A 2019 point-counterpoint debate between Cedric Yu and Thomas Bortfeld centered on disruptive innovation in radiation oncology is worth revisiting.¹ Both authors offer nuanced, thoughtful perspectives, and we won’t pick a winner here, but the TL/DR version is this: true disruptive innovation in radiation oncology is unicorn-like in its rarity.

Exactly how the oncoming societal tsunami of artificial intelligence (AI) in all of its protean forms will impact the field of radiation oncology in particular — and whether it will manifest as a disruptive innovation in health care delivery according to the original Christensen paradigm² — is open to speculation. And so, with the caveat that our crystal ball has not yet made either of us a profit,³ speculate we will…

Reimbursement 101

First, to frame the possible rippling economic consequences in radiation oncology of AI-driven technologies in the United States, for those unfamiliar with the topic, we offer a quick primer on reimbursement assessments by federally funded payers. In its simplest form, the process typically involves (1) a definition of a unique medical procedure or service accepted by the AMA’s Common Procedural Terminology (CPT) committee; (2) valuation of the service by the AMA’s Relative Value Update Committee (aka the RUC); and (3) a final determination of payments by the Centers for Medicare and Medicaid Services (CMS), usually but not always in line with RUC recommendations. Finally, there are some differences in how payments for technical services are established in freestanding versus hospital-based clinics.

Professional societies like ASTRO have committees within them that play an active role in this sequence at all points along the way, and individual physicians also have opportunities to provide valuable input. For example, in preparation for step 2 in the process, i.e., billing code valuation by the RUC, ASTRO sends out surveys to members to gauge the resources required to provide the service or procedure in question. Responding to these surveys with honest answers is a great way to have a voice in this important activity.

There are various rules and regulations that add complexity. For example, for a service that is considered a brand new technology, there is a scheduled revaluation by the RUC a few years after the initial introduction, based on the rationale that likely there are operational efficiencies that emerge after broader implementation, thus perhaps reducing resource expenditures. Other triggers for a billing code revaluation include a sudden surge in utilization or the observation that two separate codes are very frequently billed together. The latter case is viewed as a possible signal that by furnishing two services together, there might be operational efficiencies, i.e., reduced resource needs, that warrant a reduced total payment relative to what had been determined appropriate initially if the services were furnished individually. Reimbursements may then be further adjusted by CMS to satisfy statutory requirements of budget neutrality, whereby spending increases for new services are offset by equivalent decreases in payments for other services.

The RUC has come under scrutiny in recent years for various reasons, mainly concerning allegations of potential favorable bias toward certain specialties. A proper discussion of the controversy is beyond the scope of this essay, but it is easy to find lay media coverage of the RUC with a quick internet search. Regardless of what the RUC decides, though, there are occasional surprises coming out of left field, so to speak,⁴ because Medicare agency administrators or congressional legislation can override any reimbursement decisions. One random example from a few years ago was the legislative mandate for Medicare to equalize the reimbursement for radiosurgery treatment delivery whether given with a linac or GammaKnife unit, a decision that was made entirely outside the usual mechanism.

An Artificial Thought Experiment

A few years ago, a bohemian young journalist pointed out in these pages that several factors should be considered in the discussion of the possible forthcoming impact(s) of AI on radiation oncology.⁵ First and foremost, the use of computers to think faster for us is not at all new. At least in upper middle- and high-income countries, it would be hard to find a radiation oncology clinic that is not already using powerful computational hardware to perform tasks of dose calculation and optimization that would be impossible to do by hand. Furthermore, the gradual evolution from 2D calcs to high-speed adaptive technology didn’t exactly send the professions of medical dosimetry and physics to the same pastures as 1960s telephone switchboard operators. If anything, it upped the ante for requisite skill levels in each of those domains.

Nevertheless, it remains possible that AI will speed up some of the more mundane, repetitive tasks that comprise the duller side of modern radiation oncology practice — contouring target and normal tissue volumes for treatment planning is the easiest example that comes to mind. Whether that makes anything more or less costly, though, will depend on several other factors. Will the new software and hardware involved be more expensive to procure and maintain than current treatment planning systems? What will be the upfront and ongoing impact on clinic operational efficiency? Can current team members be retrained on the new technology and/or will there be a need to create new staffing categories? What, if any, added quality assurance steps might be required to maintain safety and accuracy? To what extent will payer support enable implementation of potentially useful improvements in technology?

But let’s not be too dreary in our future vision of the impact of AI in radiation oncology. Let’s imagine that we really do find ourselves in a place and time where it is easier, faster and cheaper to deliver radiation treatments of all sorts. What would we have then? Lower barriers to care for rural areas in this country and lower-income countries around the world? Hmmm. That doesn’t sound so bad.