Table 2. Performance of the proposed predictive model compared to other models.
| Model | 5-Year C-index | 5-Year Integrated Brier Score |
|---|---|---|
| LYFT Reported [4] | 0.680 | Not Reported |
| EPTS Reported [2] | 0.69 | Not Reported |
| EPTS Using the Same Cross-Validation Data as the Proposed Model | 0.697 | Not Calculated |
| Li et al. [6] Reported | 0.700 | Not Reported |
| Cox Model for Both Cohorts* | 0.706 | 0.063 |
| Random Forests for Both Cohorts* | 0.718 | 0.062 |
| Proposed Model without Donor Variables** | 0.717 | 0.060 |
| Proposed Model | 0.724 | 0.061 |
| Proposed Model using PMM Imputation | 0.724 | 0.060 |
Performance from 10 random samples of 80,000 training observations and 20,000 out-of-sample observations for all models except those marked ‘Reported’, where the metrics shown were provided in the literature for their respective models.
*These two predictive models used the variable selection techniques we used for each cohort separately but instead applied to all the data.
**Donor variables that were removed were: AGE_DON, COD_CAD_DON, COLD_ISCH_KI, DEATH_MECH_DON, HIST_DIABETES_DON, and HIST_HYPERTENS_DON.