Will AI-Based Opportunistic Screening in Medical Imaging Deliver Real-World Impact?

Understanding AI-Based Opportunistic Screening

One of the most promising applications of artificial intelligence (AI) in healthcare concerns the field of medical imaging. Opportunistic screening refers to the detection of incidental findings not directly relevant to the primary clinical indication – that is, the primary purpose for which a certain test such as an X-ray or CT was carried out.

By leveraging these incidental or opportunistic findings in a systemic approach, opportunistic screening can potentially save lives and improve patient outcomes through early detection of diseases that would otherwise have gone unnoticed. This article examines the question of how valuable AI-based opportunistic screening may be in the field of medicine.

What the 2023 Cost-Effectiveness Study Found

A 2023 study examined the cost-effectiveness of AI-based CT opportunistic screening for three conditions: cardiovascular disease, osteoporosis, and sarcopenia. The study found that the screening was cost-effective for all three conditions: The benefit that the patients gained from the screening outweighed the economic cost.

However, it is important to note that this was a simulation study and has not yet been validated in a prospective clinical trial; patient value of using AI-based software was extrapolated from outside data and may therefore vary from the actual cost.

Four Challenges Hindering Real-World Implementation

And for reasons that will be enumerated below, the real-world effectiveness of AI-based opportunistic screening is a matter of far more complexity involving many stakeholders involved either directly or indirectly in healthcare.

1. Insurance coverage

For opportunistic screening to be widely adopted, it needs to be covered by insurance. This is more likely to happen for conditions for which preventive screening is already covered under existing insurance codes. For example, the US Preventive Services Task Force (USPSTF) recommends screening for the following three cases:

(1) diabetes and prediabetes: in adults aged 35-70 who are overweight or obese
(2) osteoporosis: in women aged 65 and older, or women younger than 65 who have gone through menopause
(3) Abdominal aortic aneurysm: Once in a lifetime for male smokers aged 65-75

Conditions other than these three, as well as population groups not indicated by the above recommendation, are not likely to make the cut for insurance coverage.

One notable example is CT-based coronary artery calcium (CAC) scanning: AI can measure the level of CAC from CT images and predict the risk of cardiovascular disease. The American College of Cardiology, which is the preeminent authority on cardiovascular health in the United States, has been advocating for CAC scanning to be covered for a select group of patients. This would, according to the cardiologists, better identify patients at risk and allow for timely initiation of lipid-lowering drug treatment. Despite such expert opinion, CAC is not covered by most insurers. Aetna, however, became an exception when it announced insurance coverage for CAC screening starting in April 2024. Whether more insurers will follow remain to be seen.

2. Patient outcome

The second issue is whether opportunity screening will actually improve patient outcomes. To date, most proposed opportunity screening tests belong to the category of chronic disease management, such as cardiovascular risk and diabetes mellitus. It is known that these conditions are heavily dependent on lifestyle modification and medication compliance, which is hard to guarantee in the real-world clinical setting. When the actual patient outcome of a new screening modality is hard to ascertain, it is likely to be met with some resistance and skepticism.

3. Clinical workflow

The third and equally tangible challenge is that the information from the screening is often not handled by the same doctor who is ordering the primary imaging study. This can make it difficult to coordinate care and ensure that patients receive appropriate follow-up. Such concern is especially relevant in a setting where, for instance, overworked doctors may hesitate to order potentially unnecessary tests, fearing that incidental findings will create additional work.

4. Technical issues in reimbursement

Let us revisit cases where similar screenings have already been established as valuable health checkups (diabetes, osteoporosis, and abdominal aortic aneurysms; mentioned above under “1. Insurance coverage”). Medicare, the U.S. national health insurance for seniors, covers diabetes screening blood tests twice a year. If a patient undergoes diabetes screening blood tests twice in one year and an X-ray for suspected pneumonia incidentally reveals indications of diabetes in the same year, how should insurance coverage be handled?

The problem is that the current healthcare system is not fully interconnected and tracked real-time. Since opportunistic screenings were not originally conducted for screening purposes, it is difficult to incorporate them into reimbursement processes, adding a layer of technical and administrative complexity.

One approach could be to reimburse fees only when abnormalities are detected– specifically targeting diseases with critical and potentially imminent impact on patient health, such as pancreatic cancer, abdominal aortic aneurysm, and embolism in cancer patients. Still there is the issue of prevalence: These conditions are not that common to begin with. And even then, a large portion of these cases are likely to be diagnosed not through opportunistic screenings but rather due to symptom manifestations – such as jaundice or abdominal pain in pancreatic cancer.

High-Potential Use Cases:
Pancreatic Cancer and Osteoporosis

Detecting pancreatic cancer through non-contrast abdominal CT scans is one area where AI-based opportunistic screening may prove its value, as it is practically impossible for human doctors to do so. But for conditions like abdominal aortic aneurysms and pulmonary embolisms in cancer patients, which doctors can generally detect but sometimes miss, assigning value becomes more ambiguous. If the value of AI hinges on identifying what doctors have missed, how can this be proven?

Final Thoughts:
More Complex Than a Low-Hanging Fruit

Opportunistic screenings - many view them as easy yet valuable byproducts or low-hanging fruits obtained with the advent of AI in medical imaging. But the issues are more complex than may appear from first glance. The healthcare system is an intricate beast, and proving value and obtaining insurance reimbursement within this system is far from easy.

The challenges I have mentioned include getting insurance coverage, demonstrating the impact of screening on patient outcomes, integrating AI into the clinical workflow, and technical issues in reimbursement. Considering these issues, perhaps the detection of pancreatic cancer and osteoporosis through non-contrast CT may be viable candidates for real-world application of opportunistic screening. We shall see.

Written by Chiweon Kim, a Vice President in the Investment Team at Kakao Ventures.