Table 2.
Results of overall survival prediction (concordance index) and clustering (pairwise log-rank score) with 95% confidence intervals in the validation cohort
| Method | Clustering | Concordance index (↑) | Pairwise log-rank score (↑) |
|---|---|---|---|
| CPH | ✖ | 0.651 (0.650, 0.652) | - |
| AFT | ✖ | 0.647 (0.645, 0.649] | - |
| SSVM | ✖ | 0.651 (0.648, 0.654) | - |
| GBDT | ✖ | 0.652 (0.650, 0.655] | - |
| K-Means | ✓ | 0.584 (0.561, 0.593) | 41.97 (39.27, 44.33) |
| HAC | ✓ | 0.572 (0.568, 0.575) | 30.01 (25.07, 36.27) |
| NSC | ✓ | 0.650 (0.647, 0.652) | 56.23 (50.36, 62.77) |
| GEMS | ✓ | 0.665 (0.662, 0.667) | 69.17 (58.98, 76.98) |
The bold type indicates the best performance.
AFT accelerated failure time, CPH Cox proportional hazard, SSVM survival support vector machine, GBDT gradient-boosted decision trees, HAC Hierarchical Agglomerative Clustering, NSC Neural Survival Clustering, GEMS Graph-Encoded Mixture Survival.