|
Abstract:
|
Complex diseases are often a result of interactions between environmental and genetic risk factors. Methods that are useful in finding these interactions include the Classification and Regression Tree (CART) method and logic regression (LR). CART is a nonparametric tree-based classification method that has the ability to identify interactions among predictor variables based on branches in the tree. However, CART tends to be biased towards selection of continuous predictors and thus may fail to capture interactions with binary variables. LR, an alternative decision tree method, uses logical combinations of binary predictors for classification. LR has greater flexibility for identifying interactions relative to CART, but it is not designed for inclusion of continuous variables. We present a new algorithm called C.Logic that allows for incorporation of binary and continuous covariates in a logic regression framework, thus combining the benefits of both CART and LR. Simulation studies show that C.Logic is superior to CART in identifying interactions between continuous and binary predictors for predicting a dichotomous disease outcome.
|
