The GRACE (Global Registry of Acute Coronary Events) risk score is a well-established tool for predicting major cardiovascular events in patients with acute coronary syndrome. However, its application in acute ST-segment elevation myocardial infarction (STEMI) requires refinement to enhance its predictive accuracy in clinical settings.In this study, we conducted a retrospective analysis of the incidence of out-of-hospital all-cause death (ACD), calculated the correlation and significance of the GRACE score indicators with ACD, and reduced the scores corresponding to insignificant and low-correlation indicators to adjust the scores for survive patients. Using the adjusted GRACE score as the target variable, we trained and optimized and integrated the multiple regression models using the Stacking method(named GRACE-STEMI score adjusted model). Additionally, we performed supplementary SHAP (SHapley Additive exPlanations) analysis.The study ultimately identified nine key variables affecting the adjusted GRACE score: LA, LVEF, neutrophil percentage, lymphocytes, urea, systolic pressure, admission heart rate, age, and Killip classification. Furthermore, by employing the Stacking method to integrate the three best-performing regression models on the training set, the GRACE-STEMI score adjusted model achieved an adjusted R2 score of 0.7886, a C-index of 0.8521, an MSE (Mean Squared Error) of 250.8, and an RMSE (Root Mean Squared Error) of 15.84 on the test set. The model's interpretability was also successfully validated through SHAP analysis.The GRACE-STEMI score adjusted model offers a clinically relevant advancement, providing a more precise tool for risk stratification and mortality prediction in STEMI patients. The integration of SHAP analysis not only enhances the model's interpretability but also facilitates the identification of novel risk groups, contributing to personalized clinical decision-making.