Modified model of drug induced thrombocytopenia efficiently projects safe starting dose in human from preclinical data
C. Meille (1), M.B. Reddy (1), A. Walz (1), S. Retout (2), G. Nichols (3), K. Glenn (4), M. Pignatello (5), D. Bottino (2), J. Zhi (6), S. Middleton (3), and T. Lavé (1)
Pharma Research and Early Development (pRED), F. Hoffmann-La Roche Ltd., (1) Non Clinical Safety (NCS), DMPK, Modeling and Simulation; (2) Translational Research Sciences (TRS), Modeling and Simulation (3) Oncology Translational Medicine Group; (4) Non Clinical Safety (NCS), DMPK; (5) Non Clinical Safety (NCS), Toxicology; (6) Clinical Pharmacology (ClinPharm)
Objectives: A semi-mechanistic model quantifying hematotoxicity of anticancer agents has been proposed by Friberg1 with a chain of 5 compartments reproducing the hematopoietic maturation process. The study aim was, for a new compound RO_A inducing thrombocytopenia in cynomolgus monkey, to 1) adapt this model to describe platelet (PLT) time-course profiles, 2) project the appropriate dosing regimen in humans, and 3) calculate the risk of grade 4 thrombocytopenia at the selected starting dose.
Methods: The analysis included 244 PLT and 144 PK observations from 48 monkeys receiving RO_A oral doses from 10 mg to 500 mg tested in single or repeat dose studies. PK was modeled by a 2-compartment model accounting for an enzyme induction effect. Adaptations of the Friberg model were: zero-order production of progenitors inhibited by drug; effect compartment linking PK to PD; addition of a local feedback mechanism dependent on number of cells in the first compartment and affecting proliferation of cells in subsequent compartments. Population approach was used to estimate PK and PD parameters with MONOLIX 3.22. Model performance was assessed through AIC criterion, goodness of fit and posterior predictive check. Human PK of RO_A was predicted using physiologically based pharmacokinetic (PBPK) modeling in GastroPlus3. Human PLT profiles were simulated with doses from 30 to 2000 mg and considering that the drug potency on progenitors is translational across species. The risk of grade 4 thrombocytopenia was assessed by Monte Carlo simulation.
Results: Adding local feedback allowed a good description of a transient increase of PLT in monkeys while thrombocytopenia and subsequent rebound were well described for all the dosing regimens. Simulation of expected PLT profiles in human showed that concentrated daily administration (day 1 to day 5) for 28-day cycles was better tolerated. Risk of grade 4 thrombocytopenia during the first 2 cycles was estimated to be 4%. The overlay of observed and predicted PLT kinetics at the human starting dose was consistent.
Conclusions: A PKPD model of thrombocytopenia was applied on monkey data for human prediction. A new feedback mechanism was needed to describe monkey PLT profiles. With support of modeling, a 100 mg starting dose was selected rather than a 30 mg dose and two escalation steps were saved in the phase I study. Observed PLT profiles at the starting dose are in line with model predictions.
References:
[1] Friberg L. et al. J Clin Oncol. 2002 Dec 15;20(24):4713-21
[2] Kuhn E., Lavielle M. "Maximum likelihood estimation in nonlinear mixed effects models" Computational Statistics and Data Analysis, vol. 49, No. 4, pp 1020-1038, 2005
[3] GastroPlus v6.1, www.simulations-plus.com
