2007 - KÝbenhavn - Denmark

PAGE 2007: Methodology- Other topics
Chee Meng Ng

A Comparison of Estimation Methods in Nonlinear Mixed-effect Model for Population Pharmacokinetic-pharmacodynamic Analysis

RJ Bauer (1), S Guzy (1), CM Ng (2)

(1) XOMA (US) LLC, Berkeley, CA, (2) Institute for Drug Development, Cancer Therapy and Research Center, San Antonio, TX

Background: NONMEM is the current reference software for population pharmacokinetic/pharmacodynamic (PK/PD) analysis.  In the last ten years, a series of new tools for population PK/PD modeling have become available.  These included methods based on exact likelihood functions and three-stage Bayesian method.  However, the comparison between these methods that have been described in the literature to date involved only simple PK model.  

Objective: To provide an overview of the statistical basis of the selected estimation methods and assess the utility of these methods in various PK/PD modeling problems.

Method: Four stimulated dataset with various PK/PD problems were used to assess the performance of different estimation methods (FO and FOCE method in NONMEM, MCPEM method in SADAPT and PDX-MCPEM, SAEM method in Monolix, and three-stage MCMC method in WinBugs) in population data analysis.  First dataset is a two-compartment PK model with sparse data.  A second problem is the same two-compartment PK model with additional covariate model for CL and V1.  The third dataset was a single IV bolus two compartment PK model with direct-link Emax PD model requiring total of 8 parameters to describe the model.   Analytical functions were used to describe the model.  The last problem is a one-compartment PK model with both linear and saturated elimination, and indirect response PD model that required total of 2 differential equations and five parameters.  A total of 25 subjects with intensive PK/PD sampling were included in the analysis and PKPD data were analyzed simultaneously.   Results will be analyzed by comparing the deviation of the estimated parameters from the reference values. A crude comparison of CPU times is also provided.

Results/Conclusion: The NONMEM FO method is accurate and fast in analyzing simple PK data when the intra-and inter-individual variability is small.  The NONMEM FOCE method is slower but more accurate than FO.  However, parameters estimated using FOCE method can be biased if the data are very sparse.  The exact EM method (MCPEM and SAEM) has greater stability in analyzing complex PKPD model and provided accurate results with both sparse and rich data.  MCPEM method converge more slowly than NONMEM FOCE for simple PK model, but achieve faster convergence and has greater stability than NONMEM FOCE for complex PKPD model.  WinBugs provided accurate assessments of the population parameters for all examples. 

[1].  Bauer J, et al.  AAPS J 2007;9(1):E60-83

Reference: PAGE 16 (2007) Abstr 1096 [www.page-meeting.org/?abstract=1096]
Poster: Methodology- Other topics
Click to open PDF poster/presentation (click to open)