Assessment of improved prediction beyond traditional risk factors: when does a difference make a difference?