Comparison of Sawchuk-Zaske and Bayesian forecasting for aminoglycosides in seriously ill patients

1. Individual pharmacokinetic parameters and predicted steady-state serum concentrations of aminoglycosides were calculated by Sawchuck-Zaske (SZ) and Bayesian methods. 2. Predicted concentrations were compared with observed steady-state concentrations for 36 seriously ill patients with systemic infections. Four aminoglycoside concentrations were used for the SZ method. Differing numbers of serum aminoglycoside samples were used in the Bayesian parameter estimation: one sample Bayesian used one post-infusion concentration, two sample Bayesian used a trough plus one post-infusion concentrations and four sample Bayesian used a trough plus three post-infusion concentrations. 3. 79% of the SZ predictions were with +/- 2 mg l-1 of the observed peak concentrations, and 72% of the two sample Bayesian predictions were within the same range. 82% of SZ and the two sample Bayesian predictions were within +/- 1 mg l-1 of the observed trough concentrations. 4. A confidence interval comparison of estimated pharmacokinetic parameters and precision for the predicted concentrations showed no important differences between the SZ and the two sample Bayesian. The four sample Bayesian was the most precise method. 5. We conclude that the Bayesian forecasting method utilizing a trough plus one post-infusion concentrations is as useful as the SZ method which requires three to four serum concentrations in individualizing aminoglycoside therapy for seriously ill patients.