Prediction of gentamicin concentrations in neonates and infants using a Bayesian pharmacokinetic model.

This study retrospectively characterized population-based pharmacokinetic parameters for gentamicin in neonates and young infants, and evaluated the predictive performance of these parameters in a Bayesian forecasting program. Population parameter estimates were determined from the serum concentration-time data of 19 neonates and infants using a one-compartment open infusion model and nonlinear least-squares regression analysis. Univariate and multiple stepwise linear regression analyses were used to determine significant relationships between demographic characteristics and gentamicin pharmacokinetic parameters. Creatinine clearance and postnatal age were the most significant predictors of weight-standardized gentamicin clearance (model r2 = 0.86). The relationships between patient characteristics and population-based parameters were incorporated into the one-compartment Bayesian forecasting model. A second group of 17 neonates and infants receiving 35 courses of gentamicin therapy were used to evaluate the predictive performance of the population-based parameters and a Bayesian forecasting model. The population parameters provided accurate prediction of steady state gentamicin concentrations throughout multiple courses of therapy within the same patient. Bayesian forecasting further minimized the mean prediction error (bias) once a set of steady state peak and trough serum gentamicin concentrations became available (peak concentrations: -0.062 vs. -0.273 mg/l; trough concentrations: -0.006 vs. -0.161 mg/l). The mean absolute error (accuracy) was similar for the two sets of parameters. The observed accuracy of both the population parameters and Bayesian forecasting suggests that monitoring of serum gentamicin concentrations can be kept to minimum in neonates and infants.