Synthetic Patient Datasets

Working with the Clinia Context Engine requires clinical data, but not everyone has immediate access to a de-identified patient record. Four fully synthetic profiles — no real person, no PHI — cover four sizes of chart, from a single admission to four and a half years of complex care, so you can pick the one that matches what you want to exercise.

Each archive holds bundle.json (a FHIR R4 collection bundle) and a cda/ folder of C-CDA XML documents that span the same timeline. Ingest the bundle first, then each document, to exercise cross-format resolution; either format also stands alone. The three pediatric archives also hold document_manifest.json, which lists every file with its SHA-256. Jeanne's bundle is also available on its own as fhir-bundle.json.

Jeanne Tremblay

Jeanne is a 72-year-old French-Canadian woman from Montreal with a decade-long clinical history anchored on severe COPD. Her record spans primary care, respirology, physiotherapy, gastroenterology, radiology, and a recent emergency admission — which makes it a good stress-test for entity resolution, cross-source deduplication, and condition story generation.

Demographics

Allergies

Active conditions

Current medications

Clinical timeline

Ten episodes spanning 2017 to 2026:

1. 2017 — Bilateral knee osteoarthritis diagnosed; conservative management and physiotherapy referral.
2. 2019 (June) — Sequential pneumococcal PCV13 vaccination (previously primed with PPSV23).
3. 2019 (August) — Screening colonoscopy; two diminutive tubular adenomas resected, 10-year surveillance recommended.
4. 2020 — Annual cardiovascular review; lipid panel near target, renal function stable (eGFR 68).
5. 2022 — Acute viral upper respiratory infection without COPD exacerbation; telehealth visit, self-limited recovery.
6. 2023 (February) — Baseline DEXA scan; osteopenia at femoral neck (T-score −1.9), no pharmacologic anti-resorptive indicated.
7. 2023 (November) — Vitamin D recheck (78 nmol/L, at target); COPD inhaler reconciliation, triple inhaler regimen confirmed.
8. 2024 — Physiotherapy reassessment; worsening right knee pain limiting walking to 15 minutes, 8-week strengthening program initiated.
9. 2025 — Annual respirology GOLD reassessment; spirometry FEV1 1.10 L (44% predicted), FEV1/FVC 0.48, CAT score 14, GOLD group B, stable.
10. 2026 (May) — Acute COPD exacerbation; emergency admission with SpO2 87% on room air, purulent sputum, compensated hypercapnia (pH 7.34). Treated with controlled oxygen, nebulised bronchodilators, prednisone, and doxycycline (penicillin/sulfa allergy). Discharged after 48 hours.

Liam Carter

Liam is a 9-year-old from Westerville, Ohio, with nine years of ordinary pediatrics and then one month in spring 2026 that generated more documentation than the previous nine years combined. Three care teams wrote his chart — his pediatric practice, a children's hospital, and a home infusion service — and each restates the same problems and medications in its own words at every visit. The record is built so that the answer to "what is true now" differs from what most individual notes say.

Demographics

Conditions

Medications

Four home medications are current: fluticasone, albuterol, cetirizine and methylphenidate extended-release. Five peri-operative drugs stopped within a month and are restated as history in later notes:

Oxycodone is the one drug carried only by the documents: three notes name it, but it has no resource in the FHIR bundle. Ingest the bundle alone and it is not there. The home infusion admission note also lists oxycodone 2.5 mg as needed among his home medications, while the discharge summary says no opioid went home; the two disagree.

What to look for after ingestion

• The current problem list should read asthma, ADHD and allergic rhinitis, with the surgical cluster under resolved. The appendicitis is restated once under a second heading, Gangrenous perforated appendicitis with localized peritonitis, which the operative encounter charts active and only the hospital records imported by the home infusion service resolve — it is the same appendix, and it came out.
• The current medication list should hold the four home medications only. Every antibiotic order and statement is stopped, and the as-needed acetaminophen and ibuprofen from discharge and a five-day prednisolone course from 2023 are completed. The admission note's ceftriaxone, metronidazole and amoxicillin-clavulanate are planned orders (draft), not active ones.
• The pelvic drain and the PICC are Device resources placed and removed on known dates, and both go inactive on removal on 2026-04-24; neither belongs in the current state.

Ava Reyes

Ava is a 9-week-old from Dublin, Ohio, entirely well until RSV bronchiolitis at nine weeks; her 4-year-old brother attends daycare. Her record is one admission long: an emergency assessment, four ward progress notes, a discharge summary, and a note from her pediatrician a week later. It is the smallest chart here and the one with the least to reconcile — any approach that cannot answer questions about it correctly has no excuse.

Demographics

Conditions

The 2026-09-21 follow-up charts all three problems resolved, and the children's hospital summary imported with it, a DocumentReference in the bundle rather than a C-CDA file, agrees.

What to look for after ingestion

• No medications, no allergies, and exactly three conditions. A fourth condition is a confabulation from the family history or the sibling's wheeze.
• All three conditions should read resolved.
• The nasogastric tube was placed 2026-09-11 and removed 2026-09-13; it should sit in the history, not the current state.
• Apart from the newborn screens of 2026-07-10, every measurement comes from the admission or the 2026-09-21 visit. The current weight is the 9/21 one, 5.6 kg, not an admission value.
• Hepatitis B at birth is the only vaccine given. Nirsevimab and the 2-month series are charted not-done on 9/10 and 9/21, so "what is she due for" has an answer.

Mason Whitaker

Mason is a 4-year-old from Columbus, Ohio, born at 25 weeks and 2 days weighing 720 g. His record is his whole life: a 314-day neonatal intensive care stay, three shunt revisions, two pneumonia admissions, three pediatric intensive care stays, and a care team that grew to eleven subspecialties. Four services wrote his chart — a children's hospital, its home ventilation service, a pediatric rehabilitation center, and his primary care pediatrician — and every consultant restates the same problem list from their own angle. The record is built so that the current answer to any concrete question (tube size, valve setting, ventilator settings, anticonvulsant dose) sits beside an older, superseded version of itself, restated many times.

Demographics

Conditions

Several problems are restated under two labels that disagree: the intraventricular hemorrhage is resolved under its own name and active under the merged "grade IV left … grade II right" heading, the neonatal seizures are resolved as "secondary to grade IV IVH" and active as "neonatal seizures, clinical", and the January 2025 RSV bronchiolitis with tracheitis is never marked resolved at all. That disagreement is in the record on purpose — reconciling it is the work.

Medications

Thirteen medications are current, all but two given through the gastrostomy: levetiracetam 300 mg twice daily, clobazam 2.5 mg at bedtime, baclofen 5 mg three times daily, glycopyrrolate 0.6 mg three times daily, omeprazole 10 mg daily, polyethylene glycol 3350, cholecalciferol 1500 IU, ferrous sulfate 25 mg elemental, melatonin 3 mg, a peptide-based high-density formula, acetaminophen as needed, plus inhaled fluticasone and albuterol. Levetiracetam (100 → 150 → 300 mg), baclofen (2 → 5 mg) and glycopyrrolate (0.3 → 0.6 mg) each appear in several dose generations as weight rose; only the latest is current. Four drugs stopped along the way and are restated as history in later notes:

Phenobarbital and lansoprazole were both stopped in 2023. Later notes keep restating them, always as stopped, so the restatements outnumber the active entries without changing the answer.

Devices

Seven device families are charted as Device and DeviceUseStatement resources, and most of them changed at least once: the tracheostomy tube (4.0 Bivona cuffless, then 4.5 Bivona flex from March 2025), the gastrostomy button (12 Fr, then 14 Fr from 2024), the ventriculoperitoneal shunt valve (2.5, then 1.5 after the 2024 revision and again at the March 2026 valve replacement), the home ventilator, bilateral hearing aids, a TLSO brace and a central venous catheter placed for IV antibiotics in 2023. The superseded entries are still active where the note that created them said so — only the newest one is true today.

What to look for after ingestion

• The current problem list should be dominated by the sequelae of prematurity — hydrocephalus, cerebral palsy, epilepsy, chronic respiratory failure, feeding — with retinopathy and the neonatal cluster under resolved.
• Device state is the sharpest test: the tracheostomy tube, the gastrostomy button and the shunt valve each have a current value and an older one restated many times, mostly as active devices. Only the latest is right.
• Nocturnal ventilator settings were escalated during the January 2025 RSV admission (peak PS 18/8, backup rate 28; PS 14/7, backup rate 22 at discharge) and stepped back down to the PS 12/6, backup rate 20 baseline that the February 2026 sleep study confirms; the notes in between disagree by design.
• Levetiracetam, baclofen and glycopyrrolate each carry several dose generations; only the latest should be current.
• Growth is charted against corrected age, about 15 weeks behind chronological age; raw percentiles get the trajectory wrong.

Dataset versions

Every dataset carries a version and a SHA-256 in manifest.json, alongside the hash of each file inside the archive. Use it to confirm which generation of a dataset you hold and whether your copy still matches what this site publishes:

curl -sO https://docs.clinia.com/samples/liam-carter.zip
curl -s https://docs.clinia.com/samples/manifest.json | jq -r '.datasets[] | select(.id=="liam-carter") | "\(.archive.sha256)  liam-carter.zip"' | shasum -a 256 -c

A dataset's version changes whenever its content does; a hash tells you that something changed, the version tells you which generation you have. Benchmarks and demos that embed a copy of a dataset should pin its version and hashes rather than the files alone.

Ingest a dataset

See Ingest a FHIR Bundle and Ingest a CDA Document for the ingest commands. Use the dataset id (jeanne-tremblay, liam-carter, ava-reyes, mason-whitaker) as the registry key, or any identifier you prefer.