Dealing with Multiple Confounders in Observational Studies: Beyond Logistic Regression