physician-engineer. makes, breaks, and fixes healthcare AI
Joined October 2010
- Tweets2.2K
- Following367
- Followers522
- Likes7.6K
i agree 100%. this isn’t limited to research. it applies to innovation & improvement in general.
we’re so over capacity w/ the work of delivering care that we have no time to make it faster, safer, more effective. we spend all day working broken systems & have no time to fix
In American medicine, research is the price of admission but rarely a viable career. Trainees are told it's essential for competitive medical school and residency positions. As faculty, they confront scarce protected time, billing-driven productivity metrics, low grant-funding rates, long funding cycles, and salary-cap gaps their institutions are asked to absorb, just as family and financial obligations grow. Pragmatic clinical research is poorly supported, especially outside well-resourced centers. The incentives favor delivering costly care over improving it.
erkin ötleş retweeted
In American medicine, research is the price of admission but rarely a viable career. Trainees are told it's essential for competitive medical school and residency positions. As faculty, they confront scarce protected time, billing-driven productivity metrics, low grant-funding rates, long funding cycles, and salary-cap gaps their institutions are asked to absorb, just as family and financial obligations grow. Pragmatic clinical research is poorly supported, especially outside well-resourced centers. The incentives favor delivering costly care over improving it.
Coach is one way this goes. But some of us also need to become the technical leaders of healthcare AI, designing, testing, implementing, and evaluating these tools.
Actual study: arxiv.org/abs/2608.09861
Google proved it. Doctors are cooked.
I thought AI would need a photorealistic avatar with perfect microexpressions and conversational cadence before patients would prefer it over a telemedicine visit with a human.
Nope.
All the AI needed was a voice - no video avatar - and it was rated better than doctors by professional patients on assessing them and explaining what was going on.
And trained physician raters put it on par or better than docs across history-taking, diagnosis, management, and physical exam.
And did I mention that the visits can be anytime, anywhere, as often as necessary, as long as necessary and at negligible cost?
That the patient can ask as many questions as they want for as long as they want?
Or that it keeps their entire medical history in mind at all times and knows the entire literature?
So how do doctors adapt so we stay relevant?
We become the coach.
That’s the simplest analogy.
For the history of medicine, we’ve done it all - gathered the info (player), made the decisions (coach) and implemented the plan (player).
Now there’s a better player. A new breed. We’re obsolete.
But maybe, just maybe, we can still add value in making the decisions.
Or maybe it’s time to go to the broadcast booth.
for decades, health IT was bought by institutions and imposed on users. we absolutely need new governance. but perhaps this inversion is something we should celebrate, and protect?
Bypass of healthcare’s governance for AI @NEJM_AI ai.nejm.org/doi/full/10.1056… is a measure of the demand, of ripeness for disruption and the pent up reaction to the suppression of doctors’ agency. How it plays out determines if we are heading to EHR-clerk redux or something better. CC @kdpsinghlab
erkin ötleş retweeted
Bypass of healthcare’s governance for AI @NEJM_AI ai.nejm.org/doi/full/10.1056… is a measure of the demand, of ripeness for disruption and the pent up reaction to the suppression of doctors’ agency. How it plays out determines if we are heading to EHR-clerk redux or something better. CC @kdpsinghlab
Physician AI adoption went from 47 percent to 63 percent in about nine months, according to Doximitys 2026 survey.
No health system procurement committee approved that.
This is the most underrated fact in clinical AI. The adoption curve that actually happened ran around the institution rather than through it. Doctors signed up on their phones between patients, the same way they started using UpToDate years before anyone had a contract for it.
Medical education is now doing the same thing, less visibly. Faculty have a name for it: shadow AI. Students using models in ways nobody sanctioned, and educators worrying that reasoning which should be built is instead being rented. Both things are true at once. The tools genuinely accelerate learning, and they genuinely let you skip the part where you struggle.
Institutions still debating whether to allow this have already lost the adoption question. The only live question is whether they get any say in shaping what people were going to use regardless.
doximity.com/reports/state-o…
I chose the secret third option: keep the relationship, switch careers, and make my significant other endure my increasingly elaborate plans to bring AI to doctors.
erkin ötleş retweeted
@dmshanmugam & @DrMadhurNayan have put together a stellar pre-conference workshop at @mlforhc
Join us on August 12 at Hopkins! eventzilla.net/e/mlhc-precon…
Workshop registration is close to capacity. Join MLHC Aug 12 for a hands-on pre-conference day on clinical AI, LLMs in care, and medical imaging, with @AmerMedicalAssn, Claude Health @claudeai, and @nvidia sessions. Register: eventzilla.net/e/mlhc-precon… #MLHC2026
erkin ötleş retweeted
I’m hiring a postdoc to work with me on AI & health @ChicagoBooth! My research is *very* hands on — impact to me means getting things deployed and *working* in hospitals. Some areas I’m excited about (and application link) below ⬇️⬇️⬇️
erkin ötleş retweeted
Why is it ‘cancelled’ in the U.K. but ‘canceled’ in the U.S.?
Because we gave them that L in 1776.
erkin ötleş retweeted
🚨 The Supreme Court ruled that the Fourteenth Amendment guarantees birthright citizenship to children born in the United States, including those whose parents are in the country unlawfully or only temporarily, striking down President Trump's executive order.
