You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Global Health Epilepsy Dataset

Clinical dataset of epilepsy patients from Uganda, compiled for multi-agent LLM research on treatment recommendation. Contains longitudinal visit records across three patient cohorts with ground-truth drug prescriptions.

Dataset Summary

Patients 699 unique patients
Total entries 2,549
Cohorts CSV (279 + 53 patients), PDF (367 patients)
Visit range Visit 1–10 (longitudinal)

Cohorts

  • CSV cohort (1,040 entries): Structured clinical notes from Mulago Hospital neurology clinic. Visits 1–6.
  • PDF cohort (1,509 entries): Free-form clinical notes extracted from patient PDFs. Visits 1–10.

Dataset Structure

Each entry represents one patient visit and is fully self-contained — the input field includes the full cumulative context of all prior visits.

Fields

Field Type Description
pid string Patient identifier
visit_num int Visit number (1-indexed)
cohort string csv or pdf
input string Full cumulative clinical context (all prior visits + current visit notes)
output string Doctor's raw prescription text for current visit
prescribed list[str] Ground-truth ASMs prescribed (normalised drug names)
stopped list[str] Ground-truth ASMs stopped at this visit

Input Format

Age: 6 | Sex: F | Diagnosis: Generalised epilepsy

[Visit 1 - Clinical Notes]
Visit Date: 02/01/2022
History of Presenting Illness: ...

[Visit 1 - Prescription]
Continue Na Valproate 100mg morning × 1/52

[Visit 2 - Clinical Notes]
Visit Date: 22/06/2023
...

Tracked ASMs

carbamazepine, clobazam, clonazepam, ethosuximide, lamotrigine, levetiracetam, phenobarbital, phenytoin, topiramate, valproate

Usage

from datasets import load_dataset

ds = load_dataset("kartiksharma4/global-health", token="your_hf_token")
df = ds["train"].to_pandas()

Access

This dataset is private. Contact the dataset owner to request access.

Citation

If you use this dataset, please cite the associated paper (forthcoming).

Ethical Considerations

This dataset contains de-identified clinical notes from Ugandan epilepsy patients. Patient names may appear in free-form clinical notes (particularly the PDF cohort). Access is restricted to approved collaborators only.

Downloads last month
60
Free AI Image Generator No sign-up. Instant results. Open Now