Graduate Student Seminar

Speaker 1: Yun-Ju Cheng

Title

A review of statistical methods for assessing the generalizability results from randomized trials to target population

Abstract

Randomized trial remains the most accepted design for estimating treatment effect. While properly planned and conducted randomized trials yield accurate estimate of the treatment effect for the trial samples, it remains susceptible to a lack of external validity. This presentation briefing recently proposed statistical methods to generalize inferences from a randomized clinical trial to a specified target population that are focusing on the aspect of generalizability that is related to differences in the characteristics of trial samples and target population:  estimating effects for a target population using inverse probability-of-selection weights, a weighting- based method that is similar to Horivitz-Thompson estimation; a propensity score approach that provides summary measures of representativeness with respect to observed pretreatment characteristics.

Speaker 2: Rabab Elnaiem

Title

Causal Methods to Deal with Non-compliance in Randomized Designs

Abstract

In medicine, the term non-compliance or non-adherence has been defined as the patient’s failure to comply with the recommendations of their physician or other healthcare provider. In the particular field of randomized trials, reasons for non-compliance could be side effects in an actively treated group, lack of efficacy in a placebo group, etc.

Inability to confront the non-compliance may include inadequate inference on the effect of the treatment under consideration. We will review and compare the method most used for analyzing data when non-compliance occurs, such as: Intention-to-Treat (ITT), Per-Protocol (PP) approaches and Instrumental Variable (IV).