Félicien Le Louedec (1,2,3), Florent Puisset (1,2,3), Fabienne Thomas (1,2,3), Mélanie White-Koning (1,2), Étienne Chatelut (1,2,3)
(1) Cancer Research Center of Toulouse (CRCT), Inserm UMR1037, Toulouse, France. (2) Faculty of Pharmacy, Université Paul Sabatier Toulouse III, France. (3) Institut Claudius-Regaud, Institut Universitaire du Cancer de Toulouse - Oncopole, Toulouse, France
Objectives: Pharmacokinetic (PK) parameter estimation is a critical and complex step in the Model-Informed Precision Dosing (MIPD) approach. Maximum a posteriori Bayesian Estimation (MAP-BE) is currently the gold-standard method to infer individual PK parameters from information relative to the patient (i.e., covariates and concentrations obtained from therapeutic drug monitoring) and a population PK model which includes an a priori distribution of parameters. We developed a free open-source package, named mapbayr, able to perform MAP-BE in R, for the purpose of being used as an engine inside Shiny [1] web applications dedicated to MIPD. The objective of the current study is to validate the performance of mapbayr vs. NONMEM.
Methods: In mapbayr, model definition and differential equation solving rely on mrgsolve [2]. Administration and observation data can be included with the standard NM-TRAN format. The performances of mapbayr were assessed using two approaches. First, “test” models with different features were coded, e.g. oral or intravenous administration; first- and zero-order absorption; lag time; bioavailability; time-varying covariates; Michaelis-Menten elimination; proportional, additive, combined or exponential residual error; parent drug and metabolite; limited or large inter-individual variability. Four thousand PK profiles (combining single/multiple dosing and rich/sparse sampling settings) were simulated for each test model, and MAP-BE of PK parameters was performed in both NONMEM (first-order conditional estimation with interaction method) and mapbayr. For each patient, the exponential of maximum absolute difference on empirical Bayesian estimates (Δθi) was computed as a criterion to evaluate performances of estimation. Secondly, a similar procedure was conducted with six “real” previously published models, in order to compare mapbayr and NONMEM on a PK outcome used in MIPD.
Results: For the “test” models, almost 99% of parameter estimates obtained with mapbayr were identical (Δθi < 0.1%) to those given by NONMEM. Estimations were discordant (Δθi > 10%) in less than 1% of individuals overall. Discrepancies were mainly seen with non-linear PK models and models with large inter-individual variability, especially in sparse sampling settings. Discordances were also seen when lag times or zero-order phenomenon had to be estimated, however objective function values were often lower with mapbayr than NONMEM in this setting, suggesting mapbayr may outdo NONMEM in specific conditions. For the “real” models, a concordance close to 100% on PK outcomes was observed, meaning that dose would have been adapted identically whatever the software used for MAP-BE.
Conclusions: The mapbayr package provides a reliable solution to perform MAP-BE of PK parameters in R, with any non-linear mixed effect model defined with mrgsolve. It also includes functions dedicated to data formatting and reporting. Associated with the mrgsolve framework, it is a valuable means of simulating dose regimens in order to forecast drug exposure a posteriori. In conclusion, our tool enables the creation of standalone Shiny web applications dedicated to MIPD, whatever the model, the clinical protocol and without additional software other than R. It is available on CRAN and github.com.
References:
[1] Wojciechowski J et al. Interactive Pharmacometric Applications Using R and the Shiny Package. CPT: Pharmacometrics & Systems Pharmacology. 2015;4:146–59.
[2] Elmokadem A et al. Quantitative Systems Pharmacology and Physiologically-Based Pharmacokinetic Modeling With mrgsolve: A Hands-On Tutorial. CPT Pharmacometrics Syst Pharmacol. 2019;8:883–93.
Reference: PAGE 29 (2021) Abstr 9647 [www.page-meeting.org/?abstract=9647]
Poster: Methodology - Other topics