Electronic health records (EHR) encompass extensive personal information, diagnostic records, and medical history, enabling the prediction of disease occurrence and mortality risk. The objective of this study is to predict myocardial infarction complications and assess the risk of death by comparing the performance of various deep learning models and traditional machine learning approaches. The findings demonstrate similar performance between two kinds of models in predicting complication. The DeepFM model is commonly employed for Click-through rate (CTR) prediction. To the best of our knowledge, this is the first application of the DeepFM model to the EHR domain, and we have demonstrated its exceptional predictive performance, achieving the accuracy of 93.95%. Moreover, we further classify samples into low, intermediate, and high- risk categories with high confidence. To comprehend these results, we conduct an interpretability analysis of the models’ predictions employing SHAP values. This analysis involves ranking the significant features, and summarizing ECG-related features, which hold clinical decision-making revelance for clinicians.