Maria C. Kjellsson, Mats O. Karlsson
Division of Pharmacokinetics & Drug Therapy, Uppsala University, Sweden
Introduction & Objective Ordered categorical data are commonly used to describe subjectively scored symptoms and side effects and a majority of the observations are often at one extreme of the possible outcomes, i.e. the distribution of response is skewed. The standard approach for modelling ordered categorical data is with the proportional odds model. For repeated measures, a mixed effects modelling approach is applied by adding interindividual variability on baseline, enabling the -2 logarithm of the likelihood to be minimized. Minimizing the logarithm of the likelihood, using the Laplacian method may result in severely biased parameter estimates, related to the skewness of response distribution and magnitude of interindividual variability [1]. Because of this, we intend to investigate the bias in parameter estimates, when the proportional odds model is fitted to ordered categorical data using the Gaussian quadrature method available in the NLMIXED procedure in SAS.
Method This is a Monte Carlo simulation study where 100 original data sets were derived from a known model with fixed study design. The simulated response was a 4-category variable on the ordinal scale with categories 0, 1, 2 and 3. The model used for simulation was fitted to each data set for assessment of bias, once with the Laplacian method and once with the Gaussian quadrature method. In particular, we have focused on situations with non-even distribution of the response categories and the impact of interindividual variability.
Results & Conclusion The bias in the parameters estimated using the Gaussian quadrature method was markedly reduced, compared to the results from the estimations using the Laplacian method. Thus, the Gaussian quadrature performes well also in those situations where the Laplacian method does not.
References [1] S Jönsson, MO Karlsson (2002). Estimating Bias in Parameters for Some NONMEM Models for Ordered Categorical Data. AAPS Pharm Sci; 4: abstract W4228
Reference: PAGE 13 (2004) Abstr 529 [www.page-meeting.org/?abstract=529]
Poster: poster