Innovative Strategies for Strengthening Interpretability

Analysis of covariance serves two main purposes in the analysis of clinical trials. It provides for a more powerful statistical test through variance reduction for comparison of treatment groups; and it addresses comparisons between randomized groups with adjustment for random imbalance of covariates. Certain methods of covariance analysis such as logistic regression for ordinal response data and proportional hazards (Cox) regression for survival data are not readily interpretable with regard to their properties of variance reduction and bias adjustment. Direct variance reduction is not observed when adjustment is made for various covariates. It is a commonly held misconception that adjusting for covariates with logistic or Cox models not only increases variance estimates, but also leads to unstable treatment parameter estimates. In addition, there may be concerns that technical assumptions regarding the structure of the log-linear model are not appropriate such as the proper form of the quantitative covariables, no interaction between treatment and covariates, and the proportional odds (or hazards) assumption. This research provides non-parametric counterparts to clarify the components of bias adjustment and variance reduction. There are no formal assumptions required for how a response variable is related to the covariables. The rationale for nonparametric covariance analysis is based on the randomization in the study design. Computations for these methods are through the application of weighted least squares to fit linear models to the differences between treatment groups for the means of the response variable and the