Marginal Mixture Analysis of Correlated Bounded-Response Data with an Application to Ultrasound Risk Assessment

Document Type

Presentation Abstract

Presentation Date

2-28-2006

Abstract

Data with bounded responses are common in many areas of application. Often the data are bounded below by zero with excess zero observations. Essentially continuous responses may contain a substantial portion of zeros, either because no effects occur or due to limits of detection. Discrete data may exhibit a significantly higher percentage of zeros than expected under idealized models. In these settings ordinary generalized linear models fail. Three methods in the literature for modeling zero-inflated data are left-censored regression models, two-part models, and latent mixture models. We introduce a general class of zero-inflated mixture (ZIM) models that unifies and generalizes these three classes of models. In particular, we develop unified estimation procedures, large sample inferences and general computational algorithms. Novel diagnostics are proposed for assessing the adequacy of a ZIM model. We extend ZIM models to correlated data with excess zeros using the theory of generalized estimating equations. Risk threshold estimates are also provided for the incidence and magnitude of correlated adverse outcomes. We illustrate the issues and methodology in the context of an ultrasound safety study of the occurrence and extent of lung hemorrhage due to focused ultrasound exposure in laboratory animals.

Additional Details

Tuesday, 28 February 2006
4:10 p.m. in Math 109

This document is currently not available here.

Share

COinS