Table 1

Model discrimination*

ModelAUC (95% CI)Sensitivity†Specificity†PPVNPV
Regression models
 Stepwise selection (LR-S)0.74 (0.68 to 0.82)0.640.790.410.91
 Penalised (LASSO)0.77 (0.71 to 0.83)0.670.780.410.91
Neural networks (NN)
 One hidden layer (NN-1)0.74 (0.67 to 0.80)0.650.750.390.90
 Two hidden layers (NN-2)0.71 (0.64 to 0.79)0.610.780.370.90
Tree-based
 Random forest (RF)0.77 (0.71 to 0.83)0.750.710.370.93
 Gradient boosting (GB)0.73 (0.66 to 0.79)0.690.680.330.91
Support vector machine (SVM)
 SVM-RBF0.77 (0.70 to 0.84)0.750.740.390.93
Ensemble
 SuperLearner (SL)0.78 (0.72 to 0.84)0.710.770.410.92
  • *Average across five independent, 10-fold cross-validations.

  • †At an optimal cut-point found for each algorithm and iteration.

  • AUC, area under the curve; LASSO, least absolute shrinkage and selection operator; LR-S, logistic regression wih stepwise selection; NPV, negative predictive value; PPV, positive predictive value.