2017-04-14

You Wu, PhD Candidate, Department of Bioinformatics and Biostatistics, University of Louisville

"Examine treatment effect for ordinal outcome when confounding covariates exist"

Ordinal outcomes are frequently observed in the clinical studies, and social and economic sciences. In the randomized control trials, the parametric statistical methods (e.g., ordinal logistic regression model and cumulative probit model) and the non-parametric statistics (e.g., the Mann-Whitney U test statistics) are often used to examine the treatment effect for ordinal outcome. However, in the observational studies, the confounding variables need to be controlled or adjusted, otherwise the estimate of the treatment effect may be seriously biased. Propensity score based methods, such as matching, stratification, and inverse probability weighting method, have been applied to assess treatment effect with control of confounding covariates. However, these methods usually assign the ordinal outcome variables to numerical scores in the analysis, thus leading to lose information about the nature of ordinal variables. In this project, we develop an adjusted U statistics to control the confounding covariates by using the inverse probability weighting, which weights each subject into its study population, and the weight for a subject is obtained from the propensity score. We compared the adjusted U-statistics method with other methods such as ordinal logistic regression, unadjusted Mann-Whitney U statistics, stratification, matching, and adjusted Kolmogorov–Smirnov test. Simulation studies are performed to compare the performance of proposed adjusted U statistics and other methods. A case study is constructed to assess the effect of physical activity on diabetic status with control the confounding from other sociodemographic characters and dietary information.

Stay connected TwitterFacebookLinkedInYouTubeInstagram