Bayesian prediction of serum phenytoin concentration in a simulation study.

The predictive performance of the Bayesian weighted least-squares method (BWLS), viz., the Bayesian method, was evaluated and compared with that of the ordinary weighted nonlinear least-squares method (OWLS) in a simulation study. Phenytoin was selected as a model drug for a one-compartment nonlinear model (Michaelis-Menten model). The patient population for the phenytoin study (Km : 3.9±2.1 μg/ml, Vmax : 6.78±1.27 mg/kg/d) was generated by using a random number generator.When estimating the parameters by the Bayesian method, the Vmax estimate was more precise than the Km : Vmax could be estimated with a 15% and 10% prediction error with one and two observations, respectively, as opposed to 30% and 20% for Km. It was also inferred that the predictive performance of the one- and two-observation BWLSs was a slightly better than or equal to that of the three-observation OWLS but did not approach that of the four-observation OWLS.The present simulation study indicated that the Bayesian method could provide reliable estimations of Michaelis-Menten parameters such as Km and Vmax and clinically acceptable predictions for the steady-state serum concentration with at least one observation (the root mean squared error, 3.4 μg/ml; 95% confidence interval, 1.4-4.6 μg/ml), if appropriate population pharmacokinetic parameters (i.e., means and variances) were available.