Population pharmacokinetics of vancomycin in adult and geriatric patients: comparison of eleven approaches

A vancomycin population pharmacokinetic prediction model for adult and elderly patients was developed using NONMEM. The predictability of the model was studied and compared with ten other models.Data were collected from routine care of 141 subjects. NONMEM was used to derive a population model. After internal evaluation using the bootstrap technique, external validation was studied using an independent dataset that consisted of 95 subjects; a statistical comparison of precision and bias was conducted.A two-compartment open model was derived with body weight, age, and CLcr as covariates. The bootstrap process showed stability of the model. A comparison of subjects older and younger than 65 years found that the older group had a mean clearance of 2.24 (+/- 1.2) l/h compared to 4.03 (+/- 1.7) l/h, and a peripheral volume of 43.7 (+/- 5.1) l compared to 28.4 (+/- 5.3) l compared to younger patients. These values were modeled using CLcr in the clearance equation and Vd as a function of age. The eleven models studied showed a bias in predicting serum concentrations from the test database that ranged from 0.35 mg/l to -5.93 mg/l. Precision ranged from 4.53 mg/l to 8.05 mg/l. Our method ranked in fourth place overall and when compared statistically its bias was different from the method that ranked in second place by -1.45 (95% CI -2.46, -0.42; p = 0.005), and different from all the methods that ranked worse. The only difference in precision was with the method that ranked in eleventh place with a relative precision of 0.49 (95% CI 0.27, 0.70; p < 0.001).A two-compartment open model fitted the data with weight, age, and CLcr as covariates. The derived method ranked in fourth place overall. The two-compartment nature of two of the equations studied did not provide an advantage. A future study with more data in the distribution phase could provide a model with better predictability.