Many studies in the pharmaceutical sciences attempt to assess the effect of factors such as treatment in clinical trials on a categorical outcome variable such as the amount of improvement. Often the outcome categories have a known ordering (e.g., none, some, moderate improvement). Under this general setting, recent developments with association models (Goodman, 1991) suggest that new maximum likelihood methods may offer substantial improvements in power and flexibility over traditional approaches. In Phase I, we propose to develop a prototype for a new PC-based computer software package for the analysis of a 2-way table (TREATMENT by OUTCOME) by association models, where OUTCOME is an ordered categorical variable. In Phase I we will also demonstrate the feasibility and value of applying association models to assess treatment effects, by re-analyzing several clinical trial data sets which have been analyzed by traditional methods. Assuming that the results of our Phase I investigations demonstrate that the use of association models are both feasible and useful, in Phase II we propose to extend the software package to more general association models to include covariates/stratifications such as center (for multi-center studies), baseline measures and demographics such as age, sex and other risk factors.
