Manu Kumar Shetty1, Shekhar Kunal2, M P Girish2, Arman Qamar3, Sameer Arora4, Michael Hendrickson5, Padhinhare P Mohanan6, Puneet Gupta7, S Ramakrishnan8, Rakesh Yadav8, Ankit Bansal2, Geevar Zachariah9, Vishal Batra2, Deepak L Bhatt10, Anubha Gupta11, Mohit Gupta12. 1. Department of Clinical Pharmacology, Maulana Azad Medical College, Delhi, India. 2. Department of Cardiology, GB Pant Institute of Postgraduate Medical Education and Research, New Delhi, India. 3. Section of Interventional Cardiology and Vascular Medicine, NorthShore University Health System, University of Chicago Pritzker School of Medicine, Evanston, IL, United States. 4. Division of Cardiology, University of North Carolina School of Medicine, Chapel Hill, NC, United States. 5. University of North Carolina School of Medicine, Chapel Hill, NC, United States. 6. Department of Cardiology, Westfort Hi-Tech Hospital, Punkunnam, Thrissur, Kerala, India. 7. Department of Cardiology, Janakpuri Superspeciality Hospital, New Delhi, India. 8. Department of Cardiology, All India Institute of Medical Sciences, Delhi, India. 9. Mother Heart Care, Thrissur, Kerala, India. 10. Heart and Vascular Institute, Brigham and Women's Hospital, Harvard Medical School, Boston, MA, United States. 11. SBILab, Department of Electronics and Communications Engineering, Indraprastha Institute of Information Technology-, Delhi IIIT-D, India. Electronic address: anubha@iiitd.ac.in. 12. Department of Cardiology, GB Pant Institute of Postgraduate Medical Education and Research, New Delhi, India. Electronic address: drmohitgupta@yahoo.com.
Abstract
BACKGROUND: Risk prediction following ST-Elevation Myocardial Infarction (STEMI) in resource limited countries is critical to identify patients at an increased risk of mortality who might benefit from intensive management. METHODS: North India ST-Elevation Myocardial Infarction (NORIN-STEMI) is an ongoing registry that has prospectively enrolled 3,635 STEMI patients. Of these, 3191 patients with first STEMI were included. Patients were divided into two groups: development (n=2668) and validation (unseen) dataset (n=523). Various ML strategies were used to train and tune the model based on validation dataset results that included 31 clinical characteristics. These models were compared in sensitivity, specificity, F1-score, receiver operating characteristic area under the curve (AUC), and overall accuracy to predict mortality at 30 days. ML model decision making was analyzed using the Shapley Additive exPlanations (ShAP) summary plot. RESULTS: At 30 days, the mortality was 7.7%. On the validation dataset, Extra Tree ML model had the best predictive ability with sensitivity: 85%, AUC: 79.7%, and Accuracy: 75%. ShAP interpretable summary plot determined delay in time to revascularization, baseline cardiogenic shock, left ventricular ejection fraction <30%, age, serum creatinine, heart failure on presentation, female sex, and moderate-severe mitral regurgitation to be major predictors of all-cause mortality at 30 days (P<0.001 for all). CONCLUSION: ML models lead to an improved mortality prediction following STEMI. ShAP summary plot for the interpretability of the AI model helps to understand the model's decision in identifying high-risk individuals who may benefit from intensified follow-up and close monitoring.
BACKGROUND: Risk prediction following ST-Elevation Myocardial Infarction (STEMI) in resource limited countries is critical to identify patients at an increased risk of mortality who might benefit from intensive management. METHODS: North India ST-Elevation Myocardial Infarction (NORIN-STEMI) is an ongoing registry that has prospectively enrolled 3,635 STEMI patients. Of these, 3191 patients with first STEMI were included. Patients were divided into two groups: development (n=2668) and validation (unseen) dataset (n=523). Various ML strategies were used to train and tune the model based on validation dataset results that included 31 clinical characteristics. These models were compared in sensitivity, specificity, F1-score, receiver operating characteristic area under the curve (AUC), and overall accuracy to predict mortality at 30 days. ML model decision making was analyzed using the Shapley Additive exPlanations (ShAP) summary plot. RESULTS: At 30 days, the mortality was 7.7%. On the validation dataset, Extra Tree ML model had the best predictive ability with sensitivity: 85%, AUC: 79.7%, and Accuracy: 75%. ShAP interpretable summary plot determined delay in time to revascularization, baseline cardiogenic shock, left ventricular ejection fraction <30%, age, serum creatinine, heart failure on presentation, female sex, and moderate-severe mitral regurgitation to be major predictors of all-cause mortality at 30 days (P<0.001 for all). CONCLUSION: ML models lead to an improved mortality prediction following STEMI. ShAP summary plot for the interpretability of the AI model helps to understand the model's decision in identifying high-risk individuals who may benefit from intensified follow-up and close monitoring.