Impact of unbound fraction variations on model evaluation
To reflect what has been reported in the literature for piperacillin unbound fraction levels among critically ill patients, total concentrations obtained were applied an unbound fraction ranging from 50 to 100%, by increments of 5% [3, 13, 14]. Afterwards, piperacillin predicted concentrations were obtained by using a previously validated population PK (popPK) model and by using mean population estimates reported by the authors [15]. Prediction error (PE, equation 1) was calculated to assess the predictive performance of the model by comparing predicted concentrations with observed concentrations after application of an unbound fraction [16]. Model bias was determined with median prediction error (MDPE, equation 2) and model imprecision was determined with the median absolute prediction error (MDAPE, equation 3) [17]. The predictive performance of the model was acceptable if MDPE was between ± 20% and if MDAPE ≤ 30%. Predicted concentrations were obtained using NONMEM version 7.5 (ICON Development Solutions, Ellicott City, MD, USA), and calculations were performed on R version 4.1.2 using RStudio interface version 1.4.1717.