EHRSHOT

EHRSHOT

A benchmark for few-shot evaluation of EHR foundation models

15 clinical prediction tasks · longitudinal data · reproducible leaderboard

Leaderboard Paper Pretrained Model GitHub

Leaderboard

Loading…

See Full Leaderboard →
6,739
Patients
Longitudinal, non-ICU restricted
41.7M
Events
Structured EHR signals
921k
Visits
Real-world care trajectories
15
Tasks
Curated prediction suite

What is EHRSHOT?

New Longitudinal EHR Dataset

Our dataset contains de-identified structured data from the electronic health records of 6,729 patients from Stanford Medicine. Unlike MIMIC-III/IV, EHRSHOT is longitudinal and not restricted to ICU/ED.

Dataset

Curated Tasks for Benchmarking

Evaluate machine learning models using 15 clinical tasks covering diagnostics, patient outcomes, and resource allocation. Tasks are few-shot focused, requiring only a few labeled examples.

View Tasks

Reproducible Leaderboard & Baselines

Compare your model to strong baselines and other submissions using our leaderboard to see how they perform across 15 prediction tasks and updated easily with reproducible evaluation scripts.

Leaderboard Models › GitHub ›

Citation

@inproceedings{wornow2023ehrshot,
  title={EHRSHOT: An EHR Benchmark for Few-Shot Evaluation of Foundation Models},
  author={Wornow, Michael and Thapa, Rahul and Steinberg, Ethan and Fries, Jason and Shah, Nigam},
  booktitle={Advances in Neural Information Processing Systems},
  volume={36},
  year={2023}
}