Figure 1.

Prognostic performance of models for major adverse cardiovascular events prediction. In this high-dimension setting with 428 variables, tree-based machine learning models (random forests and gradient boosting) proved to yield significantly better results than linear models trained on the same feature space. Since better results were obtained on the training set during the nested cross-validation, in the following, we only report the metrics on the test set, as a fairer and more compact assessment of each model’s performance.