The ability to predict heart attacks remains a significant challenge in modern cardiology. Despite advancements in the field, many individuals are not screened for heart disease, which can lead to missed opportunities for early intervention. However, innovative startups such as Bunkerhill Health, Nanox.AI, and HeartLung Technologies are leveraging artificial intelligence (AI) algorithms to analyze millions of chest CT scans, seeking early indicators of heart disease.
This emerging technology represents a potential breakthrough in public health by using existing tools to identify patients at high risk of heart attacks who might otherwise go unnoticed. While the promise of AI in this context is substantial, its effectiveness on a larger scale remains unverified, raising complex questions regarding implementation and the very definition of disease.
Last year, approximately 20 million Americans underwent chest CT scans, often following incidents like car accidents or as part of lung cancer screenings. These scans frequently reveal signs of coronary artery calcium (CAC), a significant marker for heart attack risk, which may not be adequately addressed in radiology reports that prioritize other injuries or conditions. While dedicated CAC testing is available, it has not been widely utilized in assessing heart attack risk.
The development of algorithms capable of calculating CAC scores from routine chest CT scans could greatly enhance access to this important metric. Such algorithms may notify patients and healthcare providers about unusually high CAC scores, prompting necessary follow-up care. Although the current footprint of these AI-driven services is limited, their adoption is rapidly increasing, potentially uncovering high-risk patients who have traditionally slipped through the cracks of conventional care.
Historically, CAC scans were viewed as having limited benefits and were primarily marketed to individuals without existing health concerns. Even now, many insurance companies do not cover these scans. Nevertheless, perspectives may be shifting as more expert organizations advocate for CAC scores as valuable tools for refining cardiovascular risk assessments and encouraging hesitant patients to consider statin therapy.
The potential of AI-generated CAC scores is part of a wider trend in healthcare where vast amounts of medical data are analyzed to discover previously undetected diseases. However, various concerns persist. For instance, a Danish study conducted in 2022 found no significant improvement in mortality rates among patients who underwent CAC screenings.
If AI automates the identification of CAC, will it lead to better outcomes? With increased implementation, abnormal CAC scores are likely to become commonplace. Who will manage the follow-up on these findings? “Many health systems aren”t yet equipped to handle incidental calcium findings at scale,” explains Nishith Khandwala, co-founder of Bunkerhill Health. Without established protocols, there is a risk of generating more workload than actual benefit.
Another significant consideration is whether these AI-generated scores will indeed enhance patient care. For a patient exhibiting symptoms, a CAC score of zero might create a false sense of security, whereas an asymptomatic individual with a high CAC score faces unclear next steps. Beyond statins, it is uncertain whether these patients would gain from initiating costly cholesterol-lowering treatments, which could even lead to unnecessary procedures that may cause harm.
Currently, Medicare and most insurers do not reimburse for AI-derived CAC scoring as a distinct service. The financial justification for this technology is thus fraught with potential complications. This approach has the capacity to alter fundamentally how we define diseases. Adam Rodman, a hospitalist and AI specialist at Beth Israel Deaconess Medical Center in Boston, notes that AI-derived CAC scores resemble the concept of “incidentalomas,” a term introduced in the 1980s to describe unexpected findings on imaging studies. In both instances, the traditional method of diagnosis—where healthcare professionals and patients intentionally seek tests to understand specific health issues—has been fundamentally altered.
As Rodman points out, incidentalomas were identified by human reviewers of the scans. Now, we are transitioning into an age characterized by “machine-based nosology,” where algorithms autonomously define diseases. As these systems become more adept at making diagnoses, they may uncover conditions that have previously gone unnoticed. However, a two-tiered diagnostic landscape could emerge, where affluent patients opt for premium algorithms while others are relegated to lesser options.
For patients lacking risk factors or those who are disconnected from regular medical care, an AI-derived CAC score could potentially facilitate earlier detection of health issues. Nevertheless, critical questions remain regarding how these scores will be communicated, the actions taken in response, and whether they will genuinely improve health outcomes on a larger scale. As clinicians balance between patient care and algorithmic data, their role remains essential in navigating this evolving landscape.
