A method that is widely used to predict the risk of a major coronary event may over- or underestimate risk for millions of Americans, according to a new study.