AI and Documentation

[xrt123]

---

Members don't see this ad.

Anyone using AI to help extract information for documentation purposes? There has to be uses for creating end of treatment summaries, procedure notes that goes beyond templates used in the EMR. Just wondering if anyone is seeing some improved efficiencies.

 

[grenz]

Anyone using AI to help extract information for documentation purposes? There has to be uses for creating end of treatment summaries, procedure notes that goes beyond templates used in the EMR. Just wondering if anyone is seeing some improved efficiencies.

If it’s not here yet, then it’s just a matter of time. I think the real question is whether a company trains it on oncology specifically, or if it’s a general purpose AI they market to all specialties. The latter might not perform as well.

 

[randomname2018]

Yes. Abridge/Nuance to generate notes. Epic has chart summary tools in development. ChatGPT sandboxes for peer-to-peer appeals.

 

[grenz]

Epic has chart summary tools in development.

This is what I’m referring to. If Epic is training it on all physician notes, it might not do well for onc specifically. I hope they have specialty-specific models for generating summaries so it doesn’t end up bloated with extraneous information.

 

[Gfunk6]

ChatGPT sandboxes for peer-to-peer appeals.

Could you explain this a bit more?

 

[randomname2018]

This is what I’m referring to. If Epic is training it on all physician notes, it might not do well for onc specifically. I hope they have specialty-specific models for generating summaries so it doesn’t end up bloated with extraneous information.

Epic did. However, intial product had a limited word count look back. So if you had these massive bloated garbage notes, its summary wasn't great. Word count look back has been increased.

 

[randomname2018]

Could you explain this a bit more?

ChatGPT will compose letters for you with the appropriate prompt. Unless you have a local HIPAA-compliant version, you should not enter patient information into the public ChatGPT. You can say something like "write an insurance appeal letter for IMRT in a patient with left sided breast cancer". It will give you a generic letter, which generally isn't bad. If nothing else its a reasonable starting draft.

I do all letters this way now.

 

[seper]

ChatGPT will compose letters for you with the appropriate prompt. Unless you have a local HIPAA-compliant version, you should not enter patient information into the public ChatGPT. You can say something like "write an insurance appeal letter for IMRT in a patient with left sided breast cancer". It will give you a generic letter, which generally isn't bad. If nothing else its a reasonable starting draft.

I do all letters this way now.

some insurance appeals are a formality, some are are actually read and evaluated by radiation oncologists. AI-generated letters are easy to spot and thus often are a detriment to the patient

 

[OTN]

It only takes me 20-30 seconds to dictate a follow up, 40-60 seconds for a consult. Can be done while walking between patients, and I don't have to spend time reviewing them. Very tough bar for AI to clear.

Letters to insurance take me 30-45 seconds to type to our team. Also tough for AI to add much.

If I were early in my career there may have been efficiencies to gain, but not now.

 

[communitydoc13]

It only takes me 20-30 seconds to dictate a follow up, 40-60 seconds for a consult. Can be done while walking between patients, and I don't have to spend time reviewing them. Very tough bar for AI to clear.

Letters to insurance take me 30-45 seconds to type to our team. Also tough for AI to add much.

If I were early in my career there may have been efficiencies to gain, but not now.

WTF!

 

[BobbyHeenan]

It only takes me 20-30 seconds to dictate a follow up, 40-60 seconds for a consult. Can be done while walking between patients, and I don't have to spend time reviewing them. Very tough bar for AI to clear.

Letters to insurance take me 30-45 seconds to type to our team. Also tough for AI to add much.

If I were early in my career there may have been efficiencies to gain, but not now.

Yes, i'm still convinced old school dictation if you can get in a groove is the fastest. I'm closer to 90 seconds fo ra follow up, 3 minutes for consult though.

I can say "copy and past section primary diagnosis and previous treatment from my last note. next section interval history and start giving the interval history."

Unfortunately administration wants to get rid of our transcriptionists for speak to text .

 

[medgator]

It only takes me 20-30 seconds to dictate a follow up, 40-60 seconds for a consult. Can be done while walking between patients, and I don't have to spend time reviewing them. Very tough bar for AI to clear.

Letters to insurance take me 30-45 seconds to type to our team. Also tough for AI to add much.

If I were early in my career there may have been efficiencies to gain, but not now.

60s for a consult between pt rooms. Impressive

 

[OTN]

 

[BobbyHeenan]

Haha.

That’s what my staff says too “I can’t believe anyone can talk that fast.”

 

[taserlaser]

Yes, i'm still convinced old school dictation if you can get in a groove is the fastest. I'm closer to 90 seconds fo ra follow up, 3 minutes for consult though.

I can say "copy and past section primary diagnosis and previous treatment from my last note. next section interval history and start giving the interval history."

Unfortunately administration wants to get rid of our transcriptionists for speak to text .

They are closing ours in two weeks, 2 years after our transition to epic. I have my epic stats compared to the rest of the dept and I estimate I save at least one hour a week. To say I am upset is an understatement. Have been looking at AI solutions on the side but hard with device side processing for privacy.

Old school dictation with a transcriptionist is the most accurate and efficient imo for routing, formatting, everything. Especially if you have a diverse practice (includes mine) as I assume OTN has. A homogeneous practice in a single or two tumor group(s) such as breast or GU may stand to benefit from templates notes, but I am much less a fan for various reasons. Alas

 

[Bequerel]

The hospital got rid of our transcriptionist service when went to Epic 6 years ago. It’s a pia to transition, but if you’re efficient one way then you’ll figure out how to be efficient the new way with smart phrases and dragon. Being efficient is the name of the game. Amazes me how this varies so much between physicians!

 

[randomname2018]

some insurance appeals are a formality, some are are actually read and evaluated by radiation oncologists. AI-generated letters are easy to spot and thus often are a detriment to the patient

If you are good with prompt engineering, you won't have to edit and it looks like any other letter. I've done this dozens of times, seems to work just fine.

 

[taserlaser]

If you are good with prompt engineering, you won't have to edit and it looks like any other letter. I've done this dozens of times, seems to work just fine.

Any suggested prompts?

 

[medgator]

The hospital got rid of our transcriptionist service when went to Epic 6 years ago. It’s a pia to transition, but if you’re efficient one way then you’ll figure out how to be efficient the new way with smart phrases and dragon. Being efficient is the name of the game. Amazes me how this varies so much between physicians!

@OTN @BobbyHeenan you guys still have old school transcription. Very nice. I'm definitely slower now that we have dragon, templates help but still not the same, agree

 

[OTN]

@OTN @BobbyHeenan you guys still have old school transcription. Very nice. I'm definitely slower now that we have dragon, templates help but still not the same, agree

Yep, old school transcription is the best

 

[evilbooyaa]

Old school transcription definitely the best. EPIC is second.

 

[yesmaster]

AI costs too much and is too error prone.

Go take somebody else’s job Sam Altman!