"Censoring by death, principal stratification, and joint causal modeling of failure-time and longitudinal outcomes"
September 13, 2011 @ 3:30pm - 4:30pm
Marshall Joffe, M.D., Ph.D., University of Pennsylvania
Location: 701 Blockley
It is often of interest to estimate the effect of some treatment or exposure on a possibly repeated measured outcome. When some subjects die before the outcome can be observed, the measured outcome is not well-defined for those subjects and so it is difficult to define the causal effect of treatment on the measured outcome. The principal stratification approach, which has recently become popular, defines the effect of treatment on outcome for subjects who would live whether or not treated. This approach results in some bizarre paradoxes. We develop alternatives based on joint causal modeling of failure-time and measured outcomes and consider how to define the net effect of a treatment in this setting.