Population pharmacokinetic modeling and optimal sampling strategy for Bayesian estimation of amikacin in critically ill septic patients
Isabelle K. Delattre (1), Flora T. Musuamba (1,2), Joakim Nyberg (3), Roger K. Verbeeck (2), Frédérique Jacobs (4), Pierre E. Wallemacq (1)
(1) Louvain centre for Toxicology and Applied Pharmacology, (2) Unit of Pharmacokinetics and Metabolism, Université catholique de Louvain, Brussels, Belgium; (3) Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden; (4) Department of Infectious Diseases, Hôpital Erasme, Brussels, Belgium
Objectives: In severe sepsis and septic shock, appropriate antibiotic therapy plays a key role in the patient management. Since the sepsis-induced pharmacokinetic (PK) modifications need to be considered in the drug dosages [1], the present study aimed to develop a population PK model for amikacin (AMK) in critically ill septic patients, and to subsequently propose an optimal sampling strategy suitable for Bayesian estimation of the drug, taking into account clinical constraints.
Methods: Serum concentration-time profiles were obtained from 88 critically ill septic patients, during the first 24 hours of antibiotic treatment (AMK combined with a broad-spectrum β-lactam). The population PK model for AMK was developed using NONMEM (FOCEI method). Fourteen potential covariates, including demographic data, pathophysiological characteristics and co-medication, were evaluated for influence on PK parameters. Using population estimates as prior information, optimal sampling times were selected based on ED-optimality. Optimization was performed in PopED v.2.10. [2,3] on the individual level, as previously reported [4]. Taking into account clinical constraints, a two-point sampling strategy was investigated. Predictive performance of Bayesian estimates obtained with the optimal sampling strategy was assessed.
Results: A two-compartment model with first-order elimination best fitted the AMK concentrations. Population PK estimates were 19.2 and 9.34 L for the central and peripheral volume of distribution, 4.31 and 2.21 L/h for the inter-compartmental and total body clearance. Creatinine clearance (CrCL), calculated using the Cockcroft-Gault equation, was retained in the final model. Optimal sampling times were 2 replicated sampling times at 6 hours when optimizing samples between 1-6 hours. Predictive performance of Bayesian estimates obtained with this sampling strategy was satisfactory (MPE < 6%, RMSE < 30%).
Conclusions: The present study has highlighted the significant influence of the CrCL on the PK disposition of AMK, during the first hours of treatment in critically ill septic patients. Based on developed population estimates of the drug, an optimal sampling strategy has been proposed for this patient population. As it was found suitable for the Bayesian estimation, its prospective use could be considered as successful.
References:
[1] Roberts JA, Lipman J. Antibacterial dosing in intensive care. Pharmacokinetics, degree of disease and pharmacodynamics of sepsis. Clin Pharmacokinet 2006; 45: 755-73.
[2] Foracchia M, Hooker A, Vicini P, et al. POPED, a software for optimal experiment design in population kinetics. Comput Methods Programs Biomed 2004; 74: 29-46.
[3] PopED, version 2.10 (2009). http://poped.sourceforge.net/
[4] Hennig S, Nyberg J, Fanta S, Hooker A, Karlsson MO. Application of the optimal design approach to improve therapeutic drug monitoring for cyclosporine. PAGE 17 (2008) Abstr 1436.