Frailty Probit model for multivariate and clustered interval-censored data
Interval-censored data commonly arise in many follow-up medical studies with repeated observations over time. The failure time of interest is not observed exactly but is known to fall within an interval formed by the two observational times that have changed status of the failure event. In this talk, we study semiparametric regression analysis of two types of interval-censored data with additional complications: multivariate interval-censored data where multiple failure times are of interest and clustered interval-censored data where only one failure time is of interest but subjects in the study are correlated due to sharing the same clustering characteristics. We propose a novel frailty Probit model and develop fully Bayesian methods for these two types of data.  The frailty Probit model enjoys several nice properties: (1) the marginal distribution of the failure time is a semiparametric Probit model, (2) the regression parameters has an interpretation of marginal covariate effects, and (3) the association between different types of failure times is characterized by several nonparametric association measures in closed form.  We adopt monotone splines to model the unknown non-decreasing functions and use a latent variable approach to facilitate the posterior computation.  The proposed Gibbs samplers are straightforward to implement because the full conditional posterior distributions are standard distributions for all unknowns. Simulation studies suggest good performance of the proposed methods. The method is illustrated by a real-life study of sexually transmitted infections among young women in Indiana State.

Bio: Lianming Wang obtained his Ph.D degree in Statistics from University of Missouri-Columbia in 2006. He worked as a postdoc at the National Institute of Environmental Health Sciences (NIEHS) for two years after my graduation. He then joined University of South Carolina in 2008.