S1: Supplementary methods.

<p>S1.1 GISP data and data pre-processing. S1.2 Hyperparameter tuning. S1.3 Bootstrap algorithm to calculate confidence intervals for performance measures. S1.4 Model specificity and sensitivity. S1.5 Calculating the effectiveness of personalized treatment recommendations and their impact on the unnecessary use of CRO/CFX. <b>S2: Supplementary results.</b> S2.1 Permutation importance and feature selection. S2.2 Model performance across target years. S2.3 Percentage receiving effective treatment and unnecessary CRO/CFX for logistic regression and random forest. S2.4 Leave-one-out cross-validation for logistic regression and random forest</p> <p>(DOCX)</p>