Letao Li1, S.D.T. (Sebastiaan) Sassen1, T.M.J. (Tim) Ewoldt1,2, A. (Alan) Abdulla1, Nicole G. M. Hunfeld1,2, A.E. (Anouk) Muller3, B.C.M. (Brenda) de Winter1, Henrik Endeman2, Birgit C. P. Koch1
1. Department of Hospital Pharmacy, Erasmus MC-University Medical Center; 2 Department of Intensive Care, Erasmus MC-University Medical Center; 3 Department of Medical Microbiology & Infectious Diseases Erasmus MC-University Medical Center
Objectives: Meropenem is a beta-lactam antibiotic that is frequently used to treat serious infections in critically ill patients. Previous population pharmacokinetic (popPK) studies of meropenem in intensive care (ICU) patients have shown large differences in estimated PK parameters, making it difficult to select an appropriate model for clinical use. We performed external validation using real-world patient data to assess the suitability of published meropenem models for clinical application.
Methods: A literature search was conducted in PubMed and EMBASE database according to the PRISMA statement. Published meropenem popPK models in critically ill patients were screened and replicated in NONMEM® 7.4. The predictability of these popPK models was evaluated using a dataset collected from 20 ICU patients (86 blood samples) included in the EXPAT trial who received intravenous meropenem infusion. The model variability was verified by the simulation of typical patients with normal, augmented and compromised renal function. For the concentration prediction evaluation of the models after the starting dose, we compared the population prediction goodness-of-fit (GOF) plots, visual predictive checks (VPC), population prediction error (PE%). For the concentration prediction evaluation of TDM based dosing adjustment, we compared the individual prediction GOF plots, root mean square error (RMSE) and individual PE% of these models.
Results: After screening,13 models were selected for modeling and further valiation. The popPK models of meropenem included in this study differed in both model structural (e.g. number of compartments) and included covariates. These models varied a lot in the simulation of patients with different renal functions. For the initial dosing prediction evaluation, when without information about the measured concentrations and only incorporated the patient relevant covariates information, the population prediction GOF plot and VPC plot were bad. All models seemed to overestimate at concentration below 10 mg/L and underestimate the concentration above 50 mg/L in the population prediction GOF plot. In VPC analysis, most of the models underestimate the peak concentrations. In the initial dosing concentration PE% analysis, all of the models performed bad and the overall predictive power was low, with the 75% quantile range of PE% all above the ±30% threshold. For the TDM based prediction evaluation, three models showed relatively better fit in individual prediction GOF plot and lower RMSE values. When incorporating 1~3 concentrations into these models and calculated the PE%. The 3 best performed models in individual GOF plot and RMSE had 75% quantile range of PE% within ±20% only when 2 or 3 concentrations were utilized.
Conclusions: The predictive performance of the published meropenem popPK models varies greatly when validated on an external dataset of ICU patients receiving meropenem. Our research shows that in real world the critically ill patients have a large variability in meropenem PK. None of these models could have good prediction ability on concentration of the initial dose of the meropenem. On the other hand, some models might be suitable for TDM based model prediction. However, they need first to be validated and require at least two or more sampling in each dosing occasion.
Reference: PAGE 30 (2022) Abstr 10014 [www.page-meeting.org/?abstract=10014]
Poster: Drug/Disease Modelling - Infection