Performance Measures for Prediction Models and Markers: Evaluation of Predictions and Classifications

Prediction models are becoming more and more important in medicine and cardiology. Nowadays, specific interest focuses on ways in which models can be improved using new prognostic markers. We aim to describe the similarities and differences between performance measures for prediction models. We analyzed data from 3264 subjects to predict 10-year risk of coronary heart disease according to age, systolic blood pressure, diabetes, and smoking. We specifically study the incremental value of adding high-density lipoprotein cholesterol to this model. We emphasize that we need to separate the evaluation of predictions, where traditional performance measures such as the area under the receiver operating characteristic curve and calibration are useful. from the evaluation of classifications, where various other statistics are now available, including the net reclassification index and net benefit.