Table 5.
Evaluating performance of risk prediction modelsa
| Parameter | Number of studies (%) |
|---|---|
| Validation | |
| Apparent | 30 (77%) |
| Internal | 15 (38%) |
| Bootstrapping | 2 (5%) |
| Jack-knifing | 1 (3%) |
| Random split sample | 10 (26%) |
| Cross-validation | 2 (5%) |
| Temporal | 3 (8%) |
| External | 21 (54%) |
| Performance metricsb | |
| Discrimination | |
| C-statistic | 39 (100%) |
| D-statistic | 1 (3%) |
| Calibrationc | 10 (26%) |
| Hosmer-Lemeshow statistic | 8 (21%) |
| Calibration plot | 2 (5%) |
| Classification | |
| Reclassification (NRI) | 2 (5%) |
| Other (for example, sensitivity, specificity) | 31 (79%) |
aNRI,- Net Reclassification Index; bstudies can report more than one performance metric; ccalibration assessed on the basis of the development cohort in 10 studies and in the validation cohorts in 2 studies.