A model-based approach to assess unstable creatinine clearance in critically ill patients.

Creatinine clearance is an important tool to describe the renal elimination of drugs in pharmacokinetic evaluations and clinical practice. In critically ill patients, unstable kidney function invalidates the steady state assumption underlying equations such as Cockcroft-Gault. While measured creatinine clearance (mCrCL) is often used in non-steady state situations, it assumes that observed data is error-free, neglecting frequently occurring errors in urine collection. In contrast, compartmental non-linear mixed effects models of creatinine allow to describe dynamic changes in kidney function while explicitly accounting for a residual error associated with observations. Based on 530 serum and 373 urine creatinine observations from 138 critically ill patients, a one compartment creatinine model with zero-order creatinine generation rate (CGR) and first-order creatinine clearance (CrCL) was evaluated. An autoregressive approach for inter-occasion variability provided a distinct model improvement compared to a classical approach (ΔAIC -49.0). Fat-free mass (FFM), plasma urea concentration, age and liver transplantation were significantly related to CrCL, while weight and sex were linked to CGR. The model-based CrCL estimates were superior to standard approaches to estimate CrCL (or glomerular filtration rate) including Cockcroft-Gault, mCrCL, four-variable modification of diet in renal disease (MDRD), six-variable MDRD and chronic kidney disease epidemiology collaboration (CKD-EPI) as a covariate to describe cefepime and meropenem pharmacokinetics in terms of objective function value. In conclusion, a dynamic model of creatinine kinetics provides the means to estimate actual CrCL despite dynamic changes in kidney function, and it can easily be incorporated into population pharmacokinetic evaluations. This article is protected by copyright. All rights reserved.