TABLE 1.
The c-index values, mean and standard deviation (SD) (range from 0 to 100) from 200 replications for the DNN, RSF, LASSO, and True models with sparse signals under five scenarios: linear effects (scenario 1) and linear effects together with nonlinear effects (scenario 2) or with interactions (scenario 3) or with nonlinear and interaction effects (scenario 4) or with interaction and indicator effects (scenario 5)
| p | DNN | RSF | LASSO | True | |
|---|---|---|---|---|---|
| Scenario 1 | 10 | 88.0 (0.7) | 82.9 (1.0) | 88.2 (0.6) | 87.4 (0.6) |
| 50 | 85.7 (1.0) | 82.8 (1.0) | 88.2 (0.6) | ||
| 100 | 83.2 (1.0) | 82.4 (1.2) | 88.2 (0.6) | ||
| 500 | 82.2 (1.0) | 81.1 (1.1) | 88.0 (0.7) | ||
| Scenario 2 | 10 | 88.7 (0.9) | 80.9 (1.2) | 80.0 (1.0) | 89.8 (0.5) |
| 50 | 84.2 (1.6) | 80.2 (1.1) | 80.0 (1.0) | ||
| 100 | 80.6 (2.0) | 79.5 (1.1) | 79.9 (0.9) | ||
| 500 | 74.3 (3.1) | 77.9 (1.0) | 79.9 (1.0) | ||
| Scenario 3 | 10 | 93.1 (0.6) | 79.7 (1.8) | 74.0 (1.4) | 94.0 (0.4) |
| 50 | 91.4 (0.7) | 75.6 (1.5) | 73.9 (1.5) | ||
| 100 | 89.8 (0.8) | 74.4 (1.6) | 73.9 (1.4) | ||
| 500 | 81.6 (1.8) | 72.0 (1.5) | 73.7 (1.4) | ||
| Scenario 4 | 10 | 92.1 (0.8) | 80.1 (1.8) | 71.4 (1.3) | 94.4 (0.4) |
| 50 | 88.9 (1.5) | 75.6 (1.5) | 71.3 (1.4) | ||
| 100 | 84.5 (2.0) | 74.2 (1.6) | 71.4 (1.3) | ||
| 500 | 76.3 (1.8) | 71.4 (1.4) | 71.1 (1.4) | ||
| Scenario 5 | 10 | 92.4 (0.6) | 79.4 (1.7) | 73.3 (1.3) | 94.0 (0.4) |
| 50 | 90.4 (0.8) | 75.2 (1.6) | 73.1 (1.4) | ||
| 100 | 88.6 (0.8) | 74.0 (1.4) | 73.0 (1.4) | ||
| 500 | 80.4 (2.0) | 71.4 (1.5) | 72.7 (1.3) |
Note: The number of predictors is set at p = 10, 50, 100, 500.
Abbreviations: DNN, deep neural network; RSF, random survival forest.