AI Tech for the ER

[DocEspana]

---

Members don't see this ad.

I dont use AI for notes, but I've been slowly transitioning from WikEM for my "how do I manage that again" questions and towards OpenEvidence. For those unfamiliar, it seems to be an AI trained on medical literature and does a very good (though sadly not perfect) job with management questions. Especially useful is that it identifies new literature or recommendations and integrates it - along with identifying that it is referring to more recent literature that may/may not be the current standard of care.

 

[EctopicFetus]

I dont use AI for notes, but I've been slowly transitioning from WikEM for my "how do I manage that again" questions and towards OpenEvidence. For those unfamiliar, it seems to be an AI trained on medical literature and does a very good (though sadly not perfect) job with management questions. Especially useful is that it identifies new literature or recommendations and integrates it - along with identifying that it is referring to more recent literature that may/may not be the current standard of care.

The other side benefit.. you get CME from Openevidence.

I do tend to agree that it is far from perfect. I am a little unsure how they are gonna monetize their company especially because they have raised a ton of money. I will admit I am a little disappointed with the quality of OE, even fairly straightforward questions it struggles to give strong answers. I was playing with it before and asked for management of a 30 year old female for Pyelo with allergies of x,y,z and it did not do a great job.

I think it does well with providing UptoDate type info but I still use Pepid for simple questions. It is bullet point answers and gets me what I need not the super long answer that OE tends to give.

 

[thegenius]

I think OE has a ways to go, TBH. I used it for questions where the answer is either "X" or "Y", and I can't remember which one it is, and I think I know, and OE gives me the answer.

It once recommended it give azithromycin 30 mg / kg x1 for bacterial enteritis.

It's good for general knowledge updates. Frankly, if UpToDate built a chat bot into their website that summarizes all of their info, I would use that instead.

 

[thegenius]

The other side benefit.. you get CME from Openevidence.

I do tend to agree that it is far from perfect. I am a little unsure how they are gonna monetize their company especially because they have raised a ton of money. I will admit I am a little disappointed with the quality of OE, even fairly straightforward questions it struggles to give strong answers. I was playing with it before and asked for management of a 30 year old female for Pyelo with allergies of x,y,z and it did not do a great job.

I think it does well with providing UptoDate type info but I still use Pepid for simple questions. It is bullet point answers and gets me what I need not the super long answer that OE tends to give.

What is Pepid, a website or an app?

 

[EctopicFetus]

What is Pepid, a website or an app?

They have an app but i just use the website.. im an old geezer that way

 

[TooMuchResearch]

I think OE has a ways to go, TBH. I used it for questions where the answer is either "X" or "Y", and I can't remember which one it is, and I think I know, and OE gives me the answer.

It once recommended it give azithromycin 30 mg / kg x1 for bacterial enteritis.

It's good for general knowledge updates. Frankly, if UpToDate built a chat bot into their website that summarizes all of their info, I would use that instead.

I just use UpToDate. People have tried to convince me that OE is better and I tend to disagree.

 

[DocEspana]

I think OE has a ways to go, TBH. I used it for questions where the answer is either "X" or "Y", and I can't remember which one it is, and I think I know, and OE gives me the answer.

It once recommended it give azithromycin 30 mg / kg x1 for bacterial enteritis.

It's good for general knowledge updates. Frankly, if UpToDate built a chat bot into their website that summarizes all of their info, I would use that instead.

My boss loves OE. Its why I even know it exists. He had a case of mine he ended up reviewing (thankfully nothing went wrong. The radiologist just called him up randomly to change the read on my patient after I left and the change was clinically irrelevant for their dispo) and decided to decide to just review my care on the guy. Said I should have gotten renal function and PT/PTT/INR on a person who went on eliquis. I told him that, yeah, I mean it probably would have been a pretty normal thing to do but the guy just came in to confirm a DVT that his PCP already said was there but *needed* it confirmed for some reason and he had bloodwork a month ago so I know his labs are not chronically weird. But he pushed back that OE said it is required and that Eliquis needs to be renally dosed and potentially adjusted for PT/PTT

and it does need to be renally dosed.... for afib. Not DVT. Only one accepted dosing for DVT. And I didn't have time to explain to him the nuance of PT/PTT not measuring bleeding the way he thinks it does, and that any value in the last year is probably sufficient to screen for genetic coagulopathy in someone who isnt banana yellow jaundiced. He showed me those OE searches he did. It quoted some ED blogs discussing their standard practice opinions. So I showed him the actual bristol-meyer-squibb education on what is needed before starting eliquis for DVT (hint: nothing except some confirmation of no genetic coagulopathy pre-existing).

I guess its the same as any other AI tool. It can really lose the nuance needed because it is mostly ingesting information from sources that mix "actual protocols" with "things we do a lot."

also: WikEM is the best.

 

[Mount Asclepius]

My boss loves OE. Its why I even know it exists. He had a case of mine he ended up reviewing (thankfully nothing went wrong. The radiologist just called him up randomly to change the read on my patient after I left and the change was clinically irrelevant for their dispo) and decided to decide to just review my care on the guy. Said I should have gotten renal function and PT/PTT/INR on a person who went on eliquis. I told him that, yeah, I mean it probably would have been a pretty normal thing to do but the guy just came in to confirm a DVT that his PCP already said was there but *needed* it confirmed for some reason and he had bloodwork a month ago so I know his labs are not chronically weird. But he pushed back that OE said it is required and that Eliquis needs to be renally dosed and potentially adjusted for PT/PTT

and it does need to be renally dosed.... for afib. Not DVT. Only one accepted dosing for DVT. And I didn't have time to explain to him the nuance of PT/PTT not measuring bleeding the way he thinks it does, and that any value in the last year is probably sufficient to screen for genetic coagulopathy in someone who isnt banana yellow jaundiced. He showed me those OE searches he did. It quoted some ED blogs discussing their standard practice opinions. So I showed him the actual bristol-meyer-squibb education on what is needed before starting eliquis for DVT (hint: nothing except some confirmation of no genetic coagulopathy pre-existing).

I guess its the same as any other AI tool. It can really lose the nuance needed because it is mostly ingesting information from sources that mix "actual protocols" with "things we do a lot."

also: WikEM is the best.

This is in the weeds of medicine. Good Directors should dismiss these reviews and not bog down line docs with having to defend themselves over trivial medicine.

 

[thegenius]

I just use UpToDate. People have tried to convince me that OE is better and I tend to disagree.

OE is not better than UTD for factual information. Not yet. OE is quicker is much quicker so it really depends on what one is asking.

OE is good for questions like “what is the prob of passing a 6 mm kidney stone” and you kind of know the answer, but it gives you a percentage range.

 

[southerndoc]

My boss loves OE. Its why I even know it exists. He had a case of mine he ended up reviewing (thankfully nothing went wrong. The radiologist just called him up randomly to change the read on my patient after I left and the change was clinically irrelevant for their dispo) and decided to decide to just review my care on the guy. Said I should have gotten renal function and PT/PTT/INR on a person who went on eliquis. I told him that, yeah, I mean it probably would have been a pretty normal thing to do but the guy just came in to confirm a DVT that his PCP already said was there but *needed* it confirmed for some reason and he had bloodwork a month ago so I know his labs are not chronically weird. But he pushed back that OE said it is required and that Eliquis needs to be renally dosed and potentially adjusted for PT/PTT

and it does need to be renally dosed.... for afib. Not DVT. Only one accepted dosing for DVT. And I didn't have time to explain to him the nuance of PT/PTT not measuring bleeding the way he thinks it does, and that any value in the last year is probably sufficient to screen for genetic coagulopathy in someone who isnt banana yellow jaundiced. He showed me those OE searches he did. It quoted some ED blogs discussing their standard practice opinions. So I showed him the actual bristol-meyer-squibb education on what is needed before starting eliquis for DVT (hint: nothing except some confirmation of no genetic coagulopathy pre-existing).

I guess its the same as any other AI tool. It can really lose the nuance needed because it is mostly ingesting information from sources that mix "actual protocols" with "things we do a lot."

also: WikEM is the best.

Wow. I check a POC creatinine and send them out. If they're elderly I'll check a CBC. Coags for someone already on Eliquis are useless. Elevated means they recently took it and that's all.

 

[sloh]

Frankly, if UpToDate built a chat bot into their website that summarizes all of their info, I would use that instead.

Coming soon.

• Wolters Kluwer launched “AI Labs” in 2023, an internal sandbox for experimenting with generative AI built specifically on UpToDate’s trusted medical content
• These tools aim to provide AI‑enhanced search and conversational-style responses directly informed by UpToDate’s vetted articles — not generic LLM output
• AI‑Enhanced Search: Live since late 2024, it lets clinicians ask natural‑language questions (“What’s the best treatment for pneumonia in elderly?”) and get succinct, UpToDate‑sourced answers, with links to full articles. It emphasizes speed, context, and transparency

 

[EctopicFetus]

Everyone is doing it. One of the AI scribe programs i looked at has it embedded in their phone app. I just cant get over the reality that I HATE carrying my phone with me on shift. Hate it.. like 10/10.. i hate charting just a touch more so i bring it but man do i ever hate it.

 

[RustedFox]

I wish that Google's search results weren't mostly ads at the top.

Remember when the Google motto was: "Don't be evil." - ?

 

[TooMuchResearch]

Bought my first tiny amount of BTC. Had a limit order in at 101K, bought about 2K worth. I'll probably buy more every 5K it does down. BTC is currently making lower highs and lower lows...so buying at the lower lows is not a bad thing to do right now.

Just spoke to the wife about slowly acquiring BTC and she's like 1) "aren't people kidnapped and forced to give them BTC", and 2) "You are in front of the computer too much, you gotta get out more!" We had about a 3 min conversation and it moved to things like "You gotta fix the toilets in the house." and "Set aside time for next weekend because we need to clean out the garage."

LOL

Talking to your spouse. Classic mistake! (I'm joking)

 

[xaelia]

Coming soon.

• Wolters Kluwer launched “AI Labs” in 2023, an internal sandbox for experimenting with generative AI built specifically on UpToDate’s trusted medical content
• These tools aim to provide AI‑enhanced search and conversational-style responses directly informed by UpToDate’s vetted articles — not generic LLM output
• AI‑Enhanced Search: Live since late 2024, it lets clinicians ask natural‑language questions (“What’s the best treatment for pneumonia in elderly?”) and get succinct, UpToDate‑sourced answers, with links to full articles. It emphasizes speed, context, and transparency

I have been pretty unimpressed with the AI Labs version of UTD; I find OpenEvidence slightly better.

UTD seemed to do a poor job of actually locating the most relevant articles for me for their summary/suggestions. OE was a little better, usually dragging some college guideline into the chat.

 

[RustedFox]

Everyone is doing it. One of the AI scribe programs i looked at has it embedded in their phone app. I just cant get over the reality that I HATE carrying my phone with me on shift. Hate it.. like 10/10.. i hate charting just a touch more so i bring it but man do i ever hate it.

How do you send controlled Rx? I need the two-factor six-digit code on my phone app to send narcs.

 

[thegenius]

Talking to your spouse. Classic mistake! (I'm joking)

(I put that post in the wrong thread. I moved it)

 

[thegenius]

I have been pretty unimpressed with the AI Labs version of UTD; I find OpenEvidence slightly better.

UTD seemed to do a poor job of actually locating the most relevant articles for me for their summary/suggestions. OE was a little better, usually dragging some college guideline into the chat.

Why can't UTD chat bot just reference their own website? That's what I want.

I think UTD is perfect it's just too verbose. We need a "UTD Lite" for ER docs

 

[xaelia]

Why can't UTD chat bot just reference their own website? That's what I want.

I think UTD is perfect it's just too verbose. We need a "UTD Lite" for ER docs

It would be interesting to see how result generation would change if it tried to create "role-based" summaries – e.g., for the ED doc, for the hospitalist, for the PCP, etc. with some prompt massaging.

Couldn't be any worse, really.

 

[FrustratedFamDoc]

#1: AI taking over charting is yet another shot at depersonalizing and sterilizing one of the most uniquely human interactions that we have left. It’s still the people business. Always has been, but it has been tragically been corrupted and bastardized by corporations, managed care and government mandates.

#2: When corporate is pushing it, it is ALWAYS with the intent of getting more juice from the squeeze. They don’t give a shart about making anything more streamlined, efficient or less stressful for physicians.

#3: With how accurate dragon dictation has become and the shortcut buttons I have programmed in to my speech mic, I can pretty much fly through a narrative HPI, regardless of complexity, most of the time. AI trying to make sense of me having to ask the same question in a 4th way so I can actually get useful and usable information. lol. What’s the differential on someone with vibrating insides, AI? The longer you’re in the room, the longer the list of completely irrelevant information there is. AI would never get this as its ultimate goal is to suck up every word for analysis. Maybe it’ll help people come up with the name of just one medicine they say they’re taking.

The most time consuming part of my job (primary care), by far, and it’s not even close, is getting labs/notes/imaging from docs outside my network. My org, thus far has shown 0.0% interest in addressing this issue. In short, it has nothing to do with the bottom line for them, ( or so they think. I could easily up my volume 20% with easier access to records) so it’s completely irrelevant. OTOH, they love them some AI…. Wonder why?

I’m not trying to be crotchety or a ‘back in my day’ kind of guy, but I see AI as a solution to a problem that we (physicians) didn’t realize that we actually had.