3 ms·

Right now, the only products I've seen leveraging LLMs are the tools from Nuance[1] that fill in physician notes. Arguably this is not an improvement; physician

Right now, the only products I've seen leveraging LLMs are the tools from Nuance[1] that fill in physician notes. Arguably this is not an improvement; physician notes are already an incomprehensible garble of cut-and-paste content (often from other systems).[2] But at least it is unlikely to negatively impact patient care. So far, AFAICT, the drive is to cut costs by saving the time the physician spends with the EHR. This has been a goal of the EHR systems for as long as they've existed as it saves money. There is a lot of press releases about products that do more but it's not at all clear to me that these products being widely used. The UnitedHealthcare example in this article is interesting in that is was being used in production. OTOH, it sounds much more like something evaluating business rules or some heuristics rather than an LLM or even an ML based solution. [1]: https://blogs.microsoft.com/blog/2023/08/22/microsoft-and-epic-expand-ai-collaboration-to-accelerate-generative-ais-impact-in-healthcare-addressing-the-industrys-most-pressing-needs/ https://blogs.microsoft.com/blog/2023/08/22/microsoft-and-ep... [2]: https://ehrintelligence.com/news/ehr-clinical-note-length-continues-to-grow-despite-coding-changes https://ehrintelligence.com/news/ehr-clinical-note-length-co...

to11mtm 3y ago
> The UnitedHealthcare example in this article is interesting in that is was being used in production. OTOH, it sounds much more like something evaluating business rules or some heuristics rather than an LLM or even an ML based solution. I'm fairly certain that at least one insurance company VP out there is using the LLM/ML hype to trumpet what is actually a statistical model evaluation. OTOH if it provides cost savings does the business truly care what it is? Especially if they can spin it as 'The computer makes more informed decisions thanks to AI'.
- tcbawo 3y ago
  It helps keep doctors treating patients instead of bogged down in documentation and bureaucracy. More time for patients and less stressed-out doctors means better outcomes.
MandolieJhay 3y ago
[dead]
EdwardDiego 3y ago
My employer is exploring LLMs in our existing product for transcription purposes in a clinical setting, especially telehealth. It's pretty damn impressive tbh.
bradknowles 3y ago
Can we be sure that those intermediated notes aren't just hallucinating what it thinks the doctor wants to write down? How do you verify that what gets written down really is what the doctor wanted? How do you automate that to protect yourself against LLM hallucination?
scrubs 3y ago
"So far, AFAICT, the drive is to cut costs by saving the time the physician spends..." In mental health a lot of providers don't even file on customer's insurance. Patient pays cash/charge and the patient is stuck with filing claims. Providers washed their hands of that dirty work. During my last job (large firm large metropolitan city) employer subcontracted for covid testing erecting a large tent to handle 8 covid exams at once. Here's the procedure: - you badge in; subcontractor used our hw/network. - you them give them your name and DOB, which the PC already told them - they write your name on specimen collection tube, and send you to bay X. The nurse at bay X sees you coming - at the bay you are asked for your name and dob again. And you badge in again. That level of duplication can only arise as a function of billing my then employer for extra non productive labor or fraud cya protection. Either way it's damn annoying. At my last doc appointment I spent the first 20 mins answering THEIR BILLING questions: was my name xyz? Did my insurance change? Etc. No services have been rendered yet. And despite all the IT for billing and PC displays displaying common sense health tips, there was no display of how long one would wait for the doc. It was a blind date: maybe they'd show; maybe you had better things to do. In sum, whatever crappy IT health care has, it's to help them, not customers. God damn: they made the paper work hell they live in and, as I say, mental health care patients are stuck with the dirty work because they can't stand it anymore. My wife spends 2hrs a week yelling at aetena. She's pretty sure they decline claims out of willful stupidity or nefarious greed. I would not be surprised to hear crappy AI involved. I absolutely hate that business. I was solicited multiple times for interviews by health insurance corps. I told them NFW. Their pay sucks. Their concept of quality improvement sucks. They suck.

to11mtm 3y ago

> The UnitedHealthcare example in this article is interesting in that is was being used in production. OTOH, it sounds much more like something evaluating business rules or some heuristics rather than an LLM or even an ML based solution. I'm fairly certain that at least one insurance company VP out there is using the LLM/ML hype to trumpet what is actually a statistical model evaluation. OTOH if it provides cost savings does the business truly care what it is? Especially if they can spin it as 'The computer makes more informed decisions thanks to AI'.

tcbawo 3y ago
It helps keep doctors treating patients instead of bogged down in documentation and bureaucracy. More time for patients and less stressed-out doctors means better outcomes.

MandolieJhay 3y ago

[dead]

EdwardDiego 3y ago

My employer is exploring LLMs in our existing product for transcription purposes in a clinical setting, especially telehealth. It's pretty damn impressive tbh.

bradknowles 3y ago

Can we be sure that those intermediated notes aren't just hallucinating what it thinks the doctor wants to write down? How do you verify that what gets written down really is what the doctor wanted? How do you automate that to protect yourself against LLM hallucination?

scrubs 3y ago

"So far, AFAICT, the drive is to cut costs by saving the time the physician spends..." In mental health a lot of providers don't even file on customer's insurance. Patient pays cash/charge and the patient is stuck with filing claims. Providers washed their hands of that dirty work. During my last job (large firm large metropolitan city) employer subcontracted for covid testing erecting a large tent to handle 8 covid exams at once. Here's the procedure: - you badge in; subcontractor used our hw/network. - you them give them your name and DOB, which the PC already told them - they write your name on specimen collection tube, and send you to bay X. The nurse at bay X sees you coming - at the bay you are asked for your name and dob again. And you badge in again. That level of duplication can only arise as a function of billing my then employer for extra non productive labor or fraud cya protection. Either way it's damn annoying. At my last doc appointment I spent the first 20 mins answering THEIR BILLING questions: was my name xyz? Did my insurance change? Etc. No services have been rendered yet. And despite all the IT for billing and PC displays displaying common sense health tips, there was no display of how long one would wait for the doc. It was a blind date: maybe they'd show; maybe you had better things to do. In sum, whatever crappy IT health care has, it's to help them, not customers. God damn: they made the paper work hell they live in and, as I say, mental health care patients are stuck with the dirty work because they can't stand it anymore. My wife spends 2hrs a week yelling at aetena. She's pretty sure they decline claims out of willful stupidity or nefarious greed. I would not be surprised to hear crappy AI involved. I absolutely hate that business. I was solicited multiple times for interviews by health insurance corps. I told them NFW. Their pay sucks. Their concept of quality improvement sucks. They suck.