Model-informed precision dosing: can the administration method influence the predictive performance of a model? An example with piperacillin.
Ibrahim El-Haffaf (1,2), David Williamson (1,3), Djamila Hachemi (3), Thomas Pesout (1,3), Martin Albert (3,4), Amélie Marsot (1,2,5)
(1) Faculty of Pharmacy, Université de Montréal, Montreal, QC, Canada (2) Laboratoire de Suivi Thérapeutique Pharmacologique et Pharmacocinétique, Faculty of Pharmacy, Université de Montréal, Montreal, QC, Canada (3) Hôpital du Sacré-Cœur de Montréal, Montreal, QC, Canada (4) Faculty of Medicine, Université de Montréal, Montreal, QC, Canada (5) Centre de recherche, CHU Sainte-Justine, Montreal, QC, Canada
Introduction: Piperacillin is an extended-spectrum β-lactam antibiotic frequently prescribed in intensive care units (ICUs). Critically ill patients constitute a complex population with multiple pathophysiological alterations that can affect the pharmacokinetic profile of drugs such as piperacillin [1]. As a result, standard doses of piperacillin have been reported to be inadequate for proper target attainment in ICU patients [2, 3].
A recently advocated approach to optimize piperacillin therapy is through the implementation of therapeutic drug monitoring guided by population pharmacokinetic (popPK) models [4, 5]. As piperacillin popPK models can be developed with data obtained from intermittent (II) or continuous infusion (CI) methods, it is unclear whether the administration method used can influence the transferability of a model when evaluated in a dataset with a different administration method. We have previously evaluated the performance of two piperacillin popPK models, one by Klastrup et al. (developed with CI data) and another by Udy et al. (developed with II data), with a small sample size of ICU patients who received piperacillin by CI, and both models performed well [6-8]. The goal of the present external evaluation is to confirm the transferability of these models with a new dataset that consisted solely of II data.
Objectives:
- To evaluate the predictive performance of piperacillin popPK models with an independent dataset.
- Determine possible factors that could affect the predictive performance of these models.
Methods: Independent dataset: piperacillin concentrations were obtained from a prospective observational study conducted in the adult ICU of a tertiary care hospital in Canada, the Hôpital du Sacré-Coeur de Montréal. Piperacillin was administered by II and a series of blood samples were collected: 30 min after the end of the perfusion (peak), at the middle of the dosing interval (middle), and at the end of the dosing interval (trough). Demographic and clinical data were also collected.
External evaluation: piperacillin popPK models were selected based on previously exhibited predictive performance [6]. Both models included the creatinine clearance (CLCr) calculated by the Cockcroft-Gault equation as a covariate on total clearance. External evaluation was performed by calculating prediction errors between the model-predicted concentrations at population-level (PRED) and at individual-level (IPRED) and the observed concentrations to retrieve the bias (median prediction error, MDPE) and the imprecision (median absolute prediction error, MDAPE) for each model. Bias and imprecision were also stratified by sampling time and by renal function category (renal insufficiency (RI), normal and augmented renal clearance (ARC)). Evaluation was performed with NONMEM version 7.5.
Results: The independent dataset included 38 ICU patients with 175 samples: 62 at peak, 60 at middle and 53 at trough. Median (interquartile range) age was 66 (49-73) years, weight was 81.0 (69.1-90.1) kg, for CLCr was 77.9 (45.8-117.8) mL/min, with 15 patients in RI, and 10 in ARC, and main reason of admission to ICU was for trauma (n=14).
Model by Klastrup et al.: overall MDPE and MDAPE for PRED were at 0.59% and 39.66%, respectively. Stratification by sampling time and renal function category revealed that the model performed poorly for trough concentrations (MDPE/MDAPE of -12.8/51.6%) and for patients in normal or ARC (MDPE/MDAPE of -10.6/41.7% and 14.0/41.85%, respectively). IPRED values remained within the boundaries of ±20% and ≤30%.
Model by Udy et al.: overall MDPE and MDAPE for PRED were at -4.02% and 36.39%, respectively. Stratification by sampling time and renal function category revealed that the model had low bias but high imprecision for trough concentrations, (MDPE/MDAPE of -2.4/54.2%) and performed poorly for patients in normal or ARC (MDPE/MDAPE of -18.6/43.2% and -18.3/32.4%, respectively). IPRED values remained within the boundaries of ±20% and ≤30%.
Conclusions: Overall, both models had relatively low bias values, but imprecision remained consistently high for PRED concentrations, particularly for trough concentrations and for patients with normal to augmented renal function. Compared to the previous evaluation carried out with CI data, the models performed worse with II data. This shows the importance of externally evaluating a model before implementation in clinical facilities.
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] Roberts, J.A., et al., DALI: Defining Antibiotic Levels in Intensive Care Unit Patients: Are Current β-Lactam Antibiotic Doses Sufficient for Critically Ill Patients? Clinical Infectious Diseases, 2014. 58(8): p. 1072-1083.
[3] Zander, J., et al., Piperacillin concentration in relation to therapeutic range in critically ill patients – a prospective observational study. Critical Care, 2016. 20(1): p. 79.
[4] Abdul-Aziz, M.H., et al., Antimicrobial therapeutic drug monitoring in critically ill adult patients: a Position Paper#. Intensive Care Medicine, 2020. 46(6): p. 1127-1153.
[5] Guilhaumou, R., 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). Critical Care, 2019. 23(1): p. 104.
[6] El-Haffaf, I., et al., Using a Validated Population Pharmacokinetic Model for Dosing Recommendations of Continuous Infusion Piperacillin for Critically Ill Adult Patients. Clinical Pharmacokinetics, 2022. 61(6): p. 895-906.
[7] Klastrup, V., et al., Population Pharmacokinetics of Piperacillin following Continuous Infusion in Critically Ill Patients and Impact of Renal Function on Target Attainment. Antimicrobial Agents and Chemotherapy, 2020. 64(7): p. e02556-19.
[8] Udy, A.A., et al., Are standard doses of piperacillin sufficient for critically ill patients with augmented creatinine clearance? Critical Care, 2015. 19(1): p. 28.