Eric Yamga

AI + expertise paradox. More technical and specialized medical tasks -> easier for AI to substitute & outperform _average_ human experts. Case in point is echocardiogram interpretation. Yet it is those tasks that are better paid/reimbursed vs general primary care. HT @venkmurthy

Epic Systems makes the dominant electronic health record (EHR) system in America; if you're a doctor, chances are you're *required* to use it. For each hour a doctor spends with a patient, they spend *two* hours on clinically useless bureaucratic data-entry on an Epic EHR. 1/

Interested in using LLMs to automate writing clinical notes? Remember stories like this, which happen all the time. Writing = thinking. LLMs can and do help us think, but not if they are implemented thoughtlessly.

@ross_prager @VeinVidiVicid @ArgaizR Would love to see this too!

@VeinVidiVicid IVC size likely will never normalize with decongestion but @ArgaizR has the objective data on this from his TR clinic and study.

(1/x) The inferior vena cava (IVC) is 'useless' is a phrase stated confidently and incorrectly by #pocus novices. ❌ The IVC is useless at: - telling you what to eat for dinner - acting as a fuel gauge for volume status - predicting fluid responsiveness HOWEVER. The IVC is incredibly useful in helping us decide on the optimal fluid and decongestion strategy for your patients. A 🧵 (With a bonus discussion of future of #VEXUS research at the end. ) #medtwitter #foamed #echofirst

We write so much useless research that no one reads that we now have invented a new type of useless research that no one reads -- the "scoping review".

As clinicians spend more time reading/writing notes, Sharon Jiang, Barbara Lam, @MonicaNAgrawal, @shannonzshen, Nicholas Kurtzman, Steven Horng, @karger, & @david_sontag propose an ML approach that predicts relevant clinical notes in @JAMIA_Journal academic.oup.com/jamia/advance-…

How to be as "smart" as Auto-Regressive LLMs: - memorize lots of problem statements together with recipes on how to solve them. - to solve a new problem, retrieve the recipe whose problem statement superficially matches the new problem. - apply the recipe blindly and declare victory. - do not use basic logic. - do not use common sense to check your solution. - do not use a mental model of the situation as a sanity check. - do not simulate the scenario in your mind using your world model. - when someone tells you your solution is wrong, reply "I'm sorry, you are right" and apply another irrelevant recipe. Knowledge accumulation is not a substitute for actual understanding.

Pathway now ranking higher than UpToDate and Medscape in the US App Store by a large margin. Only 1 of us is top 100.

Very proud of the @PathwayMedical team for this one. Explored every drug API to build this, but nothing was quite labeled and granular enough for our needs. Built a drug database from scratch - a monumental task, unlike any other challenge we've tackled. More to come!

@DrToddLee @AlexLawandi @nickmmark Good summary of current evidence: pathway.md/ai/share/b032c…

This is still one of my favorites.

🚀 DeepSeekMath: Approaching Mathematical Reasoning Capability of GPT-4 with a 7B Model. Highlights: - Continue pre-training DeepSeek-Coder-Base-v1.5 7B with 120B math tokens from Common Crawl. - Introduce GRPO, a variant of PPO, that enhances mathematical reasoning and reduces training resources. More Details：arxiv.org/abs/2402.03300 Model Download：huggingface.co/deepseek-ai GitHub Repo：github.com/deepseek-ai/De… #DeepSeek #DeepSeekMath

@jonathanhershon Agreed with all the points made! I would also add that it is essential that whatever the solution is, it needs to integrate seamlessly into physicians workflow.

Ingredients to making AI that’s actually useful in clinical practice: 1️⃣ Use quality labeled data as your foundation 2️⃣ Be transparent about data sources 3️⃣ Involve clinicians at every level Agree / disagree? Anything I'm missing?

@itsdrhartmann @naval We do this at @PathwayMedical Our AI generates answers based on high-yield, evidence-based literature and our structured medical knowledge graph. Physicians love it so far.

The LLM-augmented doctor is indeed safer, better and more reliable. @PathwayMedical is showing early signs of this!

@khudhur_moh @drewskiMD @zakkohane @GauravSingalMD let me know your thoughts!

Unlike other AIs, every statement is backed by a reference from either the guidelines inside Pathway or plenty of other verified sources. 👇 Question: "P2Y12 inhibitors in STEMI w/ dose"

I left Gaza yesturday. My heart and my soul are still there with my patients. I remember their names and their wounds. I will fight until they receive the treatment they need and the justice they deserve. My heart is broken in ways I never knew was possible.

La désillusions en IA est évitable en mettant en place des équipes multidisciplinaires qui sauront identifier des projets prometteurs.

Triste constat qui rappelle l'aspect clé de l'IA en santé : définir des problèmes réels pour lesquels des solutions sont implémentables. #polqc ledevoir.com/societe/sante/…