A strange new dialogue is occurring somewhere between Oxford’s cardiac wards and the tiny glass screens we carry in our pockets. Physicians who previously relied solely on blood panels, stethoscopes, and the laborious calculation of cholesterol levels are now sitting across from software that says it can predict a heart attack ten years in advance. Although it sounds like science fiction, the research is actual, peer-reviewed, and subtly entering hospitals that are typically reluctant to adopt new practices.
CaRi-Heart®, created by Professor Charalambos Antoniades and his team at the University of Oxford, is the most discussed of these tools. Compared to a human radiologist, the technology interprets cardiac CT scans differently. It examines the fat around the coronary arteries rather than searching for the obvious villains, the visible plaques and narrowed arteries. It turns out that inflammation concealed in that fat is a more subtle but accurate indicator of impending problems. Approximately 350,000 cardiac scans are carried out in the UK annually, and a startling number of patients are sent home only to experience heart attacks later, according to the British Heart Foundation, which provided funding for a large portion of this study.
There’s a minor irony that merits consideration. Hospitals have been using the obvious—visible blockages, cholesterol levels, and blood pressure readings taped to clipboards—to measure heart risk for decades. However, on their initial scans, two-thirds of patients who go on to have heart attacks do not exhibit any discernible narrowing. AI appears to be working most quietly and effectively in that gap between what physicians could see and what was really going on inside the body.
Then there’s the other half of this tale, the one who doesn’t even reside in the hospital. An iOS app called Antshrike, named after a South American bird that alerts other animals to impending predators, was released by Before Health Intelligence, a Toronto-based startup. It generates a customized risk score for major ischemic events by extracting biometric data from an Apple Watch and applying two layers of machine learning. After being trained on more than two million CDC data points, the company claims to have over 85% accuracy. It’s unclear if those figures hold up at scale, but the ambition is impressive.
It’s difficult to ignore the differences between the two methods. A hospital, a CT scanner, and a radiologist with algorithmic overlay interpretation training are all necessary. The other only requires the user to wear a watch, lives on their wrist, and operates in the background while they go grocery shopping. Patients are caught between two ideas about how preventive medicine should truly function because they are approaching the same issue from different ends of the healthcare system.
The skepticism is justified. AI in medicine has a history of overpromising, with impressive early research that falters in practical applications. Noisy data is produced by wearables. It is possible to misinterpret inflammatory markers. Furthermore, the discrepancy between an 85% accuracy rate in a controlled dataset and a trustworthy alert for a 56-year-old in Manchester or Karachi is greater than what the press releases indicate. Nevertheless, something seems to have changed as we watch this play out. AI’s ability to forecast cardiac events is no longer a question. It’s whether the surrounding system—physicians, regulators, insurers, and regular people—can catch up in time to apply what it’s saying.
FAQ’s
1: What is CaRi-Heart® and how does it work?
A: It’s an AI tool developed at Oxford that analyzes CT scans of the fat around coronary arteries to detect inflammation and predict heart attack risk up to 10 years in advance.
2: How accurate is the Antshrike™ app at predicting heart attacks?
A: The app claims more than 85% predictive accuracy using biometric data collected from Apple smartwatches and phones.
3: Who developed the AI heart attack prediction technology?
A: Professor Charalambos Antoniades and his team at the University of Oxford, with funding from the British Heart Foundation.
4: When will the Antshrike™ app be available to the public?
A: The freemium app is expected to launch in the United States in 2025, according to Before Health Intelligence.
5: Why do traditional CT scans miss so many heart attack risks?
A: Because two-thirds of patients who later suffer heart attacks show no significant artery narrowing on initial scans, which is the gap AI tools aim to close.
