Johannes Starp (1,*), Antonia Leonhardt (2,*), Lea Marie Schatz (1), Alexandra K. Kunzelmann (1), Sebastian Greppmair (1), Michael Paal (3), Michael Zoller (1), Christina König (4), Jörn Grensemann (4), Christina Janßen (1), Sophie L. Stocker (5), Lana Reiter (5), Cindy Lau (5,6), Deborah Marriott (7,8) , Sebastian G. Wicha (2,#), Uwe Liebchen (1,#), *Shared first authorship #Shared last authorship
(1) Department of Anesthesiology, LMU University Hospital, LMU Munich, Germany, (2) Department of Clinical Pharmacy, Institute of Pharmacy, University of Hamburg, Germany, (3) Institute of Laboratory Medicine, LMU University Hospital, LMU Munich, Germany, (4) Department of Intensive Care Medicine, University Medical Center Hamburg-Eppendorf, Germany, (5) School of Pharmacy, Faculty of Medicine and Health, University of Sydney, NSW, Australia, (6) Department of Pharmacy, St Vincent's Hospital, Darlinghurst, NSW, Australia, (7) School of Clinical Medicine, St Vincent's Healthcare Clinical Campus, Faculty of Medicine and Health, UNSW Sydney, NSW, Australia, (8) Department of Clinical Microbiology and Infectious Diseases, St Vincent's Hospital, Darlinghurst, NSW, Australia
Objectives: Overdosing of linezolid increases the risk of adverse events such as thrombocytopenia, lactic acidosis, or neuropathy, while underdosing may lead to therapeutic failure or resistance. Model-informed precision dosing (MIPD) of linezolid can increase attainment of target exposure (2-8 mg/L) compared to standard dosing, However, the accuracy of dose predictions are dependent on the population pharmacokinetic model (popPK) selected, so the choice of the popPK model is crucial for a successful implementation of MIPD [1]. Which popPK model appropriately describes a population can only be known after external evaluation. Therefore, the aim of this study was to evaluate the predictive performance of existing popPK models of linezolid in critically ill patients .
Methods: 31 popPK models (application: 18 intravenous, 13 oral) were identified from literature and encoded in NONMEM®7.4. The predictive performance of each model was assessed using clinical data from 194 patients (162 intravenous, 28 oral, 4 both) and 594 linezolid concentrations (498 intravenous, 96 oral) obtained from 3 sites (Munich, Hamburg, Sydney). All patients were treated with linezolid and had at least 3 observed dosing occasions. Oral and intravenous administrations were evaluated separately. The predictive performance was evaluated for the third observed dosing occasion in three scenarios. First, only the patients characteristics and dosing information were used for the prediction (a priori, AP). Further, either 1 or 2 previous trough concentrations were included (Bayesian forecasting 1, B1 and Bayesian forecasting 2, B2, respectively). Forecasting performance was analyzed by different metrics. For each model and each scenario (AP, B1, B2) the prediction errors (PE), median prediction error (MPE) and median absolute prediction error (MAPE) were calculated. The target attainment (TA) was calculated as the expected percentage of samples attaining a linezolid target concentration of 2-8 mg/L based on PE.
Results: Linezolid popPK models were heterogenous: The number of compartments was 1 to 3, the most common covariates were total body weight and creatinine clearance. Few models described clearance as time-dependent. After intravenous administration for the AP scenario, the MPE ranged from -81.5% to 334.6% and the MAPE from 59.7% to 334.6% and the TA was between 17.2% and 50.3%. In the B1 scenario the MPE (-41.1% to 286.4%), MAPE (40.0% to 286.4%) and TA (19.6% to 65.0%) improved. In the B2 scenario, the MPE (‑28.5% to 229.4%) and MAPE (29.2% to 229.4%) and TA (25.2% to 79.8%) were further improved. After oral administration, for the AP scenario, MPE ranged from -81.1% to 77.3%, MAPE from 56.2% to 81.7% and TA was between 28.6% and 60.0%. In the B1 scenario, the MPE (-29.6% to 92.1%), MAPE (33.4% to 93.5%) and TA (36.7% to 70.0%) improved. In the B2 scenario, the MPE (-19.9% – 38.8%), MAPE (32.4% – 66.6%) and TA (46.7% to 78.6%) were further improved. Overall, the models from Abe et al 2009 [2], Fang et al 2021 [3], Taubert et al 2016 [4] and Wang et al 2021 [5] showed the best performance considering MPE, MAPE and TA.
Conclusion: Linezolid displays highly variable pharmacokinetics across patient populations which is reflected in the heterogeneity of the published population pharmacokinetic models. The use of linezolid concentrations (particularly more than 1) improves the accuracy and precision of model-predicted drug exposure compared to patient characteristics alone. Four published population pharmacokinetic models have adequate predictive performance to potentially be used in dose prediction software (e.g. TDMx) to inform clinical practice.
References:
[1] Wicha SG, Märtson A, Nielsen EI, Koch BCP, Friberg LE, Alffenaar J, et al. From Therapeutic Drug Monitoring to Model‐Informed Precision Dosing for Antibiotics. Clin Pharma and Therapeutics. 2021 Apr;109(4):928–41.
[2] Abe S, Chiba K, Cirincione B, Grasela TH, Ito K, Suwa T. Population Pharmacokinetic Analysis of Linezolid in Patients With Infectious Disease: Application to Lower Body Weight and Elderly Patients. The Journal of Clinical Pharmacology. 2009 Sep;49(9):1071–8.
[3] Fang J, Zhang XS, Zhang CH, Zhou ZY, Han L, Wang YX, et al. Model Based Identification of Linezolid Exposure–toxicity Thresholds in Hospitalized Patients. Front Pharmacol. 2021 Oct 5;12:732503.
[4] Taubert M, Zoller M, Maier B, Frechen S, Scharf C, Holdt LM, et al. Predictors of Inadequate Linezolid Concentrations after Standard Dosing in Critically Ill Patients. Antimicrob Agents Chemother. 2016 Sep;60(9):5254–61.
[5] Wang X, Wang Y, Yao F, Chen S, Hou Y, Zheng Z, et al. Pharmacokinetics of Linezolid Dose Adjustment for Creatinine Clearance in Critically Ill Patients: A Multicenter, Prospective, Open-Label, Observational Study. DDDT. 2021 May;Volume 15:2129–41.
Reference: PAGE 32 (2024) Abstr 11134 [www.page-meeting.org/?abstract=11134]
Poster: Clinical Applications