Artificial intelligence is poised to alter care in many medical specialties over the next several years. But few doctors are likely to see as much change as cardiologists, whose practices are already shifting with the use of algorithms to monitor patients’ hearts and produce reams of personalized data.
These data can be used not only to detect troublesome heart rhythms, but also to predict future life-threatening heart problems and inform treatment decisions that may head off those episodes.
“In the next three to five years, we will see a huge change in how we practice and how we teach,” said Dr. Samir Kapadia, chair of cardiovascular medicine at Cleveland Clinic. “Whether doctors are exposed to it directly or indirectly, it’s inevitable they will utilize information derived from AI.”
Here’s a glimpse into the future of AI-augmented cardiac care and the aspects of the field already undergoing significant change.
Using data to assess patients’ risks of developing heart disease is not a new idea. But modern machine learning algorithms are supercharging these capabilities, sharpening the acuity of doctors’ predictions.
A steady stream of research has demonstrated that AI systems can predict the onset of coronary artery disease and other problems with high levels of accuracy. One study released at a European cardiology conference this year demonstrated that an AI system was able to identify patterns in imaging data (CT and PET scans) that were correlated with heart attack and death. The system was able to predict the risk of those outcomes better than doctors.
Traditional risk assessment involves analyzing factors such as a patient’s weight, age, lifestyle, and medical history. Cardiologists said those will remain an important part of the calculus, but the ability to use machine learning to analyze images and other data promises to make these risk estimates faster and more precise.
Kapadia said the Cleveland Clinic is using advanced machine learning to help identify patients with higher risks of complications. “It helps us make our decisions in a more informed way,” he said. “We are working on it and have several [algorithms] up and running.”
Using AI to diagnose stroke offers the potential to improve outcomes for patients — and the technology is already in circulation.
Last year, the Food and Drug Administration approved an AI system that analyzes computed tomography (CT) scans to notify doctors of a potential stroke. The company that created the system, Viz.ai, sends alerts to the mobile devices of radiologists and interventional cardiologists when a possible stroke is detected, helping to speed up response times. The system’s analysis of the scan makes it easier for the radiologist to review and confirm the diagnosis and proceed with an action plan.
That can be crucial, because for the most common type of stroke, a clot-dissolving drug should be administered within three to 4.5 hours from the start of symptoms.
“The challenge we have right now is the [response] is not synchronized,” said Dr. Chris Mansi, the chief executive of Viz.ai.
Mansi said studies of the system’s impact on time to treatment have shown a large variance — anywhere from 6 to 206 minutes. That shows that doctors are just beginning to adjust to the technology, but that significant improvements are possible.
Viz.ai recently struck a deal with the device maker Medtronic to distribute its software to hospitals, which is likely to increase the uptake of its system across the country. Several other companies and private research teams have also demonstrated the ability of AI to analyze changes in the brain that may give rise to stroke and other conditions, potentially making it easier to treat patients earlier and avert emergencies.
A wave of new devices, and the algorithms that support them, are helping cardiologists keep closer watch on patients outside medical settings, where the generation of high quality data can be the difference between life and death.
Apple and companies such as AliveCor have garnered huge attention for devices designed to detect potentially problematic arrhythmias, such as atrial fibrillation. But several other firms are also digitizing traditional tools to allow for closer monitoring of people with heart failure or patients recovering from surgery.
A San Francisco-based company called Eko has developed a bluetooth-enabled digital stethoscope that can automatically transmit heart sounds and electrocardiogram data to doctors.
“We have seen an explosion in the use of our product for telehealth,” said Jason Bellet, a co-founder of Eko. “Doctors are using it to listen to heart and lung sounds remotely, along with a video platform.”
That capability makes it easier for physicians to keep tabs on patients who welcome the stepped up surveillance and the reduction in office visits. Doctors can remotely detect the onset of emergencies when, or even before, they occur. Bellet said that the company is still awaiting FDA approval of algorithms designed to interpret the data from its devices.
Earlier this year, another company, Scotland-based Current Health, received clearance for a different device that uses machine learning to track and analyze vital signs in patients with heart failure and other conditions.
Connected devices and AI are making it much easier for cardiologists to conduct research on specific patient populations.
Eko’s device is being used in a clinical trial by the Mayo Clinic to evaluate heart function declines in breast cancer patients receiving a specific type of chemotherapy.
“It goes beyond the standard of care today, which would be to give a patient a digital scale and tell them to report weight gain or complete a survey of symptoms,” Bellet said. “This can tell doctors what’s happening on a physiological level maybe even before the patient begins experiencing symptoms.”
Many hospitals are also using machine learning to recruit patients for clinical trials. Algorithms designed by doctors and outside companies facilitate combing through electronic health records to identify patients with particular cardiac abnormalities or resistance to conventional treatments. That reduces to minutes a research process that can take many hours of manual screening.
“Machine learning is super crucial for research,” Kapadia, the Cleveland Clinic cardiologist said. “If we can search regular text, put context to it, and find the right patients based on the protocol for the study, we can do research more efficiently and target the right population.”
Doctors have begun to identify voice changes that are correlated with cardiovascular disease.
A study by Mayo physicians in 2018 identified five voice characteristics associated with the presence of coronary artery disease in patients. The study was observational and limited in scope, and it focused on mostly white, middle-aged subjects. But its author said it serves as a jumping off point for further inquiry.
“The potential is very big, but like every technology we need to be careful that we’re not overstating the capability of the artificial intelligence,” said Dr. Amir Lerman, a co-author of the study. “It should not replace our clinical decision-making, it should help us make those decisions.”
He said the identification of voice signals, if verified through further research, could be used to screen patients for signs of disease, a particularly beneficial technology for rural populations that lack immediate access to care. Lerman said he is conducting another study to examine whether the voice signals can be identified across multiple languages.