A Two-stage Regression Model for Epidemiological Studies with Multivariate Disease Classification Data

• Activity Number: 72
• Type: Topic Contributed
• Date/Time: Monday, August 9, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Nilanjan Chatterjee*+
• Companies: National Cancer Institute
• Keywords: molecular epidemiology ; disease classification ; semiparametric method

Abstract

Advances in clinical and molecular characterization of cancer provide new opportunities to study "etiologic" and "treatment" heterogeneity, i.e., to determine whether effects of exposures or treatments are different for different subtypes of a disease. Polytomous logistic regression, commonly used for analyzing heterogeneity between two or more distinct subtypes of a disease, gives odds ratio estimates for each disease subtype compared to a common group of nondiseased controls, but is not suitable for evaluating the effects of risk factors when disease groupings are defined by multiple overlapping tumor characteristics or markers. We propose a novel two-stage modeling approach for analyzing epidemiologic studies with data on multiple disease markers. The first-stage model involves defining polytomous logistic regression parameters for a set of markers, comparing all possible marker-defined disease subtypes to the common control group. The second-stage model further characterizes the exposure odds ratios for the first-stage disease subtypes in terms of the underlying disease markers in terms of a regression model.