TABLE 4. Suggestions for Comparison of Models for Risk Prediction

1. To compare global model fit, use a measure based on the log likelihood function, such as the Bayes Information Criterion, in which lower values indicate better fit and a penalty is paid if the number of variables is increased.
2. Compare general indices of calibration (such as the Hosmer-Lemeshow statistic, which compares the observed and predicted risk within categories) and discrimination (such as the c-statistic).
3. If global fit is better for 1 model but general calibration and discrimination are similar, fit may be better among some individuals (for instance, those at higher risk). Determine how many individuals would be reclassified in clinical risk categories and whether the new risk category is more accurate for those reclassified.
4. For clinical use of a new invasive or expensive biomarker, determine if a higher or lower estimated risk would change treatment decisions for the individual patient.