Impressed by Open Evidence

[Palex80]

Advertisement - Members don't see this ad

Hey

I am not sure how many of you regularly use AI apps to help in clinical practice. I am quite impressed by Open Evidence.

OpenEvidence

The leading medical information platform.

www.openevidence.com

The best thing about it are the references it gives. Although sometimes not complete, they are really helpful.

An example:

 

[Gfunk6]

Haven't heard of it, but glad you posted about it. I tossed a couple of questions at it - it seems very solid.

 

[Mandelin Rain]

Discussed briefly on Ezra Klein's recent podcast on Artificial General Intelligence.

Unfortunately, the rest of the episode was a bit terrifying.

 

[A_DeMichele]

very impressive. Can say goodbye to Pubmed

 

[xrt123]

I ask AI all of my clinical cases at this point. Open evidence gives more succinct answers and Grok 3 appears to be the best in my opinion for detailed answers. This is based on about 50 clinical scenarios to Open evidence, Grok3 and Chat GPT. All AI at this point are great and all are smarter than me.

Intelligence is now a marginalized commodity whereas ownership and regulation will be what is important for staying ahead of others in society, but hopefully a high tide will raise all boats. To be honest I hope this acts to be a reliable way to decrease animosity between humans. It seems much more reliable as a way to obtain facts instead of through the lens of tradition bias media from all directions.

 

[OTN]

No mention of partial breast isn't particularly great, and only mentioning the boost when it comes to not choosing 5 fraction breast doesn't make a ton of sense. I added the words "hormonally-positive" in front of "left breast cancer" and all it did was not talk about 5 fraction whole breast at all.

 

[TheWallnerus]

No mention of partial breast isn't particularly great, and only mentioning the boost when it comes to not choosing 5 fraction breast doesn't make a ton of sense. I added the words "hormonally-positive" in front of "left breast cancer" and all it did was not talk about 5 fraction whole breast at all.



View attachment 399941

Ha yeah weird. That said I can’t fault the AI too much because there is a huge blind spot in the NCCN guidelines for 5 fraction. Were this a British AI it would know the Royal College of Radiologists recommendation for 26 Gy/5fx in this case; as we know 28.5 Gy is a bit gauche (FAST) given 27 Gy being slightly worse for side effects versus 26 Gy (FAST Forward). Gauche… and NCCN recommended! And we need to tell the AI that 28.5/5 does have 10 year data… something like 3 in 300 in breast recurrences versus 4 in 300 for 50/25 at 10y.

 

[Palex80]

 

[TheWallnerus]

No mention of partial breast isn't particularly great, and only mentioning the boost when it comes to not choosing 5 fraction breast doesn't make a ton of sense. I added the words "hormonally-positive" in front of "left breast cancer" and all it did was not talk about 5 fraction whole breast at all.



View attachment 399941

Q: "What is the best dosing and fractionation schedule for a 75 year old woman with a stage I left breast cancer after lumpectomy?"
A: "40 [or] 42.5 Gy in 15 [or] 16 fractions."

I do not think the A.I. understood the word "best." 😉

Puts me in mind of what Steve Webb said about optimization a long time ago:

 

[winstonfoot5]

Perplexity Deep Research also very impressive

 

[thesauce]

This is a classic garbage in/garbage out scenario. These are all just search engines that spit out other people’s opinions. The computer can’t tell you what the best thing is. That requires personal judgement.

Awhile back I posted that thing about subjective minimizing language in medical oncology conferences. Guaranteed that if you ask the search engine about the drugs that show death in 10%, it will say it’s well tolerated because that’s what the author said.