Masters candidate in Biostatistics, Qiuyu Yang, will present:
“Comparison of Propensity Score Methods for Observational Studies to Conventional Regression Analysis”
Plan B Adviser: Chap Le
Abstract: In observational study, how to control confounding variables is a big challenge when estimating the treatment effect on the outcome. An increasingly popular method to address this problem is propensity score. It is the conditional probability of a subject being assigned to treatment group, given a vector of observed baseline covariates, which can be calculated by logistic regression. There are three approaches to apply propensity score, one is forming matched sets of treated and untreated subjects, one is stratifying subjects into several subsets based on propensity score, and last one is using propensity score directly in the regression model as a covariate. An observational study on throat cancer is presented to compare propensity score and regression analysis. The estimation and confidence interval of these methods are similar that there is no significant difference in Quality of life score between surgery group and radiation/chemotherapy group. All methods efficiently eliminated the bias compared to the unadjusted model, among them, stratification is model-free and has similar subgroups at each level of propensity score, which may provide an alternative method in situation where randomized controlled study is impractical.
Refreshments will be served prior to the presentation.