Table 4.
Summary of the prediction performance during the training phase (CV = 10) with the Discovery cohort (N = 131), and during the replication phase of the best model with both Replication Cohort 1 (N = 113) and Replication Cohort 2 (N = 113)
| Sample | Type | Model | Accuracy | Sensitivity | Specificity | PPV | AUC | LR+ | MCC |
|---|---|---|---|---|---|---|---|---|---|
| Discovery cohort | Training | SVM | 0.63 | 0.19 | 0.89 | 0.50 | 0.36 | 1.73 | 0.11 |
| Training | RF | 0.65 | 0.25 | 0.88 | 0.55 | 0.65 | 2.08 | 0.17 | |
| Training | NB | 0.66 | 0.31 | 0.86 | 0.56 | 0.64 | 2.16 | 0.20 | |
| Replication Cohort 1 | Replication | NB | 0.63 | 0.39 | 0.81 | 0.61 | 0.64 | 2.07 | 0.22 |
| Replication Cohort 2 | Replication | NB | 0.64 | 0.38 | 0.79 | 0.50 | 0.58 | 1.75 | 0.17 |
SVM support vector machine, RF random forest, NB naive bayes, PPV positive predictive value, AUC area under de curve, LR+ positive likelihood ratio test, MCC The Matthews correlation coefficient