Aim 2: Classification and Regression Tree Analysis

For the second aim, classification and regression tree (CART) modeling was used to identify an optimal set of rules for classifying participants by diagnosis based only on itemand subscale-level data from the neurobehavioral battery. Again, R package 'tidymodels' was used in all steps of the CART analysis. First, training and test datasets were generated from the data. Then, the training dataset was used to generate a set of 10-fold crossvalidation samples for model hyperparameter tuning. The best hyperparameters were selected based on the AUC. The CART model was first fit on the training dataset, then on the test dataset to assess performance.

Aim 3: Receiver Operating Characteristic Curves and Comparisons

For the third aim, to compare the relative utility of the PCA and CART models, the AUC of both models for classifying diagnosis were compared using the bootstrap test for comparing ROC curves (R routine 'roc.test' from the package 'pROC'). Sensitivity, specificity, and positive and negative predictive values were computed for the PCA and CART models.
