Multivariate Survival Model Using Latent Variables with the Department of Veterans Affairs Diabetes Trial (VADT) Data

• Activity Number: 559
• Type: Topic Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Government Statistics
• Author(s): Gideon D. Bahn*+
• Companies: Hines VA Hospital
• Keywords: Multivariate ; Survival Model ; Latent Variables ; Veterans Affairs Diabetes Trial (VADT) ; Structural Equation Modeling

Abstract

Framework: Conventionally, the Cox survival model has one outcome as a single and non-recurring discrete variable, though often several related events are made into one composite outcome. This model has limitation; losing important measured outcomes by not counting recurring events within the same person. This problem has been dealt with in frailty models. This study, however, shows flexibility of a survival model using a latent variable (SML), measuring recurring events. The frailty models are still limited to analyzing either multiple but different events or the reoccurrence of the same event. This study utilizes the continuous time survival model in a Structural Equation Modeling (SEM) framework. First, Cox survival model and a SML are compared. Second, frailty models and SML with recurring events are compared. Then, a multivariate survival model using multiple latent variables (MSML) is investigated using SEM. The results show the similarity between the first two comparisons, but the MSML provides additional information that is different from the first two. Moreover, the MSML may be explored compared with multi-stage and/or multivariate frailty models. The Department of Vete