Machine Learning Prediction and Interpretability Analysis of High-risk Chest pain: A study from the MIMIC-IV database

Hongyi Chen¹Haiyang Song¹Hongyu Huang²Xiaojun Fang³Chen Huang⁴Qingqing Yang⁵Junyu Zhang⁶Wenjun Ding⁷Zheng Gong^2*Jun Ke^4*

• ¹Department of Emergency, Fujian Provincial Hospital, Fuzhou, Fujian Province, China
• ²Shengli Clinical Medical College, Fujian Medical University, Fuzhou, Fujian Province, China
• ³Fujian Sanbo Funeng Brain Hospital, Fuzhou, Fujian Province, China
• ⁴Fuzhou University Affiliated Provincial Hospital, Fuzhou, Fujian Province, China
• ⁵Fujian Provincial Key Laboratory of Cardiovascular Diseases, Fujian Provincial Hospital, Fuzhou, Fujian Province, China
• ⁶Guangxi Normal University, Guilin, China
• ⁷Xiamen University, Xiamen, Fujian Province, China

High-risk chest pain is a critical symptom in emergency departments, often associated with severe diseases like heart and lung conditions. Timely diagnosis is essential for improving patient survival. This study proposes a machine learning-based high-risk chest pain prediction model using the MIMIC-IV database. Feature engineering, SMOTE, and under-sampling were applied to address class imbalance, while Bayesian optimization fine-tuned the model. A total of 14,716 patients, including 1,302 with high-risk chest pain, were analyzed. Various machine learning algorithms, including Logistic Regression, Random Forest, SVM, XGBoost, LightGBM, TabTransformer, and TabNet, were compared. The LightGBM model achieved the highest accuracy (0.95), precision (0.95), recall (0.95), and F1 score (0.94). SHAP analysis identified maximum troponin and creatine kinase isoenzyme MB as the most important predictors. The LightGBM model outperformed others in predicting high-risk chest pain, offering a promising tool for clinicians to make accurate diagnoses and improve patient outcomes in emergency care.