[Can we use a Bayesian method to build a pharmacokinetic population in two steps?]

The aim of this study is to evaluate the use of a pharmacokinetic population model built by the two-stage method and individual parameters determined by a Bayesian estimation instead of nonlinear regression. We performed a retrospective analysis on 32 patient files (mean age: 82 years). First, we analysed prediction of amikacin serum levels for the Bayesian method (MAP) and nonlinear regression (MLS). Second, we compared pharmacokinetic parameter values for each patient with MAP and MLS methods for a one- or two-compartment model. For the one-compartment model, no difference in prediction performance was found (correlation coefficient: rMLS = 0.911, rMAP = 0.903, p > 0.05; precision: pMLS = 134.3, pMAP = 147, p > 0.05). A significant difference was observed only for systematic error (eMLS = -4.47, eMAP = -3.34, p < 0.05). For a two-compartment model, the Bayesian method was better for long-term prediction: 4-8 days (rMLS = 0.877, rMAP = 0.886, p > 0.05; eMLS = 5.26, eMAP = 0.04, p < 0.01; pMLS = 441.7, pMAP = 149, p < 0.05). The comparison of MAP and MLS estimated pharmacokinetic parameter values for a one-compartment model showed that the Bayesian method used to built a pharmacokinetic population in two stages does not influence pharmacokinetic parameter estimation (p > 0.05 for Vd, Kslope, Kel and t1/2). We conclude that we can use a Bayesian method to build a pharmacokinetic population in two steps in order to perform adaptative control of a drug-dosage regimen.