Individualizing vancomycin dosage regimens: one- versus two-compartment Bayesian models.

The absolute and relative predictive performances of one- and two-compartment Bayesian forecasting models were evaluated and compared. Initial population parameters were derived from 25 adult patients with stable renal function and who were being treated for presumed or documented gram-positive infections. The performance of each model was compared using these population parameters with and without steady-state or non-steady-state feedback concentrations to predict future peak and trough concentrations in an additional 20 patients. Both models tended to underpredict vancomycin peak and trough concentrations obtained at steady state. The use of a two-compartment model resulted in statistically less bias and more precise predictions of vancomycin peak concentrations when either population parameters or non-steady-state concentrations were used for future predictions. No difference in model performance was observed when steady-state concentrations were used to predict future steady-state concentrations. The results of this evaluation demonstrate that the two-compartment Bayesian model is less biased and more precise in determining future vancomycin serum concentrations given only population parameters or non-steady-state feedback information. No difference in model performance could be discerned when steady-state concentrations were used as feedback information.