Artificial intelligence-based prediction of acute myocardial infarction mortality risk

Acute myocardial infarction, a critical condition in cardiovascular disease, is associated with a high degree of morbidity and mortality. Early identification of the risk of acute myocardial infarction (AMI) can provide treatment to patients as early as possible and effectively reduce the risk of death. In this study, we constructed an AMI mortality risk prediction model based on the MIMIC-IV database and explored the relationship between relevant data characteristics and prediction effects. We included AMI patients in the database into the study sample, and used the SMOTE algorithm to correct the sample for category imbalance after preprocessing the data outliers and missing values. After three rounds of feature extraction we obtained a final dataset that could be used for model construction. We selected three algorithms, XGBoost, LightGBM, and Random Forest, to construct the prediction model, and selected the model with the best effect through a variety of evaluation indexes. Finally, we found that the prediction model constructed based on the XGBoost algorithm has better prediction results, with an accuracy of 91.6% and an AUC value of 97.5%. Subsequently, we interpreted and analyzed the model with the Shapley value, and found that the three features of hr_min, aniongap_max and respiratory_max had a high impact on the prediction results. This effectively ensures the general applicability and ease of use of the model, which is conducive to the timely detection of disease development in patients, reducing mortality and improving the prognosis of AMI patients.