Figure 1.

Schematic overview of training and test prediction procedures. (A) Cohort is randomly divided into training and test datasets. (B) Initial feature selection is performed by determining which VOIs minimize Bayesian information criterion (BIC). (C) Fivefold cross‐validation is performed within the training dataset by randomly dividing the cohort into five sets, calculating the features that achieve highest prediction accuracy, and permuting this process 1,000 times to identify the most stable VOIs for prediction. (D) Stable VOIs are entered into a power analysis in training cohort to confirm that there is a sufficient sample for test prediction. (E) ROC curve to calculate prediction accuracy in training cohort. (F) Power analysis in independent test cohort. (G) ROC curve to evaluate prediction accuracy in independent test cohort.