Table 1.
Comparison of the prediction performance of the optimal model of each algorithm.
| Model | AUC |
Youden Index |
Optimal threshold |
Sensitivity (%) |
Specificity (%) |
PPV (%) | NPV (%) |
Proportion of high-risk population (%) |
Brier score |
Homser- Lemeshow χ2 |
P-Value |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Lasso-AdaBoost | 0.798 (0.782, 0.813) |
0.472 | 0.11 | 73.09 | 74.10 | 23.5 | 96.2 | 30.4 | 0.078 (0.070, 0.086) | 13.81 | 0.09 |
| FLR-L1-LR | 0.817 (0.801, 0.832) |
0.524 | 0.11 | 73.49 | 78.86 | 27.4 | 96.5 | 26.7 | 0.076 (0.069, 0.084) | 11.51 | 0.17 |
| FLR-RF | 0.804 (0.788, 0.820) |
0.506 | 0.08 | 79.52 | 71.09 | 23.0 | 97.0 | 33.1 | 0.077 (0.070, 0.086) | 11.59 | 0.17 |
| FLR-SVM | 0.814 (0.798, 0.829) |
0.511 | 0.11 | 73.90 | 77.16 | 26.0 | 96.5 | 38.4 | 0.076 (0.069, 0.084) | 16.10 | 0.04 |
AUC, area under the receiver operating characteristic curve; PPV, positive predictive value; NPV, negative predictive value; Lasso-AdaBoost, AdaBoost with Lasso regression; FLR-L1-LR, L1 regularized Logistic regression with forward Partial Likelihood Estimation; FLR-RF, random forest with forward Partial Likelihood Estimation; FLR-SVM, support vector machine with forward Partial Likelihood Estimation.