Uncontrolled confounding in studies of screening effectiveness: an example of colonoscopy