Eun-Kyung Lee
Department of Statistics, Ewha Womans University
Objectives: In these days, NONMEM provides new estimation methods to estimate population PK/PD model parameters. These new methods – IMP, SAEM, Bayes, etc – use MCMC techniques to get the final parameter estimates. However MCMC methods use stochastic simulation, it depends on the starting point, burn-in period and stationary properties of chains. Therefore it should be carefully examined before we use the final parameter estimates. In this study, we suggest new method to check the validity of these MCMC method and the final parameters.
Methods: We present various methods to validate estimates of PK/PD model with MCMC techniques. We also propose new method for validation and compare them. Our new method modified the Bayesian model check approach. Also we develop a tool for MCMC validation methods with its own GUI. It is written in R.
Results: Various validation methods are examined. None of them is the outstanding performance for validation. Therefore user needs to use several validation methods at the same time. This developed tool is very helpful to compare several validation methods.
Conclusions: A tool for MCMC validation methods will be very helpful for pharmacometricians to compare various estimation methods, check their models and decide the final model.
References:
[1] Markov Chain Monte Carlo in Practice – Gilks and Spiegelhalter
[2] Monte Carlo Simulation Methods – Robert and Casella
Reference: PAGE 21 (2012) Abstr 2398 [www.page-meeting.org/?abstract=2398]
Poster: Model evaluation