@adamrodmanmd
Physician, educator, historian, author, podcaster, researcher at Beth Israel Deaconess Medical Center and Harvard Medical School, host of histmed podcast Bedside Rounds, associate editor at NEJM AI, studies 🤖+🧠. 🖖🚲
Congrats man!!!
Agree completely, I think in many ways APP oversight debates are most relevant here (though a lot of nuance when it comes to hand off).
yes agreed completely!!! I'm working on it :D If you wanna move to DM I can let you know some of the stuff we're working on (BIDMC is the home of MIMIC, and we continue to be up to some cool stuff!)
These are all very exciting ideas, but the evidence isn't there yet and there's a lot of foundational work to be done (both computationally, and as you point out, in clinical workflows and HCI)
One of the explicit goals of ARPA-H's ADVOCATE is to encourage development (and regulatory clearance) of such oversight systems (arpa-h.gov/explore-fund...)
... but that work is in its infancy. Similar exciting work around early decompensation but that's even earlier.
This is a great question! I do not think that GenAI CDS like OpenEvidence meaningfully changes anything (other than maybe a little more efficiency, which TBH is why it has caught on so much). There's been some interesting work with sequential e triggers (rules-based, then LLM) ...
In the next year, we're going to see more experiments like the Utah Rx refill one with Doctronic. How quickly this scales is largely going to be driven by how adventurous state regulators are going to be (and how good the data is).
Scaling in the short term is through the Lotus Health HITL model (which has LOTS of limits); human "overseer" as liability sponge. K Health is an example of a safe version of this that's already been deployed in health systems (AI as telehealth urgent care screen) including here in Boston.
... and what regulatory systems might look like. HHS/FDA is working on new regulatory pathways. But healthcare moves (often appropriately) slow, and I can't meaningfully predict beyond a year or two.
I think we're likely to get an overlap of dyadic and triadic care -- digital health DTC (like Hims/Hers or Cerebral) predate gen AI, but will likely start scaling more with the tech (a la Lotus). One of the unanswered questions is how GOOD DTC dyadic care will get in the short term ...
My perspective is colored by my research of course (clinical trial will be releasing soon). IMO what is meaningfully different about genAI from previous CDS tools is the ability to act at least somewhat autonomously (Babylon was the first at scale, and prev NLP DDSS like Isabel were getting close)
The only hope for our field is triadic care, where AI is deployed in such a way that maintains -- or even enhances -- human relationships. Otherwise, "AI doctor" startups will continue to chip away and further fragment our already struggling primary care.
New guest essay in the @nytimes.com -- patients (and doctors) need to be open with each other about our AI use. It's here, it's useful (with drawbacks). If we put our heads in the sand, we get the dyadic "AI doctor" that cuts human relationships out of medicine
In ranking the problems with medicine in the 🇺🇸; i wouldn't rank this in the top 10k.
www.acpjournals.org/doi/10.7326/...
In med school I bought a $15 noise machine to drown out the sirens and helicopters of Washington Heights.
It worked. Or at least I thought it did.
A new study suggests it may have been making things worse. 🧵 (1/13)
My latest Substack on AI Use and Depression, discussing the latest findings in a worrying JAMA Open study.
open.substack.com/pub/bleaseon...
On tonight's All Things Considered on NPR, I discuss patients' use of AI for health. While there are risks, many patients benefit.
Overall, I favor tools that guide patients through their symptoms (with iterative Q's & A's) rather than general tools like GPT or Gemini.
www.npr.org/transcripts/...
Incredibly proud of my colleagues Peter Brodeur, Liam McCoy, and Ethan Goh for all the amazing work they put into the inaugural State of AI Report. I'll try to get my thoughts into some sort of thread -- though it's VERY hard to keep up these days!
www.arise-ai.org/report
My energy level the second week in January.
(it's for a talk in two days, I know that's not today!)
If you are at MGH tomorrow and want to see me try a mystery case, please come to the noon clinicopathological case conference! The CPCs are, IMO, the most important case series since the Hippocratic Epidemics, and it's the honor of a lifetime to flail through one publicly!
Moral crumple zone is a good one too: (estsjournal.org/index.php/es...)
Screenshot of OpenEvidence investment prospectus, noting revenue of $50,000,000 with valuation of $6,100,000,000.
OpenEvidence has revenue of $50 million.
But investors value it at $6.1 BILLION.
So… a couple of questions for the “AI won’t replace doctors!” crowd:
What do you think OE’s long-term monetization pathway looks like?
And what do investors expect to happen that could justify this valuation?
As I'm writing this, I realize I'm probably not making anyone feel better😀
TL;DR: I largely agree with your premise, but it's going to take longer, and have much more slow and incremental change (and then fast and epistemic change) than the tech bros think
... since most of the proposed regulatory systems rely on tech that doesn't really exist yet.
There's also the regulatory angle. I think that SaMD absolutists are delusional (and probably not talking to anyone in the government), but companies are going to need some sort of regulatory structure in order to make investments in this space. That's a problem (right now) too...
... but I see no signs that OE is doing that, and I don't think that's happening in the short term (which, I mean, is like 5 years)
I think it's much more likely that OE wants to become the indispensable tool for doctors, and leverage the massive amount of data that they collect on us.
I also think that means that eventually, an OE API might become part of an evidence retrieval agent for a semi-autonomous HITL->HOTL system...
But that's the *easy* part. Building out workflows, developing and testing oversight systems, and then deploying and iterating -- that is a MUCH larger challenge. If OE wants to do that, they've made no signs in their hiring, or any acquisitions.