Comparison of a Bayesian Program with Three Microcomputer Programs for Predicting Gentamicin Concentrations

A recently developed Bayesian regression program was compared with three other aminoglycoside pharmacokinetic dosing programs available for clinical use. From 30 adult patients. 152 measured serum gentamicin concentrations (SGC) were evaluated retrospectively (78 peak and 74 trough). Predictive performance was compared for each method by using the first peak and trough SGC pair to predict subsequent serum concentrations, making a total of 92 predictions (48 peak and 44 trough). The two Bayesian programs (Brater and Koup) were further evaluated using only one initial peak or trough SGC to make the same predictions. Mean predicted error (ME), mean absolute error (MAE), and root mean squared error (RMSE) were calculated for each method. Prediction bias and precision were compared statistically, between each method, by calculating the 95% confidence intervals for the ΔME and ΔME, respectively. No statistically significant differences were found in the MAEs among any of the methods for predicting peak SGCs. with the exception of the Brater program, using a single trough SGC. which was statistically less precise (<0.05). There were few statistically significant differences in the MAEs for trough SGCs; however. Koup's Bayesian program using a single trough concentration yielded statistically more precise predictions than the other methods. The ME was found to differ significantly (p < 0.05) among estimates for peak and trough SGCs provided by some of the predictive methods. Errors determined for predicted versus observed peak SGC for all methods revealed the following ranges (mUg/mL): −0.31 to 1.31 ME; 1.01 to 1.62, MAE; and 1.26 to 2.15. RMSE. Errors determined for predicted trough concentrations for all methods revealed ranges of −0.22 to 0.08. ME; 0.39 to 0.66. MAE: and 0.53 to 1.02. RMSE. Although predictive performance was found to be similar for all methods, both Bayesian programs offer the potential advantage of accurate predictions based on one peak, in some cases, one trough SGC.