Speaking of the physician who extensively uses Opus and Sol (through the appropriate HIPAA channels) I do worry about overutilization of resources. Both tools regularly recommend expensive testing that may not be clinically indicated. This consumes a finite pool of resources (we keep getting more patients, but we're not getting more MRI machines)
The underlying issue is that these models tend to shotgun orders because, at their core, they want more data. More data isn't always better for the patient: it generates incidentalomas, follow-up imaging, and downstream workups etc that carry their own risks and costs.
I do like the agent triaging in basket messages, though. That's pretty clever. Where it breaks down is when things get super specialized. This works well for primary care, etc. Maybe not in your ultra/hyper specialized setting.
> This consumes a finite pool of resources (we keep getting more patients, but we're not getting more MRI machines)
I don’t see this argument’s strength. Either the MR is indicated in a particular workup and the physician didn’t realize this or the MR isn’t indicated and won’t be performed (the physician will review the AI recommendation, the radiologist will refuse the study, or the LLM used by the payor will reject the prior auth).
The argument’s natural extent is that we should not do any more testing than we currently do if we begin using LLMs, which has the problem I outline above.
APP/NP/PA’s often order more imaging (there are studies to this effect) but this is potentially due to their education not focusing on when tests should not be pursued as opposed to carrying out what they’re asked to do by physicians. This is the setting where Opus/Claude may drive higher utilization (subject to the gates I highlighted earlier).
> Healthcare is fragmented, and teams often need to move information between systems manually before a patient can get care
This actually is the main issue, but the cause is partly EPIC and partly the federated nature of health care, where each practice (even within a health system) can have different procedure for qualifying and submitting referrals.
But their use-case of AI fielding emails or listening and summarizing is a different issue: providers are being forced on pain of losing their jobs to cut the time they take with patients.
I can see how AI could relieve pressure in both cases, but that would fix nothing. If it's easier to complete referrals, referrals will become more complex, because that's how they manage their workload. If providers can be force into 20-minute visits, they'll be forced into 10-minute visits via triage.
When AI is deployed to improve quality and not efficiency, it can only make things worse.
When it's deployed for quality, it can actually help - injecting an additional test for a low-traffic diagnosis, walking generalists through complex diagnosis algorithms, etc.
All I can see happening with AI is squeezing more patients through the same broken system, with even less human face time than they get today.
OpenAI is not here to fix healthcare, their interest is in making money off of it, and having a process that nobody understands or could manually execute is a means to that end.
The problem isn't so much that the AI summarizes, but that the summary doesn't cite the source. The correct approach would be for the phrase in the draft to link to the original document; for example, mentioning that it's a note from a specific date, or that the image is from a particular source, so the doctor can quickly verify what really matters.
You're just as well, because if they did they would probably suggest surgery, which is absolutely the last thing you want.
If you're overweight, fix that. Then go see a DPT, get a rehab plan, and actually do it. There's no magic path to a strong back, you need to do the work.
Probably not. I've worked in US health tech and done many experiments integrating OpenCV and LLMs into products. Clinicians, i.e., licensed medical providers, are extremely sensitive to your products having data and information bugs. If you have even one discrepancy to what their EHR says, you irrecoverably lose trust and your contract will not be renewed the following year. Very little AI has gotten it's way into the US healthcare space because HHS actually fines and puts people in prison. I've poached many engineers from companies HHS investigates, which for all intents and purposes is a corporate death penalty.
> HHS actually fines and puts people in prison. I've poached many engineers from companies HHS investigates, which for all intents and purposes is a corporate death penalty.
Hopefully the current admin reforms this kind of unwarranted behavior. New medical advances should only be tested in real-world, large-scale, nationwide deployments to avoid confounding variables. The time of lab-scientist dictatorship, controlling who gets access to the latest and greatest treatments, is over.
Umm, I am not sure this is the flex they think it is. 99 % of tasks reviewed by a clinician were rated safe. That means 1 in a 100 patients will get hurt. That's nowhere near safe enough.
>> A patient messages their clinician about lower back pain. A computer-use agent reviews the patient’s visit notes, physical therapy records, and prior imaging in the EHR. It drafts an MRI order, an encounter note, and a reply to the patient for the clinician to review. The agent then uses the same records to prepare a prior authorization request in the insurer’s portal for an authorization specialist to review.
Um... can I please have the clinician review my file? Having the AI review my file and present a summary to the clinician/doctor means no human has looked at my actual history, only summaries provided by the AI. This is exactly how things get missed. It is exactly why doctors look at charts no matter what they are told verbally about a patient. Relying on AI to do all the chart reading is asking for trouble.
This could be wishful thinking, but OAI may just be talking about creating the prior auth. If this is the case, it would be nice if they said "prior auth encounter note" instead.
From my limited understanding, the physician would and should still review your message, chart, agent drafted PA, etc...
They mention in the video that a clinician and/or authorization specialist will review it before it's submitted. The AI is just preparing a draft. In the video, you can see a human editing the notes in the interface, after the AI has prepared it all.
It's the same thing as AI coding. The AI agent writes the code, but the coder has to review the code and edit it or propose changes if it's wrong.
The point of both of them is to reduce manual work, and allow for less-specialized work, which saves time and money and can reduce errors.
> The agent then uses the same records to prepare a prior authorization request in the insurer’s portal for an authorization specialist to review.
Did you mean: The agent then uses the same records to prepare a prior authorization request in the insurer’s portal for an authorization agent to review.
Two agent fight, hope your agent stronger to get MRI
I'm grateful to see this progress in this area. So many human errors with the current system.
Speaking of the physician who extensively uses Opus and Sol (through the appropriate HIPAA channels) I do worry about overutilization of resources. Both tools regularly recommend expensive testing that may not be clinically indicated. This consumes a finite pool of resources (we keep getting more patients, but we're not getting more MRI machines)
The underlying issue is that these models tend to shotgun orders because, at their core, they want more data. More data isn't always better for the patient: it generates incidentalomas, follow-up imaging, and downstream workups etc that carry their own risks and costs.
I do like the agent triaging in basket messages, though. That's pretty clever. Where it breaks down is when things get super specialized. This works well for primary care, etc. Maybe not in your ultra/hyper specialized setting.
> This consumes a finite pool of resources (we keep getting more patients, but we're not getting more MRI machines)
I don’t see this argument’s strength. Either the MR is indicated in a particular workup and the physician didn’t realize this or the MR isn’t indicated and won’t be performed (the physician will review the AI recommendation, the radiologist will refuse the study, or the LLM used by the payor will reject the prior auth).
The argument’s natural extent is that we should not do any more testing than we currently do if we begin using LLMs, which has the problem I outline above.
APP/NP/PA’s often order more imaging (there are studies to this effect) but this is potentially due to their education not focusing on when tests should not be pursued as opposed to carrying out what they’re asked to do by physicians. This is the setting where Opus/Claude may drive higher utilization (subject to the gates I highlighted earlier).
it's like a terrified first year resident!
> Healthcare is fragmented, and teams often need to move information between systems manually before a patient can get care
This actually is the main issue, but the cause is partly EPIC and partly the federated nature of health care, where each practice (even within a health system) can have different procedure for qualifying and submitting referrals.
But their use-case of AI fielding emails or listening and summarizing is a different issue: providers are being forced on pain of losing their jobs to cut the time they take with patients.
I can see how AI could relieve pressure in both cases, but that would fix nothing. If it's easier to complete referrals, referrals will become more complex, because that's how they manage their workload. If providers can be force into 20-minute visits, they'll be forced into 10-minute visits via triage.
When AI is deployed to improve quality and not efficiency, it can only make things worse.
When it's deployed for quality, it can actually help - injecting an additional test for a low-traffic diagnosis, walking generalists through complex diagnosis algorithms, etc.
All I can see happening with AI is squeezing more patients through the same broken system, with even less human face time than they get today.
OpenAI is not here to fix healthcare, their interest is in making money off of it, and having a process that nobody understands or could manually execute is a means to that end.
> All I can see happening with AI is squeezing more patients through the same broken system
While introducing subtle mistakes doctors may miss on review.
Any problems and they'll blame a "rogue agent"
The best thing for AI is for it to file a bunch of antitrust suits against Big Insurance, Big Pharma, and the Big Hospital/Healthcare systems
The problem isn't so much that the AI summarizes, but that the summary doesn't cite the source. The correct approach would be for the phrase in the draft to link to the original document; for example, mentioning that it's a note from a specific date, or that the image is from a particular source, so the doctor can quickly verify what really matters.
> The problem isn't so much that the AI summarizes, but that the summary doesn't cite the source
That is in fact exactly how the current products work - they link to each source document.
> lower back pain
Lol in 20 years no Dr ever has taken my back pain seriously. This ai is already delivering better care!
You're just as well, because if they did they would probably suggest surgery, which is absolutely the last thing you want.
If you're overweight, fix that. Then go see a DPT, get a rehab plan, and actually do it. There's no magic path to a strong back, you need to do the work.
Best part is that we are a hair away from some middleman reselling this to hospitals to offload liability and selling access to health insurance orgs.
OAI and ANT really are leading the charge to pull us into the worst timeline.
Probably not. I've worked in US health tech and done many experiments integrating OpenCV and LLMs into products. Clinicians, i.e., licensed medical providers, are extremely sensitive to your products having data and information bugs. If you have even one discrepancy to what their EHR says, you irrecoverably lose trust and your contract will not be renewed the following year. Very little AI has gotten it's way into the US healthcare space because HHS actually fines and puts people in prison. I've poached many engineers from companies HHS investigates, which for all intents and purposes is a corporate death penalty.
I don't know that Clinicians have much of a say, this is all B2B selling the idea that we replace people doing these jobs.
> HHS actually fines and puts people in prison. I've poached many engineers from companies HHS investigates, which for all intents and purposes is a corporate death penalty.
Hopefully the current admin reforms this kind of unwarranted behavior. New medical advances should only be tested in real-world, large-scale, nationwide deployments to avoid confounding variables. The time of lab-scientist dictatorship, controlling who gets access to the latest and greatest treatments, is over.
Umm, I am not sure this is the flex they think it is. 99 % of tasks reviewed by a clinician were rated safe. That means 1 in a 100 patients will get hurt. That's nowhere near safe enough.
... it seems entirely unclear that human clinicians have a correctness rate of >99%
>> A patient messages their clinician about lower back pain. A computer-use agent reviews the patient’s visit notes, physical therapy records, and prior imaging in the EHR. It drafts an MRI order, an encounter note, and a reply to the patient for the clinician to review. The agent then uses the same records to prepare a prior authorization request in the insurer’s portal for an authorization specialist to review.
Um... can I please have the clinician review my file? Having the AI review my file and present a summary to the clinician/doctor means no human has looked at my actual history, only summaries provided by the AI. This is exactly how things get missed. It is exactly why doctors look at charts no matter what they are told verbally about a patient. Relying on AI to do all the chart reading is asking for trouble.
This could be wishful thinking, but OAI may just be talking about creating the prior auth. If this is the case, it would be nice if they said "prior auth encounter note" instead.
From my limited understanding, the physician would and should still review your message, chart, agent drafted PA, etc...
They mention in the video that a clinician and/or authorization specialist will review it before it's submitted. The AI is just preparing a draft. In the video, you can see a human editing the notes in the interface, after the AI has prepared it all.
It's the same thing as AI coding. The AI agent writes the code, but the coder has to review the code and edit it or propose changes if it's wrong.
The point of both of them is to reduce manual work, and allow for less-specialized work, which saves time and money and can reduce errors.
How long until they are pressured to stop reviewing just like people are starting to do with code?
I bet not very long.
> The agent then uses the same records to prepare a prior authorization request in the insurer’s portal for an authorization specialist to review.
Did you mean: The agent then uses the same records to prepare a prior authorization request in the insurer’s portal for an authorization agent to review.
Two agent fight, hope your agent stronger to get MRI