In our paper Robust Benchmarking for Clinical Entity Extraction, we find that current performance of clinical entity extraction systems is still brittle, and that further work needs to be conducted. As part of this we developed a new schema, outlined in the paper, and below we provide instructions on how to access the proof-of-concept dataset we constructed.
For this dataset, we annotated notes from the MIMIC Critical Care Database. The dataset has been made publicly available, subject to completing the relevant trainings. See detailed instructions on gaining MIMIC access [here](https: /mimic.physionet.org/gettingstarted/access/).
To ensure we only allow access to those users who have completed MIMIC credentialing, we are storing the dataset in a Google Cloud bucket only accessible by users who have MIMIC access. To connect your Google Cloud account to your Physionet account, follow the step-by-step instructions available here.
Once you have gained MIMIC access and connected to Google Cloud, you should be able to access the dataset at this link.
This version of the dataset has been collated from the two annotators, and we wanted to provide it for exploration. We plan to release a more thorough version with more fleshed out synonyms later in 2020. To be contacted when this is released, please fill out this Google form with your email, and we will get in touch!
If you have any questions, please contact Monica Agrawal at magrawal [at] mit [dot] edu