Mélanie Guhl(1), François Mercier(2), Satish Sharan(3), Kairui Feng(3), Guoying Sun(4), Wanjie Sun(4), Mark Donnelly(3), Stella Grosser(4), Liang Zhao(3), Lanyan Fang(3), France Mentré(1), Emmanuelle Comets(1), and Julie Bertrand(1)
(1) University of Paris, INSERM, IAME, UMR 1137, 75006 Paris, France (2) Department of Biostatistics, Roche Innovation Center Basel, Basel, Switzerland (3) Division of Quantitative Methods and Modeling, Office of Research and Standards, Office of Generic Drugs, Center for Drug Evaluation and Research, U.S. Food and Drug Administration, Silver Spring MD 20993, USA (4) Office of Biostatistics, Office of Translational Sciences, Center for Drug Evaluation and Research, U.S. Food and Drug Administration, Silver Spring MD 20993, USA
Introduction: Model-based approaches (MB-TOST) have been advocated to test for equivalence in sparse PK studies but questions about their performance in the presence of model misspecification remain[1]. In the present work, we studied the pharmacokinetic (PK) equivalence between two formulations of gantenerumab (Roche, Switzerland), a monoclonal antibody used for the treatment of Alzheimer’s disease, as an example to test the impact of model selection on the performance of PK equivalence testing. PK equivalence is established by ensuring that the ratio of drug exposures, measured by area under the curve (AUC) and maximum concentration (Cmax), remains within predetermined limits of 80% to 125% through two one-sided tests (TOST)[2].
Objectives: To evaluate the performance of MB-TOST using a simulation study based on observed data, investigating the impact of design and model misspecification, by comparing it to the traditional noncompartmental approach (NCA-TOST).
Data: The data come from two phase I randomised parallel clinical trials of a high concentration liquid formulation versus the reference lyophilised formulation of gantenerumab, at different doses (105 and 225 mg), in healthy subjects, with 24 subjects per treatment arm and 11 or 12 sampling points per subject.
Methods: We first analysed the observed data using MB-TOST based on non-linear mixed effect models. The analysis was performed separately in the different groups (dose/study): in each group, we tested different structural models and variability structures using bayesian information criterion (BIC), combined with a second criteria of a relative SE below 50% for all parameters, on reference data.
PK equivalence was tested using asymptotic standard errors (MB-TOST Asympt) from the estimated Fisher Information Matrix (FIM), and the results were compared to those obtained via NCA-TOST. To assess the impact of the design, we created a sparse subset of the observed data using 5 optimal sampling points[3]. Selection of the model was based on the reference data and several methods for computation of SEs were compared: MB-TOST Asympt, Gallant (correction of the SEs obtained from the FIM)[4] and Post (SEs obtained from a posteriori distributions)[5].
This analysis inspired a simulation study of 1000 datasets of rich and sparse parallel design with a two-compartment PK model. Treatment effects were simulated on Cl/F and V/F, at the boundary conditions at 0.8 and 1.25 to evaluate type I error, and at 0.9, 1 and 1.11 to evaluate power.
On rich designs, we compared NCA-TOST and MB-TOST Asympt. In the MB approach, we fitted 2 two-compartment PK models, with treatment effects estimated on all parameters (2cpt_par) or only on ka and F (2cpt_F), to explore the impact of the treatment effect model.
On sparse designs, we compared MB-TOST Asympt, Gallant and Post using 2cpt par or a one-compartment model with treatment effects estimated on all parameters (1cpt_par) to investigate the impact of the structural PK model. We also explored the relevance of a model selection step between those two models on reference data using BIC.
Results: For observed data, a two-compartment model was found to be the best model. The results of the NCA and MB approaches were concordant. The PK equivalence of the two formulations could be shown at dose 105 mg, but not at dose 225 mg. The sampling design did not impact these results, though the model selected was different on sparse data (one-compartment model).
On rich simulations, NCA-TOST and MB-TOST Asympt, using the 2cpt_par model, gave consistent results, i.e. type I error close to the nominal value. The 2cpt_F led to an inflated type I error on AUC. On sparse simulations, MB-TOST Asympt and Post gave similar controlled type I error with 2cpt_par, but type I error was inflated on Cmax when using 1cpt_par. Type I error with MB-TOST Gallant tended to be too conservative. A model selection step prior to the MB-TOST almost always selected the simulated model and thereby ensured a type I error at the nominal level.
Conclusions: Given sparse design data, MB-TOST appears to be a robust alternative to NCA methods, provided that the PK model is well specified. The sparse simulation study shows how model selection is key to maintaining an appropriate type I error for equivalence testing using MB-TOST.
Acknowledgements: This work was supported by the U.S. Food and Drug Administration (FDA) under contract 75F40119C10111. The authors thank FDA for this funding. The views expressed in this abstract do not necessarily reflect the views or policies of the FDA.
References:
[1] A Dubois, M Lavielle, S Gsteiger, E Pigeolet et F Mentré (2011). Model-based analyses of bioequivalence crossover trials using the stochastic approximation expectation maximisation algorithm, Statistics in Medicine, 30(21):2582–2600
[2] U.S. Food and Drug Administration (2013). Bioequivalence studies with pharmacokinetic endpoints for drugs submitted under an ANDA. https://www.fda.gov/media/87219/download
[3] C. Dumont, G. Lestini, H. Lenagard, F. Mentré, E. Comets, T. T. Nguyen, and the PFIM group (2018). PFIM 4.0, an R program for design evaluation and optimisation in nonlinear mixed effect models, Computer Methods and Programs in Biomedicine, 156:217–229
[4] J Bertrand, E Comets, M Chenel, and F Mentré (2012). Some alternatives to asymptotic tests for the analysis of pharmacogenetic data using nonlinear mixed effects models, Biometrics, 68(1):146–155 [5] F Loingeville, J Bertrand, TT Nguyen, S Sharan, K Feng, W Sun, J Han, S Grosser, L Zhao, L Fang, K Möllenhoff, H Dette, and F Mentré (2020). New model-based bioequivalence statistical approaches for pharmacokinetic studies with sparse sampling, The AAPS Journal, 22(6):1412
Reference: PAGE 29 (2021) Abstr 9719 [www.page-meeting.org/?abstract=9719]
Poster: Drug/Disease Modelling - Other Topics