Using Principal Stratification to Account for Noncompliance in Randomized Trials with Longitudinal Data
Date: Wednesday 28 November 2012
Time: 12:45 pm - 2:00 pm
Venue: LG6, LSHTM, Keppel Street, London, WC1E 7HT, UK
Type of event: Symposium
Speaker(s): Professor Michael Elliott, University of Michigan
Much research in the health sciences aims to understand the causal relationship between an intervention and an outcome, and a variety of statistical methods have been developed to answer these questions. However, in many situations, such causal relationships are not readily obtained even in randomized clinical trials due to problems with compliance: subjects randomized to the treatment who take it may differ in both observed and unobserved ways from those who do not take it or cannot tolerate it. Conditioning on variables such as compliance that are observed post-randomization destroys the causal interpretation of treatment effects in statistical models. One approach to accommodate post-randomization variables in regression while retaining the causal interpretation of the effect of treatment is principal stratification (Angrist et al. 1996; Frangakis and Rubin 2002), which defines principal strata as the joint distribution of ''potential'' adjustment variables under different treatment arms, and estimates the causal effect of treatment within these (now pre-randomization) principal strata. In this talk I will overview some recent methods that I and my colleagues have developed to extend the principal stratification compliance approach into settings where compliance behavior changes over time. In particular I will discuss a "superclass" approach in which trial subjects are sorted into latent classes that summarize compliance behavior over time (Lin et al. 2008, 2009), and a Markov Chain approach that allows previous outcome and compliance behavior to predict current outcome and compliance behavior (Gao and Elliott 2012). Illustrative examples will be provided to explore the different types of questions each approach tries to answer.
NOTE: External attendees please report to reception and you will be directed to the designated room.
Admission: Free and open to all with no ticket required. Entry is on a first come, first served basis.
Contact: Rhian Daniel