Is there some trustworthy resource on effect of interactions if not accounted for. Especially the situation is often, let's assume heart surgery, age, gender and logistic euroscore(derived from age and gender among others) fitted into either logistic regression predicting Mortality (yes/no) or cox-regression for survival time, what is the situation here? What happens if interaction terms have not correctly been accounted for and how would you account for?
Edit: I'd also like to expand the question, is Stepwise Regression able to detect accurately for interactions, apart from it's known flaws? If stepwise including interactions is valuable, does it imply to even reduce dependent variables to input into the model to keep some reliability as much as regular dependent variables do?