Evaluation of Bayesian estimation to discriminate subpopulations of patients with altered pharmacokinetics using fragmentary data: a pilot study with pefloxacin.

Bayesian estimation of individual pefloxacin pharmacokinetic parameters was evaluated in patients with renal (n = 34) or hepatic (n = 16) impairments who participated in experimental pharmacokinetic studies during pefloxacin development. The trials involved singly intravenous dosing (300 to 850 mg as 0.5 to 1h-infusion) and serial pefloxacin plasma level measurements (n = 10-18 over 72-96 h). Bayesian estimation used only 2 measured plasma levels (end of infusion and either 12 hour (BE 12h) or 48 hour (BE 48h) levels) and a priori information obtained in a reference population of healthy volunteers (n = 19). The performance of Bayesian estimation in predicting total plasma clearance (Cl), elimination half-life (t1/2) and steady state volume of distribution (Vss) was evaluated with respect to weighted nonlinear least-square estimates using all the data points and a two-compartment model. Pharmacokinetic parameters (and their inter-individual variability) in patients differed significantly from those of healthy volunteers (e.g. t1/2 of 22 +/- 14 h, range 6-81 h vs. 11 +/- 2 h, range 8-17, p less than 0.001). BE 12h provided biased estimates of Cl and t1/2 (24 and 28% respectively) and poor precision (30 and 53% respectively), Vss being well estimated. The BE 48h estimator had good performance with unsignificant bias (less than 1%) and good precision (extremes 13 and 27% for Cl and t1/2 respectively). Patients with hepatic impairment had the more pronounced pharmacokinetics modifications and could even be discriminated from healthy volunteers using the biased BE 12h estimates. These results suggest that Bayesian estimation is quite robust to the a priori information.(ABSTRACT TRUNCATED AT 250 WORDS)