This systematic review and proof-of-concept case study examined how artificial intelligence and machine-learning methods are applied to cardiac biomarkers in patients with myocardial infarction. Researchers identified 120 eligible studies and analyzed a dataset of 152 patients. The analysis focused on prediction and prognostic modelling using multimodal inputs that combine biomarkers with clinical or functional variables.
Most studies used logistic or regularized regression, with Random Forest also being common. The FULL variant of the model achieved near-perfect discrimination in the test data, while the BIOMARKERS variant showed strong performance. However, independent external validation was reported in only 36.7% of the studies.
The main reason to be careful is that the clinical value of these models is undermined by limited external validation and incomplete calibration assessment. The study highlights weak reproducibility practices and poor transparency. Readers should understand that this work serves as a methodological demonstration rather than proof of clinical prognosis.