PURPOSE: Predictive modeling is a common analytic method in observational research. Validation of the final model is important for drawing conclusions regarding reproducibility. The object of this session is to discuss and compare different validation methods.

DESCRIPTION: Predictive models focus on identifying significant predictors of outcomes. After a predictive model is built, it needs to be validated. The validation is usually done by split-sample method, cross-validation or bootstrapping. However, each method has pros and cons, and its performance depends on factors including sample size, number of variables, and the nature of the study. In this discussion, we will compare these model validation methods, with application to real world study data.