I am doing a research about the predictive value of a bookmarked in the predicting outcome. The dependent variable is binary. The covariates can be incorporated after bivariate analysis, and only ones with certain P values e.g. Less than 0.1 be included in final model. The other way is to include all variables that are thought to interact with the bio marker and outcome, no matter their significance level in the bivariate analysis. Could someone suggest good references in dealing with such issue?