Application of Poisson Spatial Regression Models in Estimating Time-Varying and Covariate-Dependent Coefficient Functions in Survival Analysis

• Activity Number: 412
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Jianghua He*+
• Companies: University of Kansas Medical Center
• Keywords: Survival Analysis ; Time-varying coefficients ; Covariate-dependent coefficients ; Spatial Poisson models

Abstract

In survival analysis, both covariate-dependent coefficients and time-varying coefficients have been widely discussed. However, these two issues were rarely dealt with simultaneously. In reality, it is possible that a coefficient not only depends on the corresponding covariate but also depends on the follow-up time. The simple approach of modeling such coefficients using three-way interactions is far too restricted. Some software packages can estimate covariate-dependent or time-varying coefficients, but their potentials of handling complex coefficients with both features are not fully demonstrated. In this study, the author proposed to use the spatial Poison regression models to demonstrate how covariate-dependent coefficients change with follow-up time in survival analysis, which can be done using any software package that can handle generalized linear spatial models. This proposed approach is used to model the association between body mass index (BMI) and all-cause mortality, a controversial issue widely discussed in the literature. The results of using this innovative approach provide a new prospective on the issue and potentially explain some of the controversies.