Fig. 2.

The feature selection process of the RFE. Each iteration removes a feature that is considered least important and corresponds to a 3-repeated 5-fold cross-validation. After cross-validation, the average AUC of the model in the training cohort was used to determine the optimal number of features. The number of candidate features was chosen to range from 1 to 15. The feature number with maximal AUC was selected. a Two features were selected in the TBR model and b six features were selected in the TTP model. RFE, recursive feature elimination; AUC, area under the receiver operating characteristic curve