Table 3.
Performance of progression prediction models
| C-index (95% CI) | F score (95% CI) | Sensitivity (95% CI) | Specificity (95% CI) | Binomial p value* | Log-rank p value† | χ2 (95% CI)† | |
|---|---|---|---|---|---|---|---|
| Internal test set | |||||||
| Image-based model | 0·737 (0·713–0·773) | 0·790 (0·776–0·808) | 0·696 (0·664–0·718) | 0·775 (0·769–0·782) | <0·0001 | <0·0001 | 17·33 (13·73–22·02) |
| Clinical-data-based model | 0·769 (0·755–0·786) | 0·811 (0·803–0·836) | 0·656 (0·631–0·674) | 0·811 (0·801–0·817) | <0·0001 | <0·0001 | 31·77 (24·58–36·56) |
| Image and clinical data combined model | 0·805 (0·800–0·820) | 0·843 (0·836–0·863) | 0·720 (0·700–0·749) | 0·845 (0·840–0·850) | ref | <0·0001 | 26·51 (21·65–33·56) |
| Severity-score–based model | 0·696 (0·676–0·711) | 0·761 (0·752–0·775) | 0·656 (0·635–0·669) | 0·743 (0·736–0·752) | <0·0001 | <0·0001 | 18·15 (9·45–23·70) |
| Severity score and clinical data combined model | 0·781 (0·755–0·787) | 0·805 (0·798–0·832) | 0·678 (0·666–0·700) | 0·798 (0·793–0·807) | 0·0002 | <0·0001 | 42·23 (33·63–49·59) |
| External test set | |||||||
| Image-based model | 0·721 (0·700–0·727) | 0·795 (0·779–0·813) | 0·633 (0·606–0·662) | 0·791 (0·788–0·796) | <0·0001 | <0·0001 | 39·17 (28·62–48·58) |
| Clinical-data-based model | 0·707 (0·695–0·729) | 0·769 (0·756–0·780) | 0·602 (0·583–0·621) | 0·753 (0·751–0·762) | <0·0001 | <0·0001 | 31·72 (26·41–42·94) |
| Image and clinical data combined model | 0·752 (0·739–0·764) | 0·805 (0·791–0·825) | 0·667 (0·643–0·698) | 0·798 (0·791–0·803) | ref | <0·0001 | 52·04 (46·50–66·14) |
| Severity-score–based model | 0·606 (0·584–0·627) | 0·720 (0·704–0·733) | 0·528 (0·512–0·541) | 0·695 (0·686–0·701) | <0·0001 | <0·0001 | 11·65 (6·84–15·43) |
| Severity score and clinical data combined model | 0·715 (0·704–0·721) | 0·778 (0·757–0·795 | 0·667 (0·649–0·677) | 0·759 (0·756–0·765) | <0·0001 | <0·0001 | 37·62 (26·68–46·95) |
C-index for right-censored data measures the model performance by comparing the progression information (critical labels and progression days) with predicted risk scores; a larger C-index correlates with better progression prediction performance. C-index=concordance index.
*
Measures the difference in performance between the image and clinical data combined model and other prediction models; a smaller p value represents greater likelihood of a difference between the combined model and other models.
†
Shows a comparison of stratification performance of different models; a smaller p value and larger χ2 correlate with better risk stratification performance.