Clinicians around the world are stressed. Staff shortages, administrative burdens, and cognitive overload have become overwhelming. Yet, the need for disease prevention using predictive tools has never been greater. With this goal in mind, several innovative technologists have developed deep learning-based algorithms to help physicians predict out-of-hospital cardiac arrests, in-house arrests, and long-term mortality after cardiac surgery.
Japanese investigators conducted a population-based study that looked at over 1 million patients who developed out-of-hospital cardiac arrests (OHCA) using a gradient boosting algorithm to determine the impact of meteorological and chronological data. They found that combining both parameters improved their ability to predict OHCAs. Nakashima et al reported: “Sunday, Monday, holiday, winter, low ambient temperature and large interday or intraday temperature difference were more strongly associated with OHCA incidence than other meteorological and chronological variables.” In the other words, if a person experiences a sudden change in temperature on days that are very cold or very hot, their odds of experiencing a cardiac arrest increase. The investigators theorized that the increased risk was related to increased sympathetic tone and blood viscosity.
South Korean technologists are also making important contributions in the specialty. They combined a deep learning algorithm with ECG readings in a retrospective analysis that looked at over 47,000 ECGs in 25,672 adult patients admitted to 2 hospitals. The combination was able to predict cardiac arrest within 24 hours with a receiver operating characteristic curve between 93 and 98 percent. Their findings held up whether the patients had a conventional 12-lead ECG or a single-lead ECG using a wearable device that employed the algorithm.
While both studies help establish the value of AI-enhanced algorithms in cardiology, both were retrospective in nature, a methodology that lacks the rigor of a prospective clinical trial. As we mentioned in a previous blog, the Mayo Clinic’s EAGLE trial demonstrated the value of AI algorithms combined with ECG readings in a randomized controlled trial.
A new follow-up data analysis that employed the data from the EAGLE trial suggests that the AI-ECG screen used to detect low left ventricular ejection fraction (LVEF) is also capable of predicting long-term mortality among patients who have undergone valve or coronary bypass surgery at Mayo Clinic. Mohamad A. Alkhouli, MD and associates included over 20,000 patients, of whom 83 percent had a normal AI-ECG screen and compared them to 17 percent who had abnormal readings, and found “Probability of survival at 5 and 10 years was 86.2 percent and 68.2 percent in patients with a normal AI-ECG screen vs 71.4 percent and 45.1 percent in those with an abnormal screen… A novel electrocardiography-based AI algorithm that predicts severe ventricular dysfunction can predict long-term mortality among patients with LVEF above 35 percent undergoing valve and/or coronary bypass surgery.”
In an ideal world, we would be able to tell every patient their precise risk of heart disease, with no margin for error. Until that day arrives, we can all be grateful for a crystal ball that lets us see things more clearly.
Written by John Halamka, MD, president, and Paul Cerrato, senior research analyst and communications specialist at Mayo Clinical Platform, this piece was originally posted to their blog page, Digital Health Frontier.