10/30/2015

Post date: Nov 2, 2015 6:45:43 PM

Title: Instrumental variable methods in survival analysis

Speaker: Gary Chan, Associate Professor, Department of Biostatistics, University of Washington.

Abstract: Unmeasured confounding is often a serious limitation of scientific studies when perfect randomization of the exposure is not attainable. Examples include observational studies and randomized trials with noncompliance. The problem of unmeasured confounding cannot be generally solved without strong assumptions. Instrumental variables (IV) methods can identify certain exposure effects under various kinds of strong assumptions. Some general frameworks of instrumental variables have been well known for uncensored data. Two of them will be discussed: structural equation and principal stratification. We will survey the extensions to time-to-event data, and discuss the particular difficulties in extending the IV methods to survival analysis. Some solutions are offered.

(slides)