development to achieve a reliable automated extraction system. Ultimately, FB, Alb data and RCA status are suggested as covariates for further analysis with richer data, to optimise covariate PK modelling of vancomycin in these patients, and support dose optimisation.