Intelligent alarm to help doctors predict heart attacks

Machine learning

University of Copenhagen researchers are in the process of testing a new alarm that helps Danish physicians predict whether pacemaker and defibrillator patients are about to suffer a serious cardiac event. With the help of artificial intelligence and heart data from a pacemaker, the alarm can calculate any elevated risk of a cardiac event. This supports doctors in making the best possible decisions for patients.

At Denmark’s main teaching hospital, Rigshospital, the cardiology department monitors 4000 patients in the Capital Region of Denmark with pacemakers or implantable cardioverter defibrillators (ICD). Should a sudden change occur to a patient’s heart rhythm, a signal is sent to the cardiology department with data from the pacemaker or ICD. There, the doctor on duty can decide whether the patient should be contacted or hospitalized. The doctor can make this decision based on their experience and information, such as the patient’s medical history and medications.

But with the new alarm system, developed in collaboration between researchers at the Department of Computer Science and the spin-out company Vital Beats, artificial intelligence enhances medical decision making by calculating the risk of serious cardiac events. The artificial intelligence draws upon data from more than 12,000 cardiac events that it has been trained to recognize patterns in – patterns that could lead to a possibly life-threatening cardiac events.

“In just a few seconds, the alarm system refers to the historical cardiac event data available to it and predicts the degree of risk for a serious event. As such, it supports decision making by the physician, which can range from acute hospitalization to rescheduling a routine appointment to discuss a patient’s medication,” explains assistant professor Tariq Osman Andersen of the Department of Computer Science.

Looking to the future

3500 people in Denmark – some with and some without pacemakers or ICDs – die every year due to cardiac arrest. Some of these cases are due to rapid heart rhythms or ventricular flutter, which increases cardiac activity. Over time, these rapid heart rhythms destroy the heart, and in the most severe cases can lead to cardiac arrest when the heart rate reaches 200-300.

“When a patient experiences fast and potentially dangerous heart rhythms, the algorithm can predict whether there is a heightened risk of it occurring again. Our studies demonstrate that the alarm supports doctors in making faster and safer decisions about which course of action to take,” says Tariq Osman Andersen, who adds:

“The artificial intelligence can almost immediately recognize a context that would typically take a doctor much longer to see. And, time is a scarce commodity in hospitals,” says Tariq Osman Andersen.

A decision made by doctors 16,000 times a year

Rigshospital’s cardiology department receives 16,000 heart information transmissions a year. A doctor must assess and determine a course of action for each transmission, which takes time.

“Being able to predict presents an entirely new way of managing this situation. Today, doctors must assess case by case and decide what to do. The algorithm provides a bit of extra help when making these time-consuming and difficult decisions. One can say that the alarm offers what many experienced medical colleagues could come up with if they sat down and thoroughly assessed a particular case,” according to Vital Beats’ CEO and co-founder Jonas Moll.

The alarm system is currently being tested by doctors in Rigshospital’s cardiology department. The SafeHeart project is supported by EU funds and is a collaboration between the University of Copenhagen, Vital Beats and Rigshospital.

Facts:

  • The alarm system is designed to help physicians proactively manage situations when a patient is experiencing changes in their heart rate
  • Artificial intelligence can predict whether there is an increased risk of fast heart rhythms (arrhythmias) within the next 30 days
  • The artificial intelligence builds upon machine learning and is trained in over 12,000 cardiac events gathered from 1250 patients between 2015-2019
  • To validate the results of the artificial intelligence, the alarm is currently being tested on patients at Copenhagen’s Rigshospital

/Public Release. View in full here.