Assessment of NONMEM and WinBUGS performances when estimating power and sigmoid Emax models
M. Neve (1), L. Iavarone (1), M. Petrone (1), R. Gomeni (1)
(1) GlaxoSmithKline, CPMS Clinical Pharmacology Modelling and Simulation, Verona, Italy
Objectives: Modeling of dose-response and dose-exposure relationship during Phase I studies is considered. In particular, the present work focuses on a couple of frequently used models: the sigmoid Emax model applied to efficacy (e.g. PET studies) or safety (e.g. vital signs) endpoints and the power model applied to dose escalation trials. Performances of NONMEM and WinBUGS have been assessed resorting to a number of simulated datasets exploring a variety of parameter conditions.
Methods: R software package was used to simulate 50 dose-response and 50 dose-exposure datasets for each set of fixed and random effects. The sigmoidicity factor for the Emax model was assumed to be 0.5, 1 or 2, whereas for the dose-exposure model the power coefficient was set to 0.8, 1 or 1.2. Low, intermediate and high variability around parameters was used when producing simulated data-sets. Moreover, the impact of weakly and more informative experimental designs (expressed in terms of sample size and number of doses) was evaluated. In both the simulation and estimation process, inter-subject variability was assumed to be log normally distributed and an additive residual error was used. Results obtained with NONMEM (ver. VI, FOCE interaction method) and WinBUGS (ver. 1.4.3) were then compared in terms of accuracy and precision of parameter estimates.
Results: Both NONMEM and WinBUGS showed a generally comparable accuracy in estimating fixed effects with either highly or less informative datasets. However, when considering random effects (inter-subject and residual error variability) NONMEM seemed to provide slightly more accurate estimates with a lower precision.
Conclusions: In the range of the explored experimental designs, neither NONMEM nor WinBUGS offered a significant added value when estimating fixed and random effects of power and sigmoid Emax models. Similar conclusions were drawn when evaluating both real and simulated datasets [1, 2]. Nevertheless, WinBUGS has the undoubted merit of providing posterior distribution of population parameters. This work does not aim to represent an exhaustive analysis of the performances that can be obtained resorting to these two different methodologies. As such, different experimental designs or more critical conditions may be considered for future evaluations.
References:
[1] Duffull S.B, Kirkpatrick C.M.J., Green B., Holford N.H.G. (2005). Analysis of population pharmacokinetic data using NONMEM and WinBUGS. J. of Biopharmaceutical Statistics 15, 53-73.
[2] Russu A., De Nicolao G., Poggesi I., Neve M., Iavarone L., Gomeni R. (PAGE 2008). Dose escalation studies: a comparison between NONMEM and a novel Bayesian tool.
