General Latent and Mixed Models for Risk Factors in Cancer Resea

Fortaleza

Congreso; XII Escola de Modelos de Regressao; 2011

Associaccao Brasileira de Estadistica

This work proposes a new disease - model approach based on Generalized Linear Latent and Mixed Models (GLLAMM) to explore exposure disease associations from observational studies on cancer. GLLAMMs include many types of latent variables varying at different levels such as random effects, common factors and latent classes, even more latent variables can be regressed on other latent variables or other covariates. The latter scheme is used in the present work in order to inquire causal network in some of the most incident cancers in Argentina. Cancer is the most common cause of death worldwide; in Córdoba (Argentina),colorectal and bladder cancer are the thirth and fourth most incident cancer, respectively, in the overall population. This multistage disease is related with environmental conditions and dietary compounds comprise its causal network model, being some of these acting as a promoter or protective against the development of the disease. We proposed common factor models, as conrmatory factor analysis to dene new constructs that characterize the population habits, and after, a logistic regression with covariate measurement error to set up an exposure-disease model for cancer outcome. The formulation included information about life styles and habits, such as dietary pattern and physical activity as well as demographic and biological data coming from case-control studies for bladder (218/444 cases/controls) and colorectal (48/96 cases/controls) cancers in Córdoba. Our main result indicates that two new constructs, protective and promotion dimensions of the disease, can be dened from dietary habits to characterize the causal network of both cancers in Argentina