Anthéa Deschamps 1,2, Lionel Velly 3,4, Florence Gattacceca 2, Romain Guilhaumou 4,5
1 Aix Marseille University, APHM, Department of Clinical Pharmacology And Pharmacosurveillance, University Hospital Timone (Marseille, France), 2 COMPutational Pharmacology and Clinical Oncology Department, Inria Sophia Antipolis - Méditerranée, Cancer Research Center of Marseille, Inserm UMR1068, CNRS UMR7258, Aix Marseille University UM105 (Marseille, France), 3 Aix Marseille University, APHM, Department of Anaesthesiology and Critical Care Medicine, University Hospital Timone (Marseille, France), 4 Aix Marseille University, CNRS, INT, Institute Neuroscience Timone, UMR7289 (Marseille, France), 5 Aix Marseille University, CNRS, INS, Institute Neuroscience Systeme, UMR 1106 (Marseille, France)
Background
Critically ill patients display pathophysiological features that significantly affect the pharmacokinetics of drugs. In particular, renal elimination processes exhibit substantial inter- and intra-individual variability, reflecting the highly dynamic nature of renal function in this population. Such variability represents a critical determinant of pharmacokinetic behaviour for renally eliminated drugs, and may result in supra- or sub-therapeutic drug exposition. Serum creatinine (Crs), routinely measured in clinical practice, represents an accessible biomarker of renal function in intensive care units (ICU) patients, which can change due to acute kidney injury or augmented renal clearance. Mechanistic biomarker modelling can offer a structured and physiologically interpretable approach to describe these dynamics. The aim of this study was then to develop a serum creatinine biomarker model that can describe renal function evolution in ICU population.
Methods
In a prospective, multicenter, observational study (October 2015 – May 2017) conducted in ICU of Marseille, demographic, clinical, biological and pharmacological data were collected at inclusion and throughout 7 days. Crs concentrations were collected at least once daily during the 7-day follow-up. Data were analysed by nonlinear mixed-effects modeling using MONOLIX version 2024R1. Model development followed three sequential steps. First, a mechanistic turnover model was established to describe creatinine kinetics and identify periods of creatinine accumulation through modulation of degradation processes. Secondly, a probabilistic three-state model was implemented to characterize underlying renal functional states (stable, impairment, improvement), corresponding to stable, increasing, and decreasing Crs dynamics. The states were coded based on observed Crs trends and imputed as categorical (count-type) data reflecting biomarker dynamics. Finally, the Crs turnover model was structurally linked to the probability model through the instantaneous Crs slope, allowing dynamic interaction between continuous Crs concentrations and discrete renal state. Graphical (Goodness-of-fit plots, scatter plots of individual weighted residuals, visual predictive checks) and statistical criteria as well as uncertainty of parameter estimates were used to assess model performance.
Results
The study included 76 ICU hospitalized patients (31 females, 45 males, age = 57,5 ± 17,6 years). 597 serum creatinine were available for biomarker model (median = 54.0 µmol/L [min 24.0 – max 251 µmol/L]). The dynamics of Crs were described by a turnover model, in which synthesis was parameterized based on the stable equilibrium of Crs (ksyn = kdeg × Crs_stable), corresponding to a stable renal function_ The typical Crs degradation rate indicated slow dynamics (kdeg_pop 0.0086 h⁻¹) with Crs_init_pop (68.2 µmol/L) > Crs_stable_pop (41.9 µmol/L) consistent with many patients starting out of stable renal function. The turnover model adequately described stable creatinine phases but failed to capture periods of creatinine increase. The transient reduction in elimination capacity was modelled as inhibition of creatinine degradation through a binary regressor (INH), such that kdeg_eff = kdeg × (1 − Ideg × INH). The inhibition parameter Ideg_pop (0.77) supported substantial inhibition of creatinine elimination when INH is active. Inter-individual variability was estimated on kdeg, Crs_base, Crs_init, and Ideg.
The dynamic renal states was implemented with multinomial logistic model, providing good parameter estimate uncertainty. At each time point, the model estimated the probability of belonging to each state. The instantaneous Crs slope, derived from the turnover model, (slope = ksyn – kdeg_eff*Crs), was introduced as a predictor in the multinomial component, leading to improved parameter stability and visual predictive check performance of each state. The log-odds of the stable versus increasing state were modeled as log(P₁/P₂) = b10 + b11·slope and those of the decreasing versus increasing state as log(P₃/P₂) = b30 + b31·slope. The negative slope coefficients (b11_pop = −2.47; b31_pop = −6.47) indicate that increasing Crs slopes strongly shift probability toward the increasing state, while negative slopes markedly favor the decreasing state.
Conclusions
A mechanistic turnover model of Crs was developed to describe time-varying renal function in the context of adult ICU patients. Modeling transient impairment as inhibition of Crs degradation, combined with slope-driven state classification of renal function evolution offer clinically meaningful interpretation of Crs trajectories. This framework opens the perspective for individualized monitoring through a maximum a posteriori estimation procedure coupled with iterative updating of the inhibition regressor, enabling real-time renal state prediction.This framework offers strong potential for integration into population PK models to predict the inter and intra-individual variability in ICU patients.
References:
(1) Hoste, E. A. J.; Bagshaw, S. M.; Bellomo, R.; Cely, C. M.; Colman, R.; Cruz, D. N.; Edipidis, K.; Forni, L. G.; Gomersall, C. D.; Govil, D.; Honoré, P. M.; Joannes-Boyau, O.; Joannidis, M.; Korhonen, A.-M.; Lavrentieva, A.; Mehta, R. L.; Palevsky, P.; Roessler, E.; Ronco, C.; Uchino, S.; Vazquez, J. A.; Vidal Andrade, E.; Webb, S.; Kellum, J. A. Epidemiology of Acute Kidney Injury in Critically Ill Patients: The Multinational AKI-EPI Study. Intensive Care Med. 2015, 41 (8), 1411–1423. https://doi.org/10.1007/s00134-015-3934-7.
(2) De Waele, J. J.; Dumoulin, A.; Janssen, A.; Hoste, E. A. Epidemiology of Augmented Renal Clearance in Mixed ICU Patients. Minerva Anestesiol. 2015, 81 (10), 1079–1085.
(3) Dayneka, N. L.; Garg, V.; Jusko, W. J. Comparison of Four Basic Models of Indirect Pharmacodynamic Responses. J. Pharmacokinet. Biopharm. 1993, 21 (4), 457–478. https://doi.org/10.1007/BF01061691.
(4) Agresti, A. Analysis of Ordinal Categorical Data, 1st ed.; Wiley Series in Probability and Statistics; Wiley, 2010. https://doi.org/10.1002/9780470594001.
Reference: PAGE 34 (2026) Abstr 12208 [www.page-meeting.org/?abstract=12208]
Poster: Drug/Disease Modelling - Other Topics