More on Polytomous Outcome Regressions (450 Patients)

Abstract

If clinical studies have categories as outcome like for example various levels of health or disease, then linear regression is not adequate for analysis. Linear regression assumes continuous outcomes, where with each level severity increases by the same quantity. Outcome categories are very common in medical research. Examples include age classes, income classes, education levels, drug dosages, diagnosis groups, disease severities, etc. Statistics has generally difficulty to assess categories, and traditional models require either binary or continuous variables. Outcome categories are sometimes called polytomous outcomes. Various analytical methods have already been reviewed. Multinomial regression (Chap. 28) is adequate, if none of the outcome categories are underpresented. Random intercepts models (Chap. 30) often provides better power than multinomial regression. Ordinal regression (Chap. 37) is adequate, if one or two categories are underpresented. Logit loglinear regression (Chap. 39) analyzes all kinds of first and second order interactions of the predictors on the outcome categories. Hierarchical loglinear regression can analyze higher order interactions of the predictors on the outcome categories (Chap. 39). In this chapter negative binomial and Poisson regressions will be reviewed. These methods may be rather suited for binary outcomes than for polytomous outcomes, but they can, even so, be used for data with more than two outcome categories. The data are, then, assessed, as multivariate models with multiple dummy outcome variables.