"Dealing with Non-regularity in Optimal Dynamic Treatment Regimes"
February 2, 2010 @ 3:30 pm - 4:30 pm
Bibhas Chakraborty, Ph.D., Columbia University
Location: BRB 253
TITLE: Dealing with
Non-regularity in Optimal Dynamic Treatment Regimes
Dynamic treatment regimes are individually tailored treatments. They offer a way to operationalize the adaptive multistage decision making in clinical practice, thus providing an opportunity to improve such decision making. However, when using longitudinal data on patients to construct these treatment regimes, hypotheses concerning the choice of the optimal treatment at each stage may involve non-regular parameters. The non-regularity stems from the fact that parameters of interest are functions of maxima. As a result, the parameter estimates can be biased, and traditional methods of constructing confidence intervals can have poor frequentist properties. In this talk, I will present and evaluate a method that adapts to this non-regularity by the use of a thresholding (empirical Bayes) approach, and compare with other available approaches. Analysis of data from a two-stage smoking cessation trial will be presented as an illustration.
This is joint work with Susan Murphy and Victor Strecher.