Machine Learning Algorithms for Predicting in-Hospital Mortality in Patients with ST-Segment Elevation Myocardial Infar

TL;DR

This study developed and validated machine learning models using electronic medical records to predict in-hospital mortality in STEMI patients, achieving high accuracy and identifying key risk factors.

Contribution

The paper introduces a novel application of multiple machine learning algorithms on HPEMR data for early prediction of mortality in STEMI patients, with detailed analysis of associated risk factors.

Findings

Support Vector Machine achieved an AUC of 0.879.

Key risk factors include atrial fibrillation and acute renal failure.

Models demonstrated good discrimination ability in test datasets.

Abstract

Acute myocardial infarction (AMI) is one of the most severe manifestation of coronary artery disease. ST-segment elevation myocardial infarction (STEMI) is the most serious type of AMI. We proposed to develop a machine learning algorithm based on the home page of electronic medical record (HPEMR) for predicting in-hospital mortality of patients with STEMI in the early stage. Methods: This observational study applied clinical information collected between 2013 and 2017 from 7 tertiary hospitals in Shenzhen, China. The patients' STEMI data were used to train 4 different machine learning algorithms to predict in-hospital mortality among the patients with STEMI, including Logistic Regression, Support Vector Machine, Gradient Boosting Decision Tree, and Artificial Neuron network. Results: A total of 5865 patients with STEMI were enrolled in our study. The model was developed by considering 3 types of variables, which included demographic data, diagnosis and comorbidities, and hospitalization information basing on HPEMR. The association of selected features using univariant logistic regression was reported. Specially, for the comorbidities, atrial fibrillation (OR: 11.0; 95% CI: 5.64 - 20.2), acute renal failure (OR: 9.75; 95% CI: 3.81 - 25.0), type 2 diabetic nephropathy (OR: 5.45; 95% CI: 1.57 - 19.0), acute heart failure (OR: 6.05; 95% CI: 1.99 - 14.9), and cardiac function grade IV (OR: 28.6; 95% CI: 20.6 - 39.6) were found to be associated with a high odds of death. Within the test dataset, our model showed a good discrimination ability as measured by area under the receiver operating characteristic curve (AUC; 0.879) (95% CI: 0.825 - 0.933).