🏥 Excited to announce the official full release of 👂💉EHRSHOT -- a dataset of 6,739 deidentified longitudinal EHRs for few-shot eval of clinical foundation models! 🌐 Website: ehrshot.stanford.edu 💽 Dataset: https://lnkd.in/gh8x7MDT 🤗 Model: https://lnkd.in/g8P6QA38 EHRSHOT has 3 parts: 1. Dataset: 👫 6,739 patients 🩺 41M clinical events 2. Benchmark: 📈 15 prediction tasks 🔍 Few-shot eval 3. Model: 🤖 141M param decoder-style model 🏥 Trained on 2.6M deidentified EHRs 🤗 Available on HF: https://lnkd.in/g8P6QA38 We're keeping a running leaderboard for models on our website, and you can generate our results using the code at our repo. Please check out the paper / links below to learn more and get access to the dataset! 👨💻 Code: https://lnkd.in/gZswu_Gx 📄 Paper: arxiv.org/abs/2307.02028 A huge thanks to Natasha Flowers, Priya Desai, Joe Mesterhazy, Somalee Datta, and Todd Ferris at Stanford University School of Medicine Stanford Health Care for their heroic efforts in getting this dataset released, as well as co-authors Rahul Thapa, Ethan Steinberg, Jason Fries, and Nigam Shah
So interesting -- thank you for sharing!
Amazing Work Michael Wornow! So glad we could support Stanford University School of Medicine to publicly release an EHR dataset.
You're churning out cool healthcare LLM papers / projects one after another!
Awesome EHR dataset from Michael Wornow for clinical foundation models
This is awesome!
Awsome dataset Mike Keep up the great work 👍
This is so cool!
Let’s go !!
Really cool! Nice work.
Founder & CTO @ Nomic
2moCurious, why do we need HIPPA training to access the dataset but not the model?