I’m going to perform a 1-year prospective study to identify predictors performing logistic regression (as outcome treated: Yes or No). First, I was thinking to perform a stepwise procedure (e.g. forward and backward), even though the procedures have been criticized because they can be influenced by random variation in the data. Another way is to test the significance of adding one or more variables in the model using the partial F-test. However, in order for assessing the best model, which test would be the best: R-squared or Cross-validation? Noted that I’m not going to consider any theory, perhaps a model would be used that will not be obvious in the study.

Dear friends, according to the facts I would like to ask for your suggestions what steps would be the best to build up a predictive model and its assessment!!!