Table 5 Performance of the multimodal LightGBM model stratified by patient historical conditions.

From: Multimodal machine learning for 5-year mortality prediction after percutaneous coronary intervention

Subgroup

N

Mortality (%)

AUC-ROC

Precision

Recall

F1-score

No complex history

8409

12.17

0.738

0.397

0.364

0.380

High-risk subgroups

 

 History of PCI

1,581

15.62

0.750

0.404

0.393

0.398

 History of CABG

152

28.95

0.710

0.654

0.386

0.486

 History of Stroke

733

23.74

0.707

0.504

0.391

0.440

  1. The No Complex History group serves as a baseline, demonstrating robust overall performance. Notably, high-risk subgroups show improved precision for mortality prediction, indicating the model’s clinical utility in complex cases.
Back to article page