Can AI In Healthcare Save Lives?
Some visionary pundits believe the sci-fi genre has correctly predicted artificial intelligence (AI) may revolt against the human race and destroy it. At least in the short to medium term, I believe AI may save lives. As I wrote in Robot Attitude: How Robots and Artificial Intelligence Will Make Our Lives Better, AI is having a significant positive impact in healthcare. The uses vary widely including making stethoscopes smarter, predicting and reducing falling in the hospital, enhancing emergency calls, predicting fatal conditions, assuring the right medications for patients, and enhancing the accuracy of diagnoses from imaging studies. A recent new addition to the list is enhanced diagnoses from electrocardiograms (ECGs).
An ECG is a painless, noninvasive way to help diagnose many common heart problems in people of all ages. Apple has introduced an ECG capability on the Watch. It is equivalent to a one-lead measurement but can be very useful in detecting atrial fibrillation. Kardia has a small device which works with a smartphone and performs the equivalent of a six-lead ECG. Doctor offices and hospitals use a 12-lead device which is the most accurate. The cost billed for an ECG varies from $50 to several thousand. People without insurance are billed much higher amounts for the identical test given to the insured. Medicare reimburses doctors and hospitals about $15. What a system, but that is another story.
An ECG records the electrical activity of your heart at rest. It can identify numerous conditions including your heart rate and rhythm, enlargement of the heart due to high blood pressure, and evidence of a previous heart attack. Another important measurement of the heart is the ejection fraction (EF).
The heart contracts and relaxes. When the heart contracts, it pumps out (ejects) blood from the lower chambers of the heart (ventricles). When the heart relaxes, the ventricles fill with blood. Regardless of the strength of the contraction, the heart can only pump a fraction of the blood from a ventricle. The ejection fraction refers to the percentage of blood which gets pumped out with each heartbeat. A normal EF is 50% to 75%, according to the American Heart Association. A borderline EF may be between 41% and 50%. A lower EF indicates your heart is not working as well as it should and can suggest heart failure and increased mortality.
The most common test used to measure EF is an echocardiogram is. An echocardiogram uses sound waves to produce images of the heart and the blood pumping through the heart. Enter AI. The Mayo Clinic has developed an algorithm which can significantly increase the number of cases identified with a low ejection fraction. A new study has shown early identification is important because problems are then more treatable. A low EF can imply serious heart problems, but they do not always show symptoms in the early stages.
What is unique about the AI at Mayo Clinic is it can identify low EF from ECG data, not requiring an expensive and time-consuming cardiac echocardiogram. The AI looks at the data in ways which a human cannot and puts any positive results directly into the electronic health record of the patient for follow-up. A study of more than 22,600 patients received an EKG as part of their usual primary care checkups. The group was then randomly assigned to have their results analyzed either by the AI or by a physician. The algorithm produced 32% more diagnoses of low EF compared to what the physician could detect.
Peter Noseworthy, a Mayo Clinic cardiac electrophysiologist and senior author on the study said, “The AI-enabled EKG facilitated the diagnosis of patients with low ejection fraction in a real-world setting by identifying people who previously would have slipped through the cracks.” Some physicians are skeptical, but Noseworthy said, “The takeaway is that we are likely to see more AI use in the practice of medicine as time goes on. It’s up to us to figure how to use this in a way that improves care and health outcomes but does not overburden frontline clinicians.”
There are many Medical AI research projects underway, but widespread adoption in hospitals and medical centers has not grown substantially, yet. The barrier is not the technology. Shinjini Kundu, a medical researcher and physician at the University of Pittsburgh School of Medicine summed it up very well. He said, “The barrier is the trust aspect. You may have a technology that works, but how do you get humans to use it and rely on it?” The way AI is being applied in medicine is to absorb a very large amount of data, analyze it, find relationships, and make a diagnosis based on the learning. In effect, a lot of data goes into a “black box” where algorithms are applied, and out comes the diagnosis. This is quite different from how physicians diagnose. If they can’t see inside the black box, they will not trust the accuracy of the diagnosis. AI developers understand this and are addressing it. I believe we will see great progress and saved lives in the days ahead.