2002 Volume 50 Issue 11 Pages 826-829
We evaluated the accuracy of Bayesian dosing for vancomycin (VCM). Prospectively collected serum concentration data was evaluated retrospectively in 12 patients who had 2 steady-state peak and trough serum VCM concentrations obtained in routine therapeutic drug monitoring from April 1995 to July 2000. The predictive performance of the method using a 1-point trough level was compared to that using 2-point trough and peak levels. In predicting subsequent peak and trough VCM serum concentrations, the predictive accuracy of the Bayesian method using only the trough level did not differ significantly from that using both trough and peak levels. At the trough level (n=12), mean prediction error (ME) was-4.08μg/mL, mean absolute prediction error (MAE) was 4.44μg/mL, and root mean squared prediction error (RMSE) was 5.42μg/mL. At the peak level (n=11), ME was 2.87μg/mL, MAE was 7.04μg/mL, and RMSE was 8.89μg/mL for the 1-point method. The corresponding ME for the 2-point method was -3.30μg/mL, MAE was 3.90μg/mL, and RMSE was 4.93μg/mL at the trough level (n=12), and at the peak level (n=10), 0.57μg/mL for ME, 5.03μg/mL for MAE, and 6.74μg/mL for RMSE. The difference in accuracy was less than 1μg/mL at the trough level, and less than 3μg/mL at the peak level. These results indicate that therapeutic drug monitoring of only trough levels after initial VCM dosing determined by Moellering nomography would satisfactorily achieve desired serum VCM concentrations by using a Bayesian method for individualizing VCM dosage.
