Alexandre Duong (1,2), Chantale Simard (3,4), David Williamson (2,5), Amélie Marsot (1,2,6)
(1) Laboratoire de Suivi Thérapeutique Pharmacologique et Pharmacocinétique, Montréal (Qc), Canada (2) Faculté de Pharmacie, Université de Montréal, Montréal (Qc), Canada (3) Faculté de Pharmacie, Université Laval, Québec (Qc), Canada (4) Centre de Recherche de l’Institut Universitaire de Cardiologie et de Pneumologie de Québec, Québec (Qc), Canada (5) Hôpital du Sacré-Cœur de Montréal, Montréal (Qc), Canada (6) Centre de Recherche CHU Sainte-Justine, Montréal (Qc), Canada
Introduction and Objectives: In a context of anti-infective drug overuse and resistance, their proper use and the prevention of resistant strain selection is a public health priority [1,2]. The use of the right antibiotic at the right dose, especially in vulnerable populations with severe infections for which treatment is life threatening, is a priority [2]. Intensive care patients are amongst the most vulnerable populations, as they have pathophysiological conditions that influence the pharmacokinetics of drugs and more particularly antibiotics [3]. Hence, depending on the clinical context and the patient’s comorbidities, subtherapeutic or potentially toxic concentrations can be observed for the same dosage. Population-based Pharmacokinetic/pharmacodynamic (PK/PD) modeling aims to describe the exposure of a drug over time in patients as well as the different sources of variability that may influence the latter. The most robust method to evaluate a model is the external evaluation. This method consists of testing the predictive capabilities of this model with the help of an independent database. The objective of this study was to evaluate the predictive performance of gentamicin population-pharmacokinetic models (Pop-PK) with two independent datasets of critically ill adult patients.
Methods: A literature review was performed to determine the several gentamicin models in critically ill adult patients [4]. In parallel, gentamicin dosing data, information on the treatment, the patient and the bacteria were collected retrospectively in two Quebec establishments (Hôpital du Sacré-Coeur de Montréal (HSCM) and Institut Universitaire de Cardiologie et Pneumologie de Québec (IUCPQ) from 2009 to 2019 and 2014 to 2020, respectively. The external evaluations were performed using NONMEM® version 7.5 with each dataset independently and then, with both datasets combined. Predictive performance was firstly assessed based on the estimation of bias and imprecision with the help of prediction error (PE) [5]. With Rstudio®, goodness of Fit (GOF) plots, visual predictive check (VPC) and normalized-prediction distribution error (NPDE) were also performed to evaluate the stability of the model.
Results: Eleven models were identified, however only four were retained due to missing information or the usage of a software other than NONMEM®. Bias values ranged between -20.09% to 5.50%, -17.36% to 14.96% and -18.02% to 10.56%, respectively for HSCM, IUCPQ and both populations combined. Imprecision values ranged between 12.97% to 26.16%, 20.90% to 27.32% and 18.86% to 27.06%, respectively for HSCM, IUCPQ and both populations combined. For most models, results from the NPDE analyses and VPCs showed poor predicting ability of these models for each dataset independently and with both datasets combined.
Conclusions: Although the bias and imprecision values were within the acceptable range of -20% to 20% and ≤30% [6] respectively, the important variability observed across the published models and the two critically ill adult populations still remains. The latter could be explained by the differences in terms of demographic characteristics between the HSCM and IUCPQ datasets and the study populations in the literature. Although most of the characteristics were deemed similar, creatinine serum and sex were statistically different between the two establishments. This can explain the difference in the predicted concentrations and the calculated PK parameters. This study brings to light of the necessity of predictive validation of Pop-pk models, especially in vulnerable populations.
References:
[1] Déclaration des Nations Unies en 2016, https://www.un.org/press/fr/2016/ag11835.doc.htm (Access le 15/04/2019)
[2] World Health Organization, Antibiotic resistance 5 February 2018. https://www.who.int/news-room/fact-sheets/detail/antibiotic-resistance
[3] Dombrovskiy VY, Martin AA, Sunderram J, Paz HL. Rapid increase in hospitalization and mortality rates for severe sepsis in the United States: a trend analysis from 1993 to 2003. Crit Care Med. 2007May;35(5):1244-50
[4] Duong A, Simard C, Wang YL, Williamson D, Marsot A. Aminoglycosides in the Intensive Care Unit: What Is New in Population PK Modeling? Antibiotics (Basel). 2021 Apr 29;10(5):507. doi: 10.3390/antibiotics10050507. PMID: 33946905.
[5] Guang W, Baraldo M, Furlanut M. Calculating percentage prediction error: A user’s note. Pharmacological Research. 1995;32(4):241-8. doi:https://doi.org/10.1016/S1043-6618(05)80029-5.
[6] Hara M, Masui K, Eleveld DJ, Struys M, Uchida O (2017) Predictive performance of eleven pharmacokinetic models for propofol infusion in children for long-duration anaesthesia. Br J Anaesth 118(3):415 – 423
Reference: PAGE 29 (2021) Abstr 9864 [www.page-meeting.org/?abstract=9864]
Poster: Methodology - Model Evaluation