tds-research-tech/shield-sample

This is a public sample of the SHIELD dataset. It contains a small subset of the data for exploring the schema and building your pipeline. For the full dataset, see tds-research-tech/shield (requires access approval).

SHIELD

SHIELD is a dataset of clinical text with PHI (Protected Health Information) span annotations for benchmarking de-identification and named entity recognition models. The dataset contains de-identified clinical notes from Stanford Medicine with expert-annotated PHI spans across 9 entity categories.

Quickstart

1. This public sample is available without access approval — use it to explore the schema and build your pipeline
2. To access the full dataset, complete the access request process at tds-research-tech/shield

Dataset Structure

Notes Table

Column	Type	Description
note_id	string	Unique identifier for each clinical note
note_text	string	Full text of the clinical note
note_type	string	Type of clinical note (e.g., discharge summary, radiology report)

Spans Table

Column	Type	Description
span_id	string	Unique identifier for each annotated span
note_id	string	Foreign key linking to the notes table
span_start	int	Character offset where the span begins
span_end	int	Character offset where the span ends
span_label	string	PHI category label

Labels

Label	Description
age	Patient age
date	Dates (admission, discharge, procedures, etc.)
doctor	Physician or provider names
hospital	Hospital or facility names
id	Medical record numbers, SSNs, or other identifiers
location	Addresses, cities, states, zip codes
patient	Patient names
phone	Phone or fax numbers
web	Email addresses, URLs, IP addresses

Usage

Loading the public sample

from datasets import load_dataset

# Public sample (no access request needed)
sample_notes = load_dataset("tds-research-tech/shield-sample", "sample_notes")
sample_spans = load_dataset("tds-research-tech/shield-sample", "sample_spans")

Loading the full dataset (requires access)

from datasets import load_dataset

# Full dataset (requires approved access)
full_notes = load_dataset("tds-research-tech/shield", "full_notes")
full_spans = load_dataset("tds-research-tech/shield", "full_spans")

---

Access Requirements

NOTE: You must apply using your institutional email address. If you use Gmail or other personal email addresses, you will not be granted access.

Before applying, make sure you have the following:

• A Hugging Face account. Create one at huggingface.co/join if you don't have one. You will need to provide your Hugging Face username in the application.
• Verified affiliation with an Academic, Government, or Industry Research Lab. Please use your institutional email address when applying — do not use Gmail or personal emails.
• CITI Training: Complete the CITI Program course "Data or Specimens Only Research" (the same course required for MIMIC-IV access on PhysioNet). See instructions here. You will need to upload your completion certificate during the application.

The access request process has six steps:

1. Request Access on Hugging Face — Go to tds-research-tech/shield and click "Request Access" to register your Hugging Face account.
2. Complete CITI Training — If you haven't already, complete the CITI "Data or Specimens Only Research" course and download your completion certificate.
3. Submit Access Request on REDCap — Complete our REDCap verification form. Provide your name, institution, role, Hugging Face username, intended use, and upload your CITI completion certificate.
4. Sign the Data Use Agreement — Read and electronically sign the DUA. You will be automatically redirected after step 3.
5. Verify Your Institutional Email — A verification link will be sent to your institutional email address. You must click the link to confirm ownership.
6. Admin Review — Your request will be reviewed and you will be notified by email of the decision. Once approved, access will be granted to your Hugging Face account.

These data must remain on your encrypted machine. Redistribution of data is FORBIDDEN and will result in immediate termination of access privileges.

Important Notes

• Sending SHIELD data over a non-HIPAA-compliant API is a violation of the DUA.
• If you create derived research artifacts based on SHIELD (such as additional annotations or synthetic data), please contact us to discuss appropriate hosting and licensing.
• Please allow 5 business days to process applications.

---

Contact

JD Posada (jdposada@stanford.edu)

Citation

@inproceedings{YOUR_CITATION,
  title={YOUR TITLE},
  author={YOUR AUTHORS},
  year={2026},
}