Comparison of Four Renal Function Estimator-Based Models for the Prediction of Gentamicin Concentrations in Geriatric Patients by Use of Nonparametric Population Approach
N. Charhon (1), L. Bourguignon (1,2), P. Maire (1,2), R.W. Jelliffe (4), M. Neely (4), S. Goutelle (1,2,3)
(1) Hospices Civils de Lyon, Groupement Hospitalier de Gériatrie, Service pharmaceutique – ADCAPT, Francheville, France ; (2) Université de Lyon, F-69000, Lyon ; Université Lyon 1 ; CNRS, UMR5558, Laboratoire de Biométrie et Biologie Evolutive, F-69622, Villeurbanne, France ; (3) ISPB - Faculté de Pharmacie de Lyon, Université Lyon 1, Lyon, France ; (4) Laboratory of Applied Pharmacokinetics, USC School of Medicine, Los Angeles, CA, USA
Objectives: Most aminoglycoside pharmacokinetic models include an index of renal function, such as creatinine clearance to describe elimination . However, the best clinical descriptor of renal function for PK modeling of aminoglycosides has not been established. The objective of this study was to compare four gentamicin (GENT) PK models based on the Cockcroft-Gault (CG), Jelliffe (JEL), MDRD, and modified MDRD (MDRDm, adjusted to individual body surface area) formulae.
Methods: This analysis was based on 427 gentamicin concentrations from 92 geriatric patients who received intravenous GENT for various infectious diseases. Monitoring of gentamicin concentrations was part of routine patient care. Four bicompartmental models were fitted to GENT concentrations in a learning set of 64 patients using the NPAG algorithm . Each model included an index of renal function (CG, JEL, MDRD, or MDRDm) as a covariate influencing GENT serum clearance. The Akaike information criterion (AIC) was used to assess the goodness-of-fit of candidate models. Mean prediction error and mean squared prediction error were used to evaluate bias and precision, respectively. In a validation set of 28 patients, population and individual predictions were made from each of the four model nonparametric population PK parameter joint densities. Bias and precision of the four models were compared with the Kruskal-Wallis test in both the learning and validation sets.
Results: In the learning set, the CG-based model best fitted the data (lowest AIC value), followed by JEL, MDRD, and MDRDm-based models. Bias and precision of population predictions were significantly different (p < 0.001 and p = 0.027, respectively). In the validation set, bias and precision of population predictions were not significantly different between the models. However, individual predictions from the four models showed marginally different bias (p = 0.04). Overall, the CG-based model provided the best fit and predictive performance.
Conclusions: PK models of GENT based on various estimators of renal function may provide significantly different results. In this study, the model based on the CG equation predicted GENT concentrations slightly better than the JEL, MDRD, and MDRDm equations in geriatric patients. In clinical practice, one should continue to use the CG equation for model-based adaptive control of GENT dosage regimens.
 Jelliffe RW, Iglesias T, Hurst AK, Foo KA, Rodriguez J. Individualising gentamicin dosage regimens. A comparative review of selected models, data fitting methods and monitoring strategies. Clin Pharmacokinet 1991;21(6):461-78
 Leary R, Jelliffe R, Schumitzky A, Van Guilder M. An adaptive grid non-parametric approach to pharmacokinetic and dynamic (PK/PD) population models," cbms, pp.0389, 14th IEEE Symposium on Computer-Based Medical Systems (CMBS'01), 2001