A comprehensive analysis of 120 studies revealed that artificial intelligence models using cardiac biomarkers after heart attacks achieved remarkable diagnostic accuracy, with one framework reaching 99.88% discrimination between STEMI and NSTEMI types using multimodal inputs. When limited to biomarkers alone, the system still maintained 93% accuracy with fewer variables, suggesting powerful predictive potential in cardiac care. However, this systematic review exposed critical methodological flaws plaguing the field. Only 37% of studies included independent external validation, and most lacked proper calibration analyses or explainability methods. The research landscape suffers from inconsistent approaches, with logistic regression and Random Forest being most common but applied heterogeneously across studies. While these AI advances could revolutionize post-heart attack prognosis and treatment decisions, the practical clinical impact remains uncertain due to weak reproducibility practices and insufficient real-world testing. The authors developed a "leakage-aware" framework to address these limitations, but as this is a preprint awaiting peer review, these promising results require validation before clinical implementation. The findings represent incremental progress in cardiac AI, but highlight that methodological rigor must catch up with technological capabilities.