For all the incredible advancements in modern medicine, particularly in cardiology, there’s a surprising challenge we still face: accurately predicting who will suffer a heart attack. Many individuals, perhaps even millions, go about their lives completely unaware of the ticking time bomb within their arteries. They might never receive the screenings that could offer an early warning, leaving them vulnerable to an event that could change their lives forever. But what if the answer wasn’t a brand new, expensive test, but rather a smarter way to look at data we already have? Enter artificial intelligence.
A quiet revolution is brewing in medical diagnostics, spearheaded by innovative startups. These companies are harnessing the power of AI to sift through vast troves of existing medical imaging data, specifically CT scans, to uncover hidden clues about heart disease. It’s an exciting prospect that could fundamentally shift how we identify and manage cardiovascular risk, turning an old diagnostic tool into a powerful new weapon in public health. But like all powerful innovations, it comes with its own set of fascinating questions and challenges.
The Hidden Clues in Plain Sight: How AI Spots Early Risk
Think about the sheer volume of medical imaging that happens every day. Last year alone, an estimated 20 million Americans underwent chest CT scans. These scans are often performed for immediate concerns, like after a car accident to rule out internal trauma, or as part of a lung cancer screening program. The radiologist’s focus is naturally on the primary reason for the scan – looking for bony injuries, life-threatening internal bleeding, or suspicious nodules.
However, these very same scans frequently contain evidence of something else entirely: coronary artery calcium (CAC). This calcification in the heart’s arteries is a clear, objective marker for heart attack risk. It’s often “buried” within the images, not explicitly mentioned or quantified in a radiology report focused on, say, a fractured rib or a clear lung field. For years, this critical information has been sitting there, effectively “hiding in plain sight.”
This is where AI algorithms are stepping in. Startups like Bunkerhill Health, Nanox.AI, and HeartLung Technologies are developing and deploying AI that can automatically detect, identify, and even quantify CAC scores from these routine chest CTs. They’re applying sophisticated machine learning to extract insights from data that human eyes, focused on other tasks, might routinely miss or simply describe subjectively.
CAC itself forms over decades as plaque in heart arteries moves through its lifecycle, hardening from lipid-rich residue into calcium. While calcified plaque is generally stable, its presence is a strong indicator that younger, more rupture-prone plaque is likely also present. And it’s the rupture of this younger, lipid-rich plaque that typically triggers the inflammatory cascade leading to a heart attack.
Expanding Access and Shifting Perspectives
Historically, dedicated testing for CAC has been an underutilized method for predicting heart attack risk. Despite its proven predictive power, these specific heart CT scans often weren’t covered by most insurers, making them accessible primarily to the “worried well” who could afford to pay out-of-pocket. Attitudes, however, are beginning to shift, with more expert groups endorsing CAC scores as a valuable tool to refine cardiovascular risk estimates and even persuade skeptical patients to start life-saving medications like statins.
This is precisely where AI-derived CAC scores offer a monumental leap forward. By leveraging existing, routine chest CTs, these algorithms could massively expand access to this crucial metric. Imagine millions of people who underwent a CT scan for an unrelated issue suddenly receiving an alert about an abnormally high CAC score, prompting them and their doctors to seek further care. This technology has the potential to identify high-risk patients who traditionally fall through the cracks or exist on the margins of the healthcare system.
The ‘Incidentaloma’ Effect: A New Era of Diagnosis?
The concept of AI-derived CAC scores isn’t entirely new in its essence. It echoes the idea of an “incidentaloma,” a term coined in the 1980s to describe unexpected findings on CT scans. In those cases, a scan performed for one reason would reveal, say, a small, asymptomatic growth on the kidney or adrenal gland. The diagnostic process was fundamentally disrupted, moving away from a doctor and patient deliberately investigating a specific problem.
But there’s a crucial difference. As Adam Rodman, a hospitalist and AI expert at Beth Israel Deaconess Medical Center in Boston, points out, incidentalomas were still found by human eyes reviewing the scans. What we’re entering now is an era of “machine-based nosology,” where algorithms themselves define and identify diseases on their own terms. Machines, unburdened by human fatigue or oversight, might catch things we consistently miss, pushing the boundaries of what we consider a diagnosis.
Navigating the Thorny Path: Challenges and Unanswered Questions
While the promise of AI for early heart attack prediction is compelling, it’s not without its complexities. The practice of mining troves of medical data for undetected disease, while promising, raises plenty of practical and ethical questions. For instance, studies on population-based CAC screening, without AI, haven’t always shown a clear benefit in mortality rates. A 2022 Danish study, for example, didn’t find improved mortality for screened patients. Would simply automating the detection through AI fundamentally shift this calculus?
Then there’s the massive question of implementation. If AI systems were to deliver these insights automatically and at scale, abnormal CAC scores would quickly become common. Who follows up on these findings? As Nishith Khandwala, cofounder of Bunkerhill Health, aptly notes, “Many health systems aren’t yet set up to act on incidental calcium findings at scale.” Without a standardized procedure, he warns, “you risk creating more work than value.”
Furthermore, the actual benefit to patient care isn’t always straightforward. For a patient experiencing chest pain, a CAC score of zero might offer false reassurance, delaying critical investigation. Conversely, for an asymptomatic patient with a high CAC score, the optimal next steps remain uncertain. Beyond starting statins, it’s unclear if these patients would benefit from costly cholesterol-lowering drugs like Repatha or other PCSK9-inhibitors. There’s even a risk of encouraging unnecessary, expensive downstream procedures that could, paradoxically, do more harm than good. Today, AI-derived CAC scoring isn’t reimbursed as a separate service by Medicare or most insurers, leaving a gap where the business case for this technology might inadvertently rely on these potentially “perverse incentives.”
There’s also a deeper, more philosophical concern. If machines begin to define diseases, could this lead to a two-tiered diagnostic future? Rodman speculates about a scenario where the “haves” pay for sophisticated, brand-name algorithms, while the “have-nots” might settle for lesser alternatives or miss out entirely. This raises profound questions about equity and access in an AI-driven healthcare landscape.
The Human Element Remains Central
The vision of AI acting as an early warning system for heart attacks is undoubtedly powerful. For patients who have no traditional risk factors or are disengaged from regular medical care, an AI-derived CAC score could be a game-changer, catching problems far earlier than ever before and potentially rewriting their health trajectory. It’s an exciting frontier in personalized medicine, leveraging our growing data capabilities.
However, the journey from algorithm to improved patient outcomes at scale is fraught with open questions. How effectively will these scores reach the right people? What standardized actions will be taken based on these findings? And ultimately, how will we ensure this technology genuinely improves public health without exacerbating existing inequalities or creating new problems?
For now, amidst the dazzling potential of machine learning and the promise of a smarter future, the human element remains irreplaceable. Clinicians, with their nuanced understanding of individual patient contexts, their ability to interpret data with compassion, and their critical judgment, still matter profoundly. They are, after all, holding the pen, toggling between complex algorithmic outputs and the unique stories of the patients sitting before them.
