"Assessing Mediation Effects for Survival Types of Outcomes"
Bin Huang, PhD, Children's Hosp.Med.Ctr., Cincinnati
April 21, 2009 @ 3:30 pm - 4:30 pm
Location: BRB - 251
Biostatistics
TITLE: Assessing Mediation Effects for Survival Types of Outcomes
Assessing mediation effect helps elucidate
causal pathway from an exposure to an outcome. For Survival types of outcomes,
Lin et al (SIM, 1997) method is commonly used for assessing mediation effect. Their method defines the mediation effect following PE ratio based on two Cox
models, which subject to model misspecification. The model misspecification issue
was addressed by establishing conditions under which the method is valid.
Recently, Ghosh (SIM 2007) proposed jointly modeling two AFT models, and
estimating mediation effect semi-parametrically by a U statistics. It overcomes
model misspecification, but is computational intensive and lack of flexibility
to incorporate additional covariates. Here, the proposed new approach jointly
modeling two Cox models, is able to overcome model misspecification problem,
and is flexible and computationally feasible to incorporate additional
covariates. Our simulation results are comparable to Lin's under their
specified conditions, and show clear advantages in other settings. Case study will
be conducted using ACTG019 data, evaluating mediation effect of CD4 count to
the treatment effect of an HIV/AIDS clinical trial.
