Bayesian forecasting of gentamicin pharmacokinetics in pediatric intensive care unit patients

The predictive performance of a one compartment Bayesian forecasting program was evaluated in pediatric intensive care unit patients with normal renal function. Gentamicin pharmacokinetic parameters were determined in 44 PICU patients (0.8 month to 14 years old) from all available serum concentrations and doses by nonlinear least squares regression. Population pharmacokinetic parameter estimates were established from 27 of the PICU patients. Mean prediction error (ME) and mean absolute error (MAE) for 2 future sets of peak and trough gentamicin serum concentrations with the use of the population parameter estimates with and without feedback were evaluated in the remaining 17 patients. Mean clearance (+/- SD) and volume of distribution for all 44 patients were 0.123 +/- 0.041 liter/hour/kg and 0.424 +/- 0.116 liter/kg, respectively. Bayesian forecasting of the second set of peak and trough concentrations with feedback from the first set of peak and trough concentrations resulted in smaller bias (peak ME, -0.15 mg/liter; trough ME, 0.13 mg/liter) and better accuracy (peak MAE, 0.91 mg/liter; trough MAE, 0.28 mg/liter) compared with the population parameter estimates alone (peak ME, 0.4 mg/liter; trough ME, 0.28 mg/liter; peak MAE, 1.21 mg/liter; trough MAE, 0.57 mg/liter). This study indicates that gentamicin volume of distribution in PICU patients is larger than non-PICU literature values. The Bayesian program, with specific population parameter estimates for PICU patients, provides accurate initial and subsequent predictions of gentamicin serum concentrations.