Predictive Performance of Bayesian Method in a Simulation Work: One-Compartment Open Linear Model Containing an Oral Route

The Bayesian weighted least-squares method (BWLS) was evaluated by simulation work. One hundred simulated patients were created by random number generator through a logarithm transformation. Theophylline was examined as a typical drug which obeys a one-compartment open linear model: total body clearance (CL), 0.04 ±0.02 l/hr/kg; volume of distribution (V_d), 0.50 ±0.09 l/kg; rate constant of absorption (k_a), 0.96 ±0.62 hr^-1. A non-uniform but periodic dosage regimen containing an oral route, designed to give a range of 10 to 20 μg/ml of serum drug concentration (SDC) in the steady-state period, was assumed. The BWLS with one SDC measurement provided time-dependent precision and bias in both CL and V_d estimates ; by increasing the time between the sampling and the onset of therapy, the precision in CL esimates increased but the precision in Vd decreased, and BWLS etimated the CL and V_d with a prediction error of approximately 10% and 20% respectively, after 14 hr from the onset of therapy (mean half-life was 10.5 hr). The k_a could be estimated inaccurately (>110% RMSE). In the predictions of steady-state SDCs, the precision and bias also showed the time-dependent profile as a function of the sampling times and, moreover, the prediction times: 1.6 to 8.6 μg/ml as a root mean squared error (RMSE), –2.9 to 1.1 μg/ml as a mean prediction error (ME). The utilization of two SDC measurements partly improved the predictive performance of BWLS for estimating CL and V_d. From the simulation work it was concluded that the BWLS would provide reliable estimations of parameters such as CL and V_d, and clinically acceptable predictions of the subsequent SDCs, given appropriate population parameter values, means and variances, and sampling times, even when as few as one or two SDC measurements are available.