Table 3.
Ensemble discrimination performance over training and testing data.
| Covariates used in model | Overall survival (OS) | Recurrence free survival (RFS) | ||||||
|---|---|---|---|---|---|---|---|---|
| C-index | AUC | C-index | AUC | |||||
| Train | Test | Train | Test | Train | Test | Train | Test | |
| Clinical | .66 | .62 | .66 | .62 | .63 | .64 | .70 | .66 |
| Clinical + rsf (top 3) | .72 | .75 | .75 | .80 | .65 | .73 | .72 | .77 |
| Clinical + rsf (top 5) | .73 | .71 | .76 | .75 | .65 | .75 | .71 | .80 |
| Clinical + rsf (top 10) | .76 | .71 | .80 | .76 | .66 | .75 | .72 | .80 |
| Clinical + cox | .73 | .67 | .76 | .70 | .67 | .73 | .75 | .79 |
| Clinical + 2 Clusters | .81 | .75 | .85 | .79 | .73 | .66 | .82 | .70 |
| Clinical + 3 Clusters | .81 | .72 | .86 | .75 | .79 | .72 | .88 | .77 |
| Clinical + 4 Clusters | .87 | .74 | .92 | .75 | .91 | .75 | .95 | .80 |
Comparison of ensemble performance over Train and Test data using C-Index and AUC for both OS and RFS outcomes. Each row in the table corresponds to the ensemble model using different covariates. The Clinical baseline is the model where only clinical covariates are included. The subsequent rows include additional covariates into the baseline model: top n selected radiomic features using rsf (+ rsf (top n)), selected radiomic features using coxnet (+ cox), and the proposed cluster labels (Clinical + N Clusters). The best test results are highlighted in bold. The best test results for OS are obtained by the Clinical + rfs (top 3) (C-Index: .75, AUC: .80) while the best test results for RFS are obtained by Clinical + 4 Clusters (C-Index: .75, AUC: .80).