Towards model-informed precision dosing of piperacillin: multicenter systematic external evaluation of pharmacokinetic models in critically ill adults with a focus on Bayesian forecasting

Abstract

Purpose Inadequate piperacillin (PIP) exposure in intensive care unit (ICU) patients threatens therapeutic success. Model-informed precision dosing (MIPD) might be promising to individualize dosing; however, the transferability of published models to external populations is uncertain. This study aimed to externally evaluate the available PIP population pharmacokinetic (PopPK) models. Methods A multicenter dataset of 561 ICU patients (11 centers/3654 concentrations) was used for the evaluation of 24 identified models. Model performance was investigated for a priori (A) predictions, i.e., considering dosing records and patient characteristics only, and for Bayesian forecasting, i.e., additionally including the first (B1) or first and second (B2) therapeutic drug monitoring (TDM) samples per patient. Median relative prediction error (MPE) [%] and median absolute relative prediction error (MAPE) [%] were calculated to quantify accuracy and precision. Results The evaluation revealed a large inter-model variability (A: MPE − 135.6–78.3% and MAPE 35.7–135.6%). Integration of TDM data improved all model predictions (B1/B2 relative improvement vs. A: |MPE| median_all_models 45.1/67.5%; MAPE median_all_models 29/39%). The model by Kim et al. was identified to be most appropriate for the total dataset (A/B1/B2: MPE − 9.8/− 5.9/− 0.9%; MAPE 37/27.3/23.7%), Udy et al. performed best in patients receiving intermittent infusion, and Klastrup et al. best predicted patients receiving continuous infusion. Additional evaluations stratified by sex and renal replacement therapy revealed further promising models. Conclusion The predictive performance of published PIP models in ICU patients varied considerably, highlighting the relevance of appropriate model selection for MIPD. Our differentiated external evaluation identified specific models suitable for clinical use, especially in combination with TDM. Supplementary Information The online version contains supplementary material available at 10.1007/s00134-023-07154-0.

Achievement of therapeutic antibiotic exposures using Bayesian dosing software in critically unwell children and adults with sepsis

“…Finally, some of the models used in the MIPD software may have been unable to accurately predict target exposures as the models used total antibiotic concentrations while the investigators used unbound concentrations for TDM measurements [ 28 ]. Commentators have highlighted the absence of external validation of some of the models used by the software [ 29 ], while a recent multicentre evaluation of 24 popPK piperacillin models illustrated substantial inter-model variability leading to highly variable predictive performance [ 30 ]. However, meta-analyses of individualised antimicrobial dose optimisation confirm the achievement of higher rates of target exposures and have suggested reductions in treatment failure and adverse outcomes, such as nephrotoxicity [ 31 , 32 ].…”

Modeling Pharmacokinetics in Individual Patients Using Therapeutic Drug Monitoring and Artificial Population Quasi-Models: A Study with Piperacillin

“…Several piperacillin population pharmacokinetic models are available in the literature, predominantly with the consideration of single- or, more often, two-compartment models including creatinine clearance as a covariate [ 22 , 23 , 24 ]. Nevertheless, it has recently become apparent that the suitability of these models for making estimates of individual pharmacokinetic properties can be highly variable.…”

“…A multicenter study revealed only three models which provided acceptable estimates and absolute prediction errors. Interestingly, the authors of this study concluded that the accuracy of estimates was gender-dependent [ 24 ].…”

Achievement of therapeutic antibiotic exposures using Bayesian dosing software in critically unwell children and adults with sepsis

“…Finally, some of the models used in the MIPD software may have been unable to accurately predict target exposures as the models used total antibiotic concentrations while the investigators used unbound concentrations for TDM measurements [25]. Commentators have highlighted the absence of external validation of some of the models used by the software [26], while a recent multicenter evaluation of 24 popPK piperacillin models illustrated substantial inter-model variability leading to highly variable predictive performance [27]. Meta-analyses of individualised antimicrobial dose optimisation con rm the achievement of higher rates of target exposures and have suggested reductions in treatment failure and adverse outcomes such as nephrotoxicity [28,29].…”