Table 4.
Estimates of discrimination, optimism and shrinkage for fitted models.
| Modela | Optimismb | Shrinkageb | Complete caseAUC (95% CI) | Complete case adjusted foroptimism AUC (95% CI) | Multiple imputationAUC (95% CI) | Multiple imputation adjustedfor optimism AUC (95% CI) |
|---|---|---|---|---|---|---|
| Model 1 | 0.012 | 0.93 | 0.871 (0.822, 0.919) | 0.859 (0.810, 0.907) | 0.845 (0.799, 0.891) | 0.833 (0.787, 0.879) |
| with TUBB6 | 0.015 | 0.91 | 0.875 (0.826, 0.923) | 0.860 (0.811, 0.908) | 0.848 (0.802, 0.894) | 0.833 (0.787, 0.879) |
| Model 2 | 0.012 | 0.91 | 0.930 (0.892, 0.967) | 0.918 (0.880, 0.955) | 0.892 (0.849, 0.934) | 0.880 (0.837, 0.922) |
| with TUBB6 | 0.015 | 0.88 | 0.932 (0.894, 0.970) | 0.917 (0.879, 0.955) | 0.894 (0.852, 0.937) | 0.879 (0.837, 0.922) |
| Model 3 | 0.021 | 0.90 | 0.682 (0.614, 0.750) | 0.661 (0.593, 0.729) | 0.696 (0.640, 0.751) | 0.675 (0.619, 0.730) |
a
Model 1 compared neoplasia with control; Model 2 compared dysplasia with control; Model 3 compared matched non-neoplastic with control.
b
Optimism and shrinkage were estimated from internal validation using bootstrap sampling.