External Validation of Population Pharmacokinetic Models for Unbound Cefazolin in Patients Receiving Prophylactic Dosing.

This study aimed to evaluate published population pharmacokinetic models of unbound cefazolin to assess their predictive performance using an independent dataset. A systematic literature search was conducted on PubMed to identify studies evaluating the population pharmacokinetics of unbound cefazolin in patients. Subsequently, the selected models were used for external validation. Predictive bias was visually assessed by plotting the prediction errors (PEs) and relative PEs. Predictive precision was evaluated by calculating the mean absolute error (MAE), root mean square error (RMSE), and mean relative error (MRE). The predictive performance of the 4 unbound population pharmacokinetic models was evaluated using clinical data from 64 patients and 218 unbound concentration samples. The PEs for unbound cefazolin concentrations in the Komatsu model indicated a positive bias, while the RPEs demonstrated similar predictive distributions along the y = 0 line, regardless of the predicted values. In contrast, the other 3 models showed a negative bias for both PE and RPE at unbound cefazolin concentrations. The best MAE, RMSE, and MRE (%) values were 4.71, 9.02, and 30.2 in Komatsu et al.'s model, while the next best values were 11.5, 16.1, and 107.2 in Chung et al.'s model. Both models, which performed best regarding bias and accuracy, were also utilized in studies on unbound concentrations and the correlation between total concentrations and protein-binding sites. This study identified these models as the most suitable for predicting unbound cefazolin concentration profiles in surgical patients.