Request access to this dataset
Synthetic data built for the TechIreland National AI Challenge 2026. No real patient, clinician or hospital is represented. Access is granted individually while the project is in development. You will be asked to share your username and email address.
Log in or Sign Up to review the conditions and access this dataset content.
Synthetic Irish Hospital Referral Prioritisation Dataset
Every record here is synthetic. No real patient, clinician or hospital is represented. Hospital names are invented and their codes use a reserved range that matches no real facility.
A hospital outpatient waiting list: people waiting for a clinic appointment, what is wrong with them, how urgent a clinician judged them, and how full the wards and clinics are. It exists to answer one question: among patients already marked equally urgent, who should be seen next, and why?
What is in it
Six hospitals. Four public, which have waiting lists; two private, which have beds and clinics but no queue of their own.
| Referral | One person waiting for one appointment. Identified by hospital_hipe + pathway_number. Numbers are only unique within a hospital. |
| Triage category | How urgent a clinician judged it: Urgent, Semi-Urgent or Routine. Assigned once, and nothing in the data reorders across categories. |
| Timeframe | How long that category should wait: 28 days if Urgent, 91 if Semi-Urgent. |
| Wait | Four different clocks, because they give different answers. adjusted_wait_days excludes time when care was bought privately and the clock paused. Use it for breach checks. |
| Conditions and observations | What is wrong with the patient, and their vital signs. |
| Capacity | Wards, bed occupancy three times a day, and clinic appointment slots. |
Vitals do not tell you who is urgent. Roughly two thirds of the urgent referrals here have entirely unremarkable observations. A suspected melanoma is urgent because of the referral pathway, not the physiology. That is measured from real triage data, not assumed.
ground_truth.csv is the answer key: which patients deteriorated while waiting. It is
held out and should never be loaded alongside the rest. Its risk model is synthetic and
unvalidated, and results depending on it must say so.
The two folders
| Size | Hospitals | Referrals | |
|---|---|---|---|
sample/ |
1.3 MB | 2 | 609 |
full/ |
11 MB | 6 | 5,200 |
sample/ is a real slice of full/, closed under its foreign keys: every table it
references, it contains. Use it to check a pipeline works; use full/ for anything you
report.
Pin a tag, never a branch. v1.1 is current.
More
The repository holds the generator, the schema, the tests and the full specification: what every column means, where every number came from, and what the real data disproved along the way.
github.com/NonsoCynthia/Tus-Aite-TechIreland-Challenge,
branch dataset_branch. The repository is private and open to the project team only, so
the link will not resolve unless you have been added to it.
Start with README.md there, then docs/GETTING_THE_DATA.md to load it and
DATASET_README.md for the column-by-column specification.
- Downloads last month
- 7