Results

A succeeded job lists its results in results, each with the URL to download it from:

"results": [
  { "name": "fhir",   "filename": "fhir_bundle.json", "content_type": "application/fhir+json",
    "size_bytes": 136319, "url": "/v1/jobs/…/results/fhir" },
  { "name": "notes",  "filename": "notes.zip",        "content_type": "application/zip",
    "size_bytes": 5848,   "url": "/v1/jobs/…/results/notes" },
  { "name": "result", "filename": "result.json",      "content_type": "application/json",
    "size_bytes": 63127,  "url": "/v1/jobs/…/results/result" }
]

Each result can be downloaded once; a second request returns 410 result_expired. Results not downloaded are deleted 24 hours after the job finishes (results_expire_at).

FHIR bundle

fhir_bundle.json is a FHIR R4 transaction Bundle following US Core conventions. Resources refer to each other by urn:uuid full URLs, so you can POST the bundle to a FHIR server as one transaction.

Cultures, and study measurements

A culture is a DiagnosticReport with category MB. Its results are a Gram stain Observation and one Observation per organism, each organism’s valueCodeableConcept carrying its SNOMED CT code. An organism’s hasMember lists its susceptibility Observations: the antibiotic’s LOINC code, the MIC as a valueQuantity in ug/mL with a comparator, and an S/I/R interpretation.

ECG and pulmonary function measurements are Observations with category procedure, referenced by the study’s DiagnosticReport:

Document types and their LOINC codes

Each DocumentReference.type carries the LOINC code for its kind of document, so you can map a document back to the document_types value that asked for it:

Cancer

Each cancer’s details follow the conventions of mCODE, the HL7 standard for oncology data:

• Stage: the clinical stage at diagnosis, and the pathologic stage after surgery. Each stage group (LOINC 21908-9 clinical, 21902-2 pathologic) links to its T, N and M categories. AJCC values are coded with SNOMED CT.
• Biomarkers: for example hormone receptors, HER2, PD-L1 or gene mutations, each with its result.
• ECOG performance status (LOINC 89247-1), at diagnosis and at each assessment.
• Disease status (LOINC 97509-4) at each restaging or surveillance assessment, coded with SNOMED CT (not detected, improved, stable, worsened).
• Radiation therapy: each course as a Procedure (SNOMED CT 1217123003), with its treated site, dates, dose and number of fractions.
• Systemic therapy: each drug as a MedicationRequest that gives the cancer as its reason.

Each of these refers to the cancer’s Condition.

Resource ids are derived from the patient, so the same patient always produces the same ids.

Documents

notes.zip holds one text file per document, numbered in date order:

001_2025-03-14_progress_note.txt
002_2025-09-02_history_and_physical.txt
003_2025-09-03_progress_note.txt
004_2025-09-05_discharge_summary.txt
manifest.json

manifest.json describes each document: its file, note_type, LOINC document type, date written, author and specialty, facility, and the encounter’s setting and id, which match the FHIR bundle and result.json.

notes_pdf.zip (output pdf) holds the same documents as PDFs, laid out as a hospital’s clinical documents would be, plus chart.pdf: the whole chart in one file, with a bookmark for each document. Every PDF is marked as synthetic.

The full record

result.json is the patient’s complete record as JSON:

{
  "record": {
    "patient_id": "…",
    "demographics": { … },
    "social_history": { … },
    "conditions": [ … ], "procedures": [ … ], "medications": [ … ], "allergies": [ … ],
    "immunizations": [ … ], "encounters": [ … ], "observations": [ … ],
    "cultures": [ … ], "coverage": [ … ], "cancers": [ … ], "notes": [ … ]
  },
  "seed": 20260630,
  "complexity": 7,
  "requested_conditions": { "heart failure with reduced ejection fraction": { … } }
}

Each coded item has its text and its codings (system, code, display), and each item links to the encounter it happened at. A condition the chart is mainly about carries primary: true, and a hospital stay’s encounter carries diagnosis_condition_id — the id of the condition chiefly responsible for it, the principal diagnosis, or the most responsible diagnosis in Canada. notes holds every document with its text, type, author and dates. requested_conditions shows how each condition you asked for was coded.

cancers holds each cancer’s course: primary site, histology and grade, clinical and pathologic stage with TNM, biomarkers, performance status, each treatment (operations, chemotherapy, endocrine, targeted and immunotherapy, and radiation with site, technique, dose and fractions), with its dates, cycles, response and side effects, and each restaging assessment. It also holds the workup — how the cancer was found, diagnosed and staged, and the tests done before treatment, each with its date and its procedure — and, for breast and gynecologic cancers in women, menopausal_status.

A diagnostic study in procedures carries its result: the findings, impression and key measurements its report gives.

cultures holds each culture: the specimen, when it was collected and finally reported, the Gram stain, the result, and each organism with its susceptibilities — organisms coded with SNOMED CT, antibiotics with RxNorm, tests with LOINC. coverage holds the patient’s health plans, primary first: payer, plan, member ID, group, subscriber, relationship, dates and benefits.

Bulk Data

A cohort’s bulk result, cohort_fhir_bulk.zip, is the whole cohort as FHIR Bulk Data, the format of a FHIR server’s population export:

AllergyIntolerance.ndjson
Condition.ndjson
DiagnosticReport.ndjson
DocumentReference.ndjson
Encounter.ndjson
…
Patient.ndjson
Practitioner.ndjson
manifest.json

Each .ndjson file holds one resource per line (NDJSON is the same as JSONL). References are relative (Patient/<id>), as in a server export. The health system’s clinicians and facilities are shared by the cohort’s patients and appear once. manifest.json follows the Bulk Data export manifest, listing each file with its resource type and count.