Predictive performance of reported population pharmacokinetic models of vancomycin in Chinese adult patients

What is known and objective: There are numerous studies on population pharmacokinetics of vancomycin in adult patients. However, there is no such research for Chinese adult patients. This study was conducted to evaluate the predictive performance of reported population pharmacokinetic models of vancomycin in Chinese adult patients and to identify some models appropriate for our population.

Methods: A literature search was conducted in PubMed to obtain the population pharmacokinetic models of vancomycin published between December 2010 and September 2012. The models were assessed using concentration data collected from Chinese patients for external validation. Models with relatively poor predictability were excluded from further analysis. The performance of the remaining models was evaluated in patients with different levels of creatinine clearance, age, body weight and sex by Bayesian method. This method was also used to compare the predictive performance based on peak concentration and trough concentration and the predictability based on different number of observed concentrations.

Results: One hundred and sixty-five blood concentrations from 72 Chinese adult patients were collected retrospectively to serve as the test data set. The evaluated models included all those reported in the seven publications reviewed by Marsot et al. and three other studies published after December 2010. Three models with poor performance on external validation were excluded from the next Bayesian analysis. The distribution of covariates in the model building data set had an important effect on prediction. The predictability based on peak/trough concentration was similar among the evaluated models, and no significant difference was found using our data set except for Roberts' model. As expected, an increased number of samples improved the performance of the Bayesian prediction.

What is new and conclusion: With our data set, the performance of the evaluated models varied. The characteristics of the patient population and distribution of covariates should be given more consideration when choosing a model to predict blood concentrations. The model developed by Purwonugroho et al. using a data set from patients similar to ours is appropriate for Bayesian dose predictions for vancomycin concentrations in our population of Chinese adult patients.