Table 4.
Internal 10-fold cross-validation of MI models evaluated with precision, recall and AUC.
| MI | ||||
|---|---|---|---|---|
| Model | Precision (95%CI) | Sensitivity (95%CI) | Specificity (95%CI) | AUC (95%CI) |
| SCORE model22 | 0.716 (0.664–0.741) | 0.725 (0.691–0.767) | 0.691 (0.652–0.731) | 0.719 (0.681–0.737) |
| Random Forest including PRSCAD | 0.735 (0.708–0.782) | 0.756 (0.726–0.801) | 0.739 (0.702–0.777) | 0.741 (0.725–0.775)a |
| Random Forest including both eye-specific Df | 0.756 (0.732– 0.802) | 0.778 (0.762–0.831) | 0.758 (0.721–0.795) | 0.763 (0.750–0.802)a |
| Random Forest including mean Df and PRSCAD | 0.733 (0.716–0.770) | 0.779 (0.743–0.814) | 0.756 (0.717–0.797) | 0.748 (0.722– 0.773)a |
| Random Forest including Df and PRSCAD | 0.770 (0.734–0.805) | 0.790 (0.757–0.826) | 0.764 (0.728–0.800) | 0.770 (0.751–0.802)a |
aAUC estimates significantly different (Wilcoxson signed-rank test P-value < 0.005) from the ones obtained with the SCORE model. The obtained Wilcoxon signed-rank P-value for each model comparison is included in Supplementary Table 7.