Approaching Singularity, Part IV: Guardrails or Vision?
Pharma is Betting Trillions on Curing Cancer. What is Radiation Oncology's Vision?
Vision First:
When you see an AI story as it relates to our field - ask yourself:
Is this a vision piece?
Or is it just small stuff?
Does it boldly push forward?
Or encourage a current lane?
Today, I’ll argue we need a higher arching vision for our field relating to AI, advocating for our value, and voicing our role as leaders in medical technology.
If we don’t, or if we try to slow the pace of the events around us - I see us never quite reaching our potential.
This piece will examine how I see us approaching AI and how people within our field are describing the impact it will have on our specialty. Maybe it gives perspective - maybe some see this series as distraction from clinical data. But the goal for the site remains pretty constant - write things I hope will hold up in five years.
The Approaching Singularity: Part I
We set the stage - two converging singularities on a collision course. It’s a large picture view of our field illustrating the long arc of progress in radiation is towards a simple concept: Put more dose in the target and put less dose elsewhere - our Singularity so to speak. Today, that long directional arc is going to collide with an explosion in AI investment.
The Approaching Singularity: Part II
We look at more specific examples and what this intersection might look like. We look at theranostics (Pluvicto), the new Reflexion treatment codes (billable at $3750 per fraction), CMS 2026 billing changes and consider how these “outside” issues pair with our own field’s “Singularity.”
The Approaching Singularity: Part III
We look at ways for the field to prepare for rapid change. We discuss items that create diversity in your career arc - so that, when change happens, you are more able to adapt. Importantly, this preparation step allows you to embrace AI and aim for a new vision.
In the past two weeks, I’ve seen two leading voices talk about AI and how it is likely to interact with our field. One was focused on more of a laundry list of possible applications, and then, the subsequent medical legal implications - discussing the need to have very structured approaches - layers of compliance, regulations, laws, and protocols (it was quite depressing if I’m being honest). The second was a practical explanation of where AI can be seen today and where it might be - say in the next year or two seeing more clinical exposure - a very traditional “radiation oncology” view of what is occurring but focusing well within the walls of our field.
From my perspective, each missed on vision - and more than ever, I believe we need some level of vision.
This isn’t completely unexpected. In fact, this is consistent with our field at large. We are conservative and measured. Our field, as a whole, has moved slower than every other specialty I have close contact with. We ascribe it to being data-driven, but it is a step more than that - after all - I don’t think many specialties would consider themselves “not data driven”. It is inherently cautious. For whatever reason, we are overly (my term) careful - to the point of struggling to argue for our value.
2026 Code Changes: Just one example - but I could build a list.
Look at our response regarding the payment changes (announced in July 2025). It has taken about nine months for us to even acknowledge the impact of CMS changes. For months we ran with “1% impact” - many now understand how wrong that was. Even today, many are arguing for restraint in interpretations of what defines “complex” vs.
”intermediate” treatment.
It makes little sense to me why we choose this path - it is as if we thought it wise to not ruffle feathers - again, the cautious path.
(Note: I think the code as written today is ambiguous - of course, you must bill appropriate codes for services rendered, but these items are often quite gray - far too large of a topic for today.)
So here with AI - our response is expected. AI is brand new. AI operates much more like a “black box” than many people are comfortable with. And small pieces and specific examples are easier to present and explain that larger conceptual ideas. Add that to the fact that change is disruptive and “scary” and my money for our field would be to approach this wall carefully.
And yet, for all of its issues, it’s so powerful, we can’t afford not to use it.
We ARE Technology:
We are a technology based field - measurably and demonstrably the most heavy CapEx field in all of medicine and it isn’t close (our practice overhead runs 70%-75% vs a more typical 45%-50% level in other “high capex” specialties). In that position, we must really work hard to lead in implementation of AI in the clinic. Everything we do today is computer driven. We have CTs. We have plans and delivered dose - all digital and all in our charts. We have QA scans showing progress during the course of treatment and we have a ton of imaging data and labs both pre and post relatively short courses of treatment.
We sit on a gold mine of data and we are creating more gold daily - if we use it.
Our window for this response data gap we hold is shrinking by the day. ctDNA testing is grabbing hold. You are beginning to see early studies showing correlations between - what is a lab draw - and outcomes across a number of disease sites (Refs: 1-Lung,2-Renal,3-NCSLC). Within 10 years, I will not be surprised to see ctDNA equivalent to, or surpassing imaging response predictive value.
Like I said - today we have a measurable gap lead in data response to treatment. That gap is almost certain to rapidly narrow with a real chance other specialties sprint ahead if we don’t utilize our inherent treatment data more efficiently. AI and compute is offering real opportunity for our specialty that few others have today.
A few years ago this may have been “too much data” - today that is not an excuse.
My Concern: Guardrails as Step One.
Here’s a quote from leaders in our field from the recent NEJM article looking at SBRT and the impact that treatment has had on our field. Below is a final line from the conclusion, that I think speaks to our field’s mindset:
“Yet radiation oncologists continue to prioritize avoidance of side effects altogether, a strategy that allows radiotherapy to retain its reputation as an effective and safe treatment for a broad scope of malignant conditions.”
Citrin and Timmerman - NEJM
Yet…
On my read a purposeful, subtle qualifier. I believe other leaders see what I am saying. This is how our field has operated for decades. Too measured and cautious. Even when there is tremendous opportunity to improve our impact, we choose the safer path - not my words, theirs.
Admittedly, “AI” represents new tools and we work in a highly regulated environment. What we do is incredibly important and errors simply can’t be tolerated. Thirty years into a clinical career and this has been my standard: it needs to be good enough for your family - your mom, dad, wife, kid - that is your standard for each and every case.
Too Focused: Just One Example: AI Scribes
AI scribes are a perfect example of misplaced focus. They dominate "AI in Oncology" posts despite being, at best, a minor efficiency play. If the tool saves you time, use it. Be HIPAA-compliant - as you should be with every action regardless. And if it's imperfect, fine - copy-paste slop isn't a high standard either, even if it never has a typo. The point is simple: whether you type your note or AI summarizes it is an efficiency discussion, not a "cure cancer" discussion. It doesn't move that needle.
Outside Our Walls:
For the first time in history, the pace of academic institutions can simply be bypassed. PhD work will continue, but it better move fast. Biotech companies are moving as much of drug development as they can onto silicon - bets sized in the hundreds of millions, backed by venture capital, with a simple mandate: iterate fast, break stuff, go again. And their goals reflect it. Cut development time to one-fifth of what it is today. Create drugs across all diseases that extend life and improve quality of it. In cancer - cut death rates in half. This is a meaningful shift: the largest pools of capital and compute in history are being aimed not at financing models or social media, but at curing disease.
Consider that two of the leading AI frontier models are led by individuals with neuroscience backgrounds - both “aiming” their institutions and trillions of dollars at biology, and medicine, and improving health through science. This will have real impact - again, they are not creating social networking sites - they are focused on using massive dollars and compute to solve disease.
In contrast to those outside our field, I see us taking the old road - the safe path. And I completely understand that we can’t have errors. But realistically, I’m also certain we can create so many guardrails within our specialty that these tools can’t do more than move the needle via “5% efficiency gains.”
We will not compete if our vision is limited.
Our Voice Today:
Look at the recent ASTROnews release on AI. We’ll focus here as they have a massive platform within our specialty.
The article is completely fine.
My guess is, they feel like it presented a “vision”, but from my perspective, it falls well short where we must aim. Clearly, this comes from within a large “regulated and structured” entity and I’m sure they purposefully wrote something “mainstream”. Maybe they believe this is a “required” stance in a highly regulated environment operating as a “non-profit professional association”. I truly understand all of those concerns.
But. This is our voice and, from where I sit, it errs quite badly on the side of measured caution.
Read two of the closing statements and really consider the goals as this entity does speak for our field.
The next chapter will be defined by how well we design AI-enriched workflows that keep patients safe, keep humans accountable, and keep quality measurable….
The forthcoming white paper will provide more detail, but the message for now is simple: when adopted intentionally and with the right safety guardrails, AI can help us deliver more consistent, efficient and patient-centered care.
Safe, accountable, measurable, intentional, consistent, efficient, “patient-centered care”.
This is OUR goal for AI. The only one with any hint of vision is the one that can’t be defined - “patient-centered”.
If the tumor is in the patient and the isocenter is in the tumor - it is, by definition “patient-centered.”
Meanwhile trillions of dollars are being poured into Biotech pharma companies to use AI to drive innovation with stated goals of “Solve all Disease” (Isomorphic Labs).
My guess is that type of language is too bold for us - but should it be?
Maybe buried beneath “efficiency” and better “safety”, improved outcomes might be implied, but wow… As a vision for AI within our field - on the heels of getting handed a specialty defining restructuring of our payment model - this misses.
In contrast to safe, accountable, measurable, intentional, consistent, efficient, I’d like to see us have vision:
AI is going to have massive impacts on medicine and healthcare. As a technology driven field we must lead. Our primary goal will be to use AI and technology to improve patient outcomes - higher cure rates, less toxicity. Primary initial drivers will be greater consistency, safer delivery, and improved efficiency to increase patient access. The longer term goal is an AI driven closed loop system that is autonomous, verifiable, and produces large scale measurable improvements over today’s outcomes. We will work relentlessly to make sure our specialty is the first in medicine to accomplish these lofty goals. We’ve got a ton of work to get done for our patients, but our position as THE technology leaders of medicine today gives us great opportunity - let’s get moving.
As I said previously, if we don’t move quickly and aim a bit higher:
Pharma will define our Singularity.
Which path do you believe leads our specialty to Better?
Which would you rather pour your Heart and Soul into?
Safe, accountable, intentional, efficient care,
or…
Solving Cancer.
As always, thoughts of one. Keep advocating for our value and keep pushing for Better.
ASTRO lacks vision you say? Well you did point out the near blindness of what has happened with reimbursement instead of the projected 1% cuts which were pushing right behind ROCR froth. Not sure how AI will help radiation therapy if our specialty is cut to the bone with the endless march to lower fractions.. FLASH poof and we're gone. For now, and for the next 5-10 years we remain viable. Rural radonc is desperate for help and along with locums to fill it pays well with grateful patients. But 'regular jobs in regular places' will become as fabled as neutrons but for a lucky few. Glad I got to experience the 'golden era' and can move around easily to work and live separately. Maybe zietman was right and leadership should have migrated to Interventional Oncology two decades ago.