Anne Ravix (1), Annie E. Cathignol (2,3), Chantal Csajka (1,4,5), Yann Thoma (3,*), Monia Guidi (1,4,6,*)
(1) Center for Research and Innovation in Clinical Pharmaceutical Sciences, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland. (2) Faculty of biology and medicine, University of Lausanne, Lausanne, Switzerland. (3) School of Engineering and Management Vaud, University of Applied Sciences and Arts Western Switzerland (HES-SO), 1401 Yverdon-les-Bains, Switzerland. (4) Institute of Pharmaceutical Sciences of Western Switzerland, University of Geneva, University of Lausanne, Geneva & Lausanne, Switzerland. (5) School of Pharmaceutical Sciences, University of Geneva, University of Lausanne, Geneva, Switzerland. (6) Service of Clinical Pharmacology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland. (*) Equal contribution.
Objectives: Therapeutic Drug Monitoring (TDM) is increasingly used in clinical practice to adjust dosage regimens based on specific patient characteristics and drug concentrations. The Bayesian approach relying on population pharmacokinetic (PK) models enhances classic TDM by predicting individual PK parameters, thus enabling more precise and robust dosing adjustments. Tucuxi, a Model-Informed Precision Dosing (MIPD) software, has been developed by the School of Engineering and Management of the canton Vaud (HEIG-VD) and the University Hospital of Lausanne (CHUV) in Switzerland to help clinicians optimize dosing to achieve the targeted therapeutic interval [1]. The aim of this study was to validate the numerical predictive performance of the Tucuxi software compared to NONMEM, considered as one of the gold standard software for Bayesian PK prediction.
Methods: The methodology used to validate the Tucuxi software was based on previous work proposed by F. Le Louedec et al. [2]. Fictional models were created to mimic various pharmacokinetic scenarios following oral or intravenous (IV) administration. One- and two-compartment models with linear absorption and elimination, including or not a lag time, interindividual variabilities ranging from 45% to 141%, as well as correlations between clearance and volume of distribution, were studied. Alternative types of error models, i.e. proportional, additive, or mixed, were also investigated. For each scenario, a virtual population of 4000 patients receiving doses ranging from 10 to 120 mg every 24 hours was created and divided into two groups: one under a single-dose regimen, and the other under multiple-dose regimen. Sparse and rich profiles were simulated, with either one or four samples per patient taken at 24 ± 10h and 1h ± 1h, 4h ± 1h, 8h ± 2h and 24h ± 6h respectively. The simulated individual concentrations, generated by the population PK models with residual error in NONMEM, constituted the “observed” concentrations in this in-silico validation study.
Tucuxi and NONMEM were then used to predict the concentrations at the time of sampling and estimate PK parameters for each popPK model, dosage regimen and sample scenario: maximum concentration (Cmax), minimum concentration (Cmin), and area under the curve from 0 to 24 hours (AUC0-24h). The predicted concentrations of Tucuxi and NONMEM were compared using the Mean Prediction Error (MPE) and the Root Mean Square Error (RMSE); the closer these parameters are to 0, the more similar the predictions of the two software are. Predicted concentrations and PK parameters were compared by calculating the relative difference between the two estimated values. Tucuxi predictions were considered excellent if the relative error was less than 0.1% compared to the ones predicted by NONMEM, acceptable if it ranged between 0.1% and 10%, and discordant if it exceeded 10%.
Results: A total of 19 (5 IV and 14 oral) models were validated. Overall, concentrations predicted by Tucuxi were very close to those of NONMEM, with a median among all models of 100% [Confidence interval (CI) 95%: 99%, 100%] estimated as excellent, a median MPE and RMSE << 0.01 ‰ with non-significant confidence intervals. The models where inter-individual variability was set at 141% for several parameters showed a slight divergence in concentration predictions. However, this difference remained acceptable as only 1% of predictions were discordant. Similar results were obtained for all PK parameter estimations, with a median of 100% [99%, 100%] excellent predictions, 0% [0%, 1%] acceptable predictions and 0% [0%, 1%] discordant predictions. The latter occurred in models with high interindividual variability.
Conclusions: Tucuxi predictions matched closely with NONMEM, validating the MDPI software ability to accurately forecast patients PK parameters. Further assessment will explore additional kinetics, such as Michaelis Menten elimination or Erlang absorption. The software, available as open source, holds promise for clinical use.
References:
[1] Dubovitskaya A, Buclin T, Schumacher M, Aberer K, Thoma Y, editors. Tucuxi – An intelligent system for personalized medicine: from individualization of treatments to research databases and back. ACM-BCB. 2017.
[2] Le Louedec F, Puisset F, Thomas F, Chatelut É, White‐Koning M. Easy and reliable maximum a posteriori Bayesian estimation of pharmacokinetic parameters with the open‐source R package mapbayr. CPT: Pharmacometrics & Systems Pharmacology. 2021;10(10):1208-20.
Reference: PAGE 32 (2024) Abstr 10806 [www.page-meeting.org/?abstract=10806]
Poster: Methodology - Other topics