A Systematical Approach to Bridge the Two-Stage Parametric Expectation Maximization Algorithm and Full Bayesian Three-Stage Hierarchical Nonlinear Mixed Effect Methods in Complex Population Pharmacokinetic/Pharmacodynamic Analysis: Troxacitabine-induced Neutropenia in Cancer Patients
Ng CM (1), Bauer RJ (2), Beeram M (1), Takimoto CH (1), Lin C (1), Patnaik A (1)
(1) Institute for Drug Development, Cancer Therapy and Research Center, San Antonio, TX. (2) XOMA (US) LLC, Berkeley, CA.
Background: The full Bayesian approach has been suggested as a suitable method for population pharmacokinetic/pharmacodynamic (PK/PD) modeling. However, to this day, published examples of its application to real population PK-PD problems are very limited due to time/labor intensive, and difficulty in achieving model convergences. Monte-Carlo parametric expectation maximization (MCPEM) is a two-stage hierarchical method that used Monte-Carlo integration method for obtaining exact likelihood function and has been used successfully in analyzing complex population PK/PD data.
Objective: To develop a systematical approach to bridge the two-stage MCPEM algorithm and full Bayesian three-state hierarchical model in complex population PK/PD analysis.
Method: The analysis was based on PKPD data from 31 subjects who received troxacitabine/cisplatin combination where troxacitabine was administered as an intravenous infusion every 28 days at five different dose levels. The PD model was based on a drug-sensitive progenitor cell compartment, linked to the peripheral blood compartment, through three transition compartments representing the maturation chain in the bone marrow. The model included a feedback mechanism to capture the rebound phenomena. The troxacitabine affected the proliferation of sensitive progenitor cells through an inhibitory sigmoidal Emax model. A three-compartment linear PK model was used to describe the troxaicibatine concentration-time profile. First, the PKPD model was developed using MCPEM algorithm implemented in the SADAPT program and fit simultaneously to the data. The SADAPT program then automatically generated several different files (including dose/dosing time, observations, priors, initial parameters, and model template files) needed for the WinBugs program for full Bayesian analysis. The individual and population parameters estimated from MCPEM algorithm were served as initial values for the WinBugs program, and the PKPD data were analyzed simultaneously.
Results and Conclusions: Based on the initial information provided by the final parameters estimated from the MCPEM algorithm, the full Bayesian model developed using WinBugs program was stable and required less than ten thousands iterations to generate reasonable final parameter estimates. The proposed systematic bridging approach offers practical solution for using full Bayesian three-Stage hierarchical nonlinear mixed effect method in complex population PKPD analysis.
References:
[1]. Jonsson F, et al. J Biopharm Stat 2007;17(1):65-92
[2]. Kathman SJ, et al. Clin Pharmacol Ther 2007;81(1):88-94
[3]. Ng CM, et al. Pharm Res 2005;22(7):1088-1100
[4]. Friberg L, et al. Invest. New Drugs 2003;183-94
