BOB体育 researchers propose AI model to predict mortality in coronary artery disease patients
Md Mahmudul Hasan, Ph.D., an assistant professor in the UF College of Pharmacy Department of Pharmaceutical Outcomes and Policy and the UF Warrington College of Business Department of Information Systems and Operations Management.
GAINESVILLE, Fla. 鈥� An interdisciplinary team led by University of Florida researchers has proposed an artificial intelligence, or AI, model to predict long-term mortality in patients with coronary artery disease. The study was published in the .
Characterized by the narrowing of coronary arteries due to the buildup of fatty plaque, coronary artery disease, or CAD, limits blood flow to the heart. The most common form of heart disease, CAD caused more than 350,000 deaths in the United States in 2022, according to the National Heart, Lung, and Blood Institute.
A team led by an assistant professor in the and the , hopes to improve treatment outcomes and reduce mortality in CAD patients with their lightweight graph neural network.
Traditional methods for predicting CAD mortality, like the Framingham Risk Score, use only established risk factors like age, smoking status and cholesterol level, Hasan said, limiting their accuracy and adaptability to diverse populations. He noted that existing AI models can make accurate predictions by incorporating diverse sets of data, but they also lack the ability to capture one critical ingredient: causality.
With the AI model devised by Mohammad Yaseliani, a graduate student in Hasan鈥檚 lab, clinicians will be able to determine which factors most prominently contribute to a patient鈥檚 risk of death.
鈥淥ur proposed framework provides the option to identify causal features and include them in the AI-based prediction framework,鈥� Hasan said. 鈥淥ur graph neural network can also extract data from neighboring patients, which is useful in understanding how demographic and clinical characteristics collectively influence a patient鈥檚 risk. We also proposed an explainer tool that can help clinicians identify the most important and influential factors putting a patient at risk.鈥�
Analysis performed by Hasan鈥檚 team found that their model outperformed all traditional models in correctly predicting which patients would survive the disease. Before translating the model into a web-based tool for practitioner use in patient care settings, Hasan said it must be validated with demographically diverse external datasets and evaluated for ethical and privacy issues.
鈥淭his model can enhance clinical decision-making,鈥� he said. 鈥淐linicians can identify and stratify patients based on their risk, prioritizing those who need immediate or intensive care. By identifying causal features, our model helps clinicians target and personalize their interventions, make better use of limited health care resources and streamline the clinical workflow.鈥�
The model was developed by a team of scientists, engineers and clinicians from the UF colleges of Pharmacy and and the , as well as Northeastern University and Georgia Tech University.
鈥淚 cannot overstate the power and importance of interdisciplinary collaboration in this research,鈥� Hasan said. 鈥淎fter all, I have an engineering background, yet I鈥檓 embedded in the health care world. Combining the complementary expertise of my colleagues has facilitated innovative problem-solving, ensuring that our model and support systems are both scientifically advanced and clinically relevant. I鈥檓 grateful we were able to create a robust and practical solution that can ultimately improve health outcomes for patients with coronary artery disease.鈥�
