Combined Benefit of Prediction and Treatment: A Criterion for Evaluating Clinical Prediction Models