A meta-analysis of protein binding of flucloxacillin in healthy volunteers and hospitalized patients. (March 2022)

• Record Type: Journal Article
• Authors: Wallenburg, Eveline; Brüggemann, Roger J.M.; Roberts, Jason A.; Jager, Nynke G.L.; Ulldemolins, Marta; Wilkes, Sarah; Schouten, Jeroen; Chin, Paul K.L.; ter Heine, Rob
• Is Part Of: Clinical microbiology and infection. Volume 28:Number 3(2022)
• Page Start: 446.e1
• Page End: 446.e7
• Publication Date: 2022-03
• Subjects: Dose individualization -- Flucloxacillin -- Mechanistic model -- Pharmacokinetics -- Protein binding -- Therapeutic drug monitoring -- Unbound fraction; Medical microbiology -- Periodicals; Diagnostic microbiology -- Periodicals; Communicable diseases -- Periodicals; Infection -- Periodicals; 616.01
• DOI: 10.1016/j.cmi.2021.06.039
• Languages: English
• ISSNs: 1198-743X

Abstract

Abstract: Objectives: The aim of this study was to develop a mechanistic protein-binding model to predict the unbound flucloxacillin concentrations in different patient populations. Methods: A mechanistic protein-binding model was fitted to the data using non-linear mixed-effects modelling. Data were obtained from four datasets, containing 710 paired total and unbound flucloxacillin concentrations from healthy volunteers, non-critically ill and critically ill patients. A fifth dataset with data from hospitalized patients was used for evaluation of our model. The predictive performance of the mechanistic model was evaluated and compared with the calculation of the unbound concentration with a fixed unbound fraction of 5%. Finally, we performed a fit-for-use evaluation, verifying whether the model-predicted unbound flucloxacillin concentrations would lead to clinically incorrect dose adjustments. Results: The mechanistic protein-binding model predicted the unbound flucloxacillin concentrations more accurately than assuming an unbound fraction of 5%. The mean prediction error varied between –26.2% to 27.8% for the mechanistic model and between –30.8% to 83% for calculation with a fixed factor of 5%. The normalized root mean squared error varied between 36.8% and 69% respectively between 57.1% and 134%. Predicting the unbound concentration with the use of the mechanistic model resulted in 6.1% incorrect dose adjustments versus 19.4% if calculated with a fixed unbound fraction of 5%. Conclusions: Estimating the unbound concentration with a mechanistic protein-binding model outperforms the calculation with the use of a fixed protein binding factor of 5%, but neither demonstrates acceptable performance. When performing dose individualization of flucloxacillin, this should be done based on measured unbound concentrations rather than on estimated unbound concentrations from the measured total concentrations. In the absence of an assay for unbound concentrations, the mechanistic binding model should be preferred over assuming a fixed unbound fraction of 5%.