Yixuan Cao 1, Inna Galvidis 2, Akmal Alimov 2,3, Joseph F Standing 1, Maksim Burkin 2, Yury Surovoy 4,5
1 University College London ( London WC1E 6BT, United Kingdom), 2 I. Mechnikov Research Institute for Vaccines and Sera (Moscow 105064, Russia), 3 Federal Center for Treatment and Rehabilitation Ministry of Health (Moscow 125367, Russia), 4 Middlesex University (London NW4 4BT, United Kingdom), 5 University College London Hospital (London NW1 2BU, United Kingdom)
Objectives:
Polymyxin B (PMB) is among the few antibiotics retaining activity against carbapenem-resistant Gram-negative pathogens [1], yet its pharmacokinetics (PK) in critically ill patients remain poorly characterised. Critical illness introduces profound physiological derangements, including altered protein binding, fluid shifts, and organ dysfunction, which make standard population PK approaches challenging due to small, heterogeneous cohorts and rapidly evolving pathophysiology [2]. Physiologically based pharmacokinetic (PBPK) modelling offers a mechanistic framework to integrate drug physicochemical properties with patient-specific physiology for predicting drug exposure [3]. The objective of this study was to develop the PBPK model for PMB, calibrated on healthy volunteer and end-stage renal disease (ESRD) data, to predict loading dose exposure in critically ill patients and to evaluate probability of target attainment (PTA) for systemic and pulmonary infections.
Methods:
A whole-body PBPK model was developed in PK-Sim 12.1 [4] using a bottom-up approach. Drug-dependent physicochemical parameters were obtained from published sources. Two key disposition processes were incorporated: transporter-mediated hepatic uptake and renal tubular reabsorption, reflecting the predominantly non-renal elimination of PMB [5]. The model was first calibrated against mean plasma concentration-time profiles digitised from two published studies in healthy volunteers receiving different PMB doses [6, 7]. It was then adapted for ESRD by incorporating disease-specific modifications including reduced glomerular filtration rate, decreased albumin and haematocrit, altered blood cell pH, increased interstitial fluid fraction, and reduced transporter expression [7]. For critically ill patients, 15 individual simulations were constructed using prospectively collected clinical and demographic data, incorporating pathophysiological alterations including increased unbound fraction (35% versus 6% in healthy subjects) [8], reduced albumin and haematocrit, and increased interstitial fluid fraction reflecting sepsis-related tissue oedema [9]. Model evaluation included comparison of predicted and observed non-compartmental PK parameters, with prediction accuracy assessed using predicted-to-observed ratios, mean absolute prediction error (MAPE), and root mean squared error (RMSE). PTA analysis was performed using 1000 virtual critically ill patients across four loading/maintenance dose regimens (1.5/0.75, 2/1, 2.5/1.25, and 3/1.5 mg/kg) with a pharmacodynamic target of fAUC₀₋₂₄/MIC greater than 20 [10], evaluated against MIC values ranging from 0.1 to 8 mg/L.
Results:
In healthy volunteers and ESRD patients, the model demonstrated good predictive performance, with predicted-to-observed ratios for AUC₀₋ₗₐₛₜ, Cmax, and clearance ranging from 0.89 to 1.27. In critically ill patients, population-level predictions were accurate, with predicted AUC₀₋ₗₐₛₜ of 26.16 mg·h/L versus observed 26.65 mg·h/L (ratio 0.98) and predicted Cmax of 8.15 mg/L versus observed 6.77 mg/L (ratio 1.2). Individual-level prediction accuracy showed MAPE of 26.7% and RMSE of 7.7%, with moderate correlation (r = 0.47). PTA analysis revealed that for systemic infections with MIC of 1 mg/L, only loading doses of 2.5 mg/kg and above achieved PTA greater than 90%. For pulmonary interstitial exposures, predicted concentrations were approximately two-fold lower than plasma, and PTA greater than 90% at MIC of 1 mg/L was achieved only with the highest evaluated dose. No dosing regimen achieved adequate PTA for MIC values of 2 mg/L or above.
Conclusions:
This study presents the PBPK model for PMB in critically ill patients demonstrating good prediction of population-level PK parameters and supporting a loading dose of at least 2.5 mg/kg to achieve adequate exposure against pathogens with MIC up to 1 mg/L. The model provides a mechanistic framework that can be refined as PMB distribution, metabolism, and excretion mechanisms become better characterised, and may inform individualised dosing strategies in this challenging patient population.
References:
[1] Velkov T et al. Future Microbiol 2013;8:10.2217/fmb.13.39.[2] Roberts JA et al. Lancet Infect Dis 2014;14:498–509.
[3] Jones H, Rowland-Yeo K. CPT Pharmacometrics Syst Pharmacol 2013;2:e63.
[4] Willmann S et al. BIOSILICO 2003;1:121–124.
[5] Manchandani P et al. Antimicrob Agents Chemother 2016;60:1029–1034.
[6] Liu X et al. J Infect 2021;82:207–215.
[7] Fang Y et al. Clin Transl Sci 2024;17:e70110.
[8] Surovoy YA et al. J Antimicrob Chemother 2022;dkac021.
[9] Radke C et al. Clin Pharmacokinet 2017;56:759–779.
[10] Tsuji BT et al. Pharmacotherapy 2019;39:10–39.
Reference: PAGE 34 (2026) Abstr 12292 [www.page-meeting.org/?abstract=12292]
Poster: Drug/Disease Modelling - Absorption & PBPK