Health, patients at risk of cardiac arrest: AI identifies them more effectively than doctors

A new AI model is far better than doctors at identifying patients at risk of cardiac arrest. The key is the system’s ability to analyze long-underused cardiac images, along with a wide range of medical records, to reveal previously hidden information about a patient’s cardiac health. The federally funded work, led by researchers at Johns Hopkins University, could save lives and spare many people from unnecessary medical interventions, including the implantation of unnecessary defibrillators. “We currently have patients who die in their prime because they’re not protected, and others who endure defibrillators for the rest of their lives with no benefit,” says senior author Natalia Trayanova, a researcher specializing in the use of AI in cardiology.

We have the ability to predict with very high accuracy whether a patient is at very high risk of sudden cardiac death or not.” The findings are published today in Nature Cardiovascular Research. Hypertrophic cardiomyopathy is one of the most common inherited heart diseases, affecting one in 200 to 500 people worldwide and a leading cause of sudden cardiac death among young people and athletes. Many patients with hypertrophic cardiomyopathy lead normal lives, but a percentage of them are at significantly higher risk of sudden cardiac death. It has been nearly impossible for doctors to identify these patients. Current clinical guidelines used by doctors in the United States and Europe to identify patients most at risk of fatal heart attacks have about a 50 percent chance of identifying the right patients — “not much better than rolling the dice,” Trayanova explains.

The team's model significantly exceeded clinical guidelines across all demographic groups.

The Multimodal AI for Ventricular Arrhythmia Risk Stratification (MAARS) predicts a patient’s risk of sudden cardiac death by analyzing a variety of data and medical records and, for the first time, exploring all the information contained in contrast-enhanced MRI images of a patient’s heart. People with hypertrophic cardiomyopathy develop fibrosis, or scarring, throughout their hearts, and that scarring increases their risk of sudden cardiac death. Although doctors couldn’t interpret the raw MRI images, the AI ​​model zeroed in on the critical scarring patterns. “No one has ever used deep learning on those images,” Trayanova says. “We can extract this hidden information in the images, which is usually not considered.” The team tested the model on real patients treated with traditional clinical guidelines at Johns Hopkins Hospital and the Sanger Heart & Vascular Institute in North Carolina. Compared to clinical guidelines that were accurate about half the time, the AI ​​model was 89 percent accurate across all patients and, crucially, 93 percent accurate for people ages 40 to 60, the population of HCM patients most at risk for sudden cardiac death. The AI ​​model can also describe why patients are at high risk, so that doctors can tailor a care plan to their specific needs. “Our study shows that the AI ​​model significantly improves our ability to predict who is at highest risk compared to our current algorithms and therefore has the potential to transform clinical care,” says coauthor Jonathan Crispin , a cardiologist at Johns Hopkins.