UBC Theses and Dissertations
Population pharmacokinetics and Bayesian forecasting of vancomycin in neonates requiring intensive care Wrishko, Rebecca Ellen
PURPOSE The primary objective of this investigation was to develop a population-based pharmacokinetic model of vancomycin in neonates that can be utilized in the individualization of drug therapy. The second objective was to evaluate the accuracy and precision of a Bayesian forecasting method, based on an optimum population pharmacokinetic model, for predicting serum vancomycin concentrations in neonates. METHODS Patients All neonates with a post-conceptional age (PCA) of < 44 weeks admitted to the special care nursery (SCN) of Children's and Women's Health Centre of British Columbia (C & W) between January 01, 1996 and December 31, 1999 and prescribed vancomycin by their attending physicians were eligible for enrollment. Population Pharmacokinetic Modeling Population pharmacokinetic models, using an iterative stepwise approach, were developed for vancomycin with data from 185 patients using a nonlinear mixed effects modeling program (NONMEM). Significant covariates were those that resulted in a decrease in the minimum value of the objective function (MOF) of > 6.6 points. Final one- and twocompartment models were evaluated with data from a naive cohort of 65 patients. Following model validation, combined population pharmacokinetic models were fully developed using data from all 250 patients. As with the original model development, an iterative process was implemented to generate base, full, and final models. Bayesian Forecasting Serum vancomycin concentration predictions based on Bayesian estimates were provided in a NONMEM generated output using the POSTHOC function. Vancomycin concentrations were independently supplied as feedback observations to the final, one-and two-compartment models to obtain case-specific predictions of vancomycin peak and trough concentrations.' The precision and accuracy of Bayesian predictions were assessed using mean absolute error and mean error, respectively, and compared using 95% confidence intervals. RESULTS At all sequential stages, the one-compartment model appeared inferior to the twocompartment model. The minimum values of the objective function (MOF) from the onecompartment unadjusted, base model and revised model, were respectively, 438.52 and 29.84 points greater than the comparable two-compartment values. Weight and PCA (relative to term gestation), modeled as power functions, yielded significant reductions in the MOF when included as covariates on vancomycin clearance. Dopamine exposure was associated with a 34% decrease in vancomycin clearance. Patient weight was modeled as a linear function on the central volume of distribution. Chronic lung disease was associated with a 276% increase in the peripheral volume (Vp). The Vp represented 50% of the volume of distribution at steady-state in the youngest patients, but only 9% in the oldest patients. Model validation demonstrated better accuracy of the two-compartment model. The final, combined models were similar, except that indomethacin was associated with a 16% decrease in vancomycin clearance in the twocompartment model. The two-compartment model was more accurate than the one-compartment model in the Bayesian prediction of initial peak and trough concentrations in neonates < 36 weeks PCA. Bayesian predictions using trough samples as feedback yielded relative mean errors of < 3% for both initial and future peak concentrations. Relative mean absolute error was 6% and 12% for initial and future peak concentrations, respectively. CONCLUSIONS The two-compartment model was superior to the one-compartment model, particularly in neonates < 36 weeks PCA. The better specified two-compartment model also generated more accurate Bayesian predictions of peak and trough concentrations in neonates < 36 weeks PCA. Single trough samples using the two-compartment model and Bayesian forecasting appear to be clinically useful for therapeutic drug monitoring of vancomycin in the SCN population.
Item Citations and Data