Ibrahim El-haffaf 1,2, David Williamson 1,3, Rosalie Voyer 1, Alexandros Cavayas Yiorgos 3, Marc-Alexandre Duceppe 4, Chantale Simard 5, Han Ting Wang 6, Zoé Thiboutot 6, Amélie Marsot 1,2
1 Faculty of pharmacy, Université De Montréal (Montreal, Canada), 2 Laboratoire de suivi thérapeutique pharmacologique et pharmacocinétique (Montreal, Canada), 3 Hôpital du Sacré-Cœur de Montréal (Montreal, Canada), 4 Centre universitaire de santé McGill (Montreal, Canada), 5 Université Laval (Quebec City, Canada), 6 Centre hospitalier de l’Université de Montréal (Montreal, Canada)
Introduction
ECMO (extracorporeal membrane oxygenation) therapy provides temporary cardiac and/or respiratory support to critically ill patients, and anti-infective agents are frequently used in ECMO patients to treat underlying infections. Piperacillin is an extended-spectrum β-lactam antibiotic frequently prescribed in intensive care units (ICUs) and in ECMO patients. Critically ill patients show multiple pathophysiological alterations that can affect the pharmacokinetic (PK) profile of drugs such as piperacillin, and ECMO can further exacerbate these alterations [1]. Standard doses of piperacillin have thus been reported to be inadequate for this subcategory of patients [2].
Population pharmacokinetic (popPK) modeling is an approach recommended by certain societies to describe the PK and explain the variability of drugs in critically ill patients [3]. Despite these recommendations, data remains limited on the impact of ECMO on piperacillin PK, with few models having studied this subpopulation. Moreover, no external evaluation has previously been performed to assess the robustness and transferability of the available piperacillin models in the literature on an independent ECMO dataset.
Objective
To evaluate the predictive performance of piperacillin popPK models with an independent dataset of ECMO patients and to verify whether general ICU (non-ECMO) models perform similarly to ECMO models.
Methods
Dataset
Piperacillin concentrations were obtained from a prospective observational study conducted in three Canadian ICUs to enrol adult ECMO patients receiving piperacillin. Samples were taken at the end of infusion, 1h after, at 50%, 75%, and 100% of the dosing interval. Demographic, clinical and ECMO data were collected.
External evaluation
Piperacillin popPK models were selected based on previous literature reviews [4, 5] and through another review to include new publications up to January 2026. Models were included if they were developed with critically ill patient data using a parametric approach and had sufficient data for code transcription. Predictive performance was assessed by calculating the prediction errors between the model-predicted concentrations at population-level and the observed concentrations to retrieve the bias (median prediction error) and the imprecision (median absolute prediction error) for each model, as well as by graphical assessment. Bias and imprecision values were stratified according to specific parameters: sex, renal function group, renal replacement therapy and study site. Model performances were also compared by ECMO vs non-ECMO models. Models had satisfactory performance if bias was between ±20% and if imprecision was below 30% [6].
If no model displayed adequate performance, we re-estimated models built with ECMO data, and bias and imprecision were recalculated to verify performance. If no model was acceptable after this step, we used the re-estimated models to select initial PK parameter values and optimal model structure to pursue model refinement in a next step with the dataset.
Results
The dataset contained 86 plasma concentrations from 20 patients with a median (interquartile range) age of 55 (38-65) years, bodyweight of 90 (82-107) kg and creatinine clearance of 66.5 (53.1-80.5) mL/min. Nine patients had venovenous (VV) ECMO vs. 11 venoarterial (VA) ECMO, and 7 patients were under renal replacement therapy. Patients received piperacillin mainly by bolus infusion (n=15).
Nineteen popPK models were selected and evaluated with the independent dataset. The predictive performance of the models varied greatly, with bias ranging from -12.9% to 122.2%. Imprecision ranged from 35.6% to 85.5%. The model by Fillâtre et al. displayed the best performance (bias/imprecision of 12.9%/35.6%), which was a model built with data from ECMO and non-ECMO patients [7].
Since no model had adequate performance, we re-estimated two models that included ECMO patients (Fill̢tre et al. and Kim et al.) [7,8]. Re-estimation did not improve their performance (bias/imprecision for Fill̢tre and Kim of Р1.3%/39.9% and 8.8%/47.2%, respectively) and graphical evaluation showed notable inaccuracy in predictions of high concentrations. However, the re-estimated model by Kim et al. offered good estimate precision compared to the one by Fill̢tre et al.
Conclusions
Overall poor predictive performance was observed from the available models in the literature in this cohort of ECMO patients despite model re-estimation. These results highlight potential differences between ECMO and non-ECMO patients that may not be considered in the available models and model refinement may be necessary to better describe this ECMO cohort. We intend to explore potential new covariates to help improve model predictions in the future with the re-estimated model by Kim et al.
References:
1. Blot, S.I., F. Pea, and J. Lipman, The effect of pathophysiology on pharmacokinetics in the critically ill patient — Concepts appraised by the example of antimicrobial agents. Advanced Drug Delivery Reviews, 2014. 77: p. 3-11.
2. Duceppe, MA., Kanji, S., Do, A.T. et al. Pharmacokinetics of Commonly Used Antimicrobials in Critically Ill Adults During Extracorporeal Membrane Oxygenation: A Systematic Review. Drugs 81, 1307–1329 (2021). https://doi.org/10.1007/s40265-021-01557-3
3. Guilhaumou, R., Benaboud, S., Bennis, Y. et al. Optimization of the treatment with beta-lactam antibiotics in critically ill patients—guidelines from the French Society of Pharmacology and Therapeutics (Société Française de Pharmacologie et Thérapeutique—SFPT) and the French Society of Anaesthesia and Intensive Care Medicine (Société Française d’Anesthésie et Réanimation—SFAR). Crit Care 23, 104 (2019).
4. El-Haffaf, I., J.-A. Caissy, and A. Marsot, Piperacillin-Tazobactam in Intensive Care Units: A Review of Population Pharmacokinetic Analyses. Clinical Pharmacokinetics, 2021. 60(7): p. 855-875.
5. Greppmair, S., Brinkmann, A., Roehr, A. et al. Towards model-informed precision dosing of piperacillin: multicenter systematic external evaluation of pharmacokinetic models in critically ill adults with a focus on Bayesian forecasting. Intensive Care Med 49, 966–976 (2023). https://doi.org/10.1007/s00134-023-07154-0
6. Varvel JR, Donoho DL, Shafer SL. Measuring the predictive performance of computer-controlled infusion pumps. J Pharmacokinet Biopharm. 1992;20(1):63–94
7. Fillâtre P, Lemaitre F, Nesseler N, Schmidt M, Besset S, Launey Y, Maamar A, Daufresne P, Flecher E, Le Tulzo Y, Tadie JM, Tattevin P. Impact of extracorporeal membrane oxygenation (ECMO) support on piperacillin exposure in septic patients: a case-control study. J Antimicrob Chemother. 2021 Apr 13;76(5):1242-1249. doi: 10.1093/jac/dkab031. PMID: 33569597
8. Yong Kyun Kim, Hyoung Soo Kim, Sunghoon Park, Hwan-il Kim, Sun Hee Lee, Dong-Hwan Lee, Population pharmacokinetics of piperacillin/tazobactam in critically ill Korean patients and the effects of extracorporeal membrane oxygenation, Journal of Antimicrobial Chemotherapy, Volume 77, Issue 5, May 2022, Pages 1353–1364, https://doi.org/10.1093/jac/dkac059
Reference: PAGE 34 (2026) Abstr 12095 [www.page-meeting.org/?abstract=12095]
Poster: Drug/Disease Modelling - Infection