A model-based analysis of the predictive performance of different renal function markers for cefepime clearance in the ICU

Abstract

Objectives: Several population pharmacokinetic models for cefepime in critically ill patients have been described, which all indicate that variability in renal clearance is the main determinant of the observed variability in exposure. The main objective of this study was to determine which renal marker best predicts cefepime clearance. Methods: A pharmacokinetic model was developed using NONMEM based on 208 plasma and 51 urine samples from 20 ICU patients during a median follow-up of 3 days. Four serum-based kidney markers (creatinine, cystatin C, urea and uromodulin) and two urinary markers [measured creatinine clearance (CLCR) and kidney injury molecule-1] were evaluated as covariates in the model. Results: A two-compartment model incorporating a renal and non-renal clearance component along with an additional term describing haemodialysis clearance provided an adequate description of the data. The Cockcroft-Gault formula was the best predictor for renal cefepime clearance. Compared with the base model without covariates, the objective function value decreased from 1971.7 to 1948.1, the median absolute prediction error from 42.4% to 29.9% and the between-subject variability in renal cefepime clearance from 135% to 50%. Other creatinine- and cystatin C-based formulae and measured CLCR performed similarly. Monte Carlo simulations using the Sanford guide dose recommendations indicated an insufficient dose reduction in patients with a decreased kidney function, leading to potentially toxic levels. Conclusions: The Cockcroft-Gault formula was the best predictor for cefepime clearance in critically ill patients, although other creatinine- and cystatin C-based formulae and measured CLCR performed similarly.