Efficient Estimation of the Risk of a Disease by Quartile-Categories Using Generalized Additive Models

• Activity Number: 59
• Type: Contributed
• Date/Time: Sunday, August 3, 2003 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Craig B. Borkowf*+ and Paul S. Albert
• Companies: Centers for Disease Control and Prevention and National Cancer Institute
• Keywords: cancer ; estimation ; odds ratio ; epidemiology ; generalized additive model ; quantile-category

Abstract

Suppose that one wishes to make inference to the risk of a disease by the population quartile-categories of a particular risk factor. In the standard approach, one categorizes the continuous risk predictor variable by its empirical quartiles. One then constructs a generalized linear model (GLM), perhaps with the logistic link, to relate the probability of disease to a linear combination of known functions of predictor variables, including the empirical quartile-categories of the chosen risk factor. Alternatively, one may construct a GAM, which nonparametrically estimates the functional form of the contribution of the risk factor to the GLM. Unlike the standard approach, this method employs information about the actual values or empirical percentiles of the risk factor, which makes the estimation process more efficient. One can then integrate under the estimated curve to estimate the risk of disease by quartile-category and, in turn, one can compute odds ratios and other desired statistics. Graphical methods enable one to make inferences about the true relationship between the disease and the risk factor. An example from nutritional epidemiology is presented.