نبذة مختصرة : Multivariate logistic regression is often used within the field of epidemiology to describe the relationship between disease occurrence and an exposure sus-pected to be associated with the disease. Additional effects are added to the model if they confound the disease-exposure relationship. Traditional model selection procedures, which focus on selecting models that are good predictors of the dependent variables, are not necessarily the most appropriate for epidemiological research questions. The proposed backwards-manual selection macro, %bms, attempts to select logistic regression models more suitable for epidemiological research. The macro consists of two main stages, (1) backwards selection of effect-modifiers, and (2) selection of main effects based on their confounding potential and influence on overall model-fit. During the first stage, the macro generates all first-order effect modifiers for the variables provided by the user, and PROC LOGISTIC’s backwards selection option is used to remove non-significant effect-modifiers. The second stage begins by removing the least significant potential confounder from the model. If this does not cause a change in the relationship between disease and exposure or the overall model fit, it remains out of the model. This process continues until all of the potential confounders not included in an effect-modifier have been evaluated.
Processing Request
No Comments.