FHIR4DS

FHIR for Data Science - A unified suite of high-performance tools for working with FHIR healthcare data in analytical environments.

Overview

FHIR4DS provides high-performance tools for querying and transforming FHIR data. Built on DuckDB for efficient analytical processing, it offers native support for FHIRPath expressions, CQL clinical quality measures, and SQL-on-FHIR v2 ViewDefinitions.

Packages (Unified Namespace)

The project is now organized under the fhir4ds namespace for a clean, professional API.

Non-Python Components

• C++ Extensions: Located in extensions/fhirpath and extensions/cql. These provide native high-performance DuckDB extensions.
• Web Demos: Located in web/wasm-demo and web/website.

Quick Start

Installation

pip install fhir4ds-v2

# With optional measure evaluation dependencies (numpy, pandas, etc.)
pip install "fhir4ds-v2[measures]"

# With HAPI PostgreSQL materialization support
pip install "fhir4ds-v2[hapi]"

# With Mongo materialization support
pip install "fhir4ds-v2[mongo]"

Unified Connection

The easiest way to get started is the create_connection helper, which returns a DuckDB connection with all FHIRPath and CQL UDFs pre-registered.

import fhir4ds

con = fhir4ds.create_connection()

# Load FHIR data
con.execute("CREATE TABLE resources AS SELECT * FROM 'data/*.parquet'")

FHIRPath Queries

result = con.execute("""
    SELECT
        fhirpath_text(resource, 'Patient.id') AS patient_id,
        fhirpath_text(resource, 'Patient.name.given[0]') AS given_name
    FROM resources
    WHERE resourceType = 'Patient'
""").fetchdf()

CQL Measure Evaluation

result = fhir4ds.evaluate_measure(
    library_path="./CMS165.cql",
    conn=con,
    output_columns={
        "ip": "Initial Population",
        "denom": "Denominator",
        "numer": "Numerator",
    }
)
print(result.df())

SQL-on-FHIR ViewDefinitions

import json

view_definition = {
    "resource": "Patient",
    "select": [{"column": [
        {"path": "id", "name": "patient_id"},
        {"path": "gender", "name": "gender"},
    ]}]
}

sql = fhir4ds.generate_view_sql(json.dumps(view_definition))
patients_flat = con.execute(sql).df()

Standalone FHIRPath

from fhir4ds.fhirpath import evaluate

patient = {
    "resourceType": "Patient",
    "id": "123",
    "name": [{"given": ["John"], "family": "Doe"}]
}

result = evaluate(patient, "Patient.name.given")
# result: ["John"]

Test Coverage & Compliance

The unified engine is rigorously tested against official specifications.

Architecture

The project follows a Feature-First hierarchy:

• fhir4ds/: Unified Python package root.
  • fhirpath/: Core FHIRPath engine.
    • duckdb/: DuckDB-specific FHIRPath integration.
  • cql/: Core CQL-to-SQL translator.
    • duckdb/: DuckDB-specific CQL integration.
  • viewdef/: SQL-on-FHIR v2 ViewDefinition implementation.
  • sqlquery/: SQL-on-FHIR v2 Analytics Layer (SQLQuery / SQLView Library profiles).
  • dqm/: Digital Quality Measure orchestration.
• extensions/: High-performance C++ source for DuckDB extensions.
• web/: Documentation website and interactive WASM demos.

Development

Setup

# Clone the repository
git clone https://github.com/fhir4ds/fhir4ds.git
cd fhir4ds

# Install in editable mode with development dependencies
pip install -e ".[dev]"

Running Tests

Pytest is configured in pyproject.toml. The config disables the auto-loaded pytest-benchmark plugin because it can hang pytest startup/collection in this workspace; normal benchmark runs use the scripts in benchmarks/ instead.

# Run all Python tests
pytest fhir4ds/

# Run specific subpackage tests
pytest fhir4ds/fhirpath/tests/unit/
pytest fhir4ds/dqm/tests/

# Run official spec conformance suites
python3 conformance/scripts/run_all.py

WASM Demo Release Process

The interactive CQL playground at web/wasm-demo/ embeds a Python wheel served to Pyodide. After any Python source change or version bump, update the WASM demo:

# 1. Build the Python wheel
hatch build -t wheel

# 2. Copy to public/ and remove the old version
cp dist/fhir4ds_v2-NEW_VERSION-py3-none-any.whl web/wasm-demo/public/
rm web/wasm-demo/public/fhir4ds_v2-OLD_VERSION-py3-none-any.whl

# 3. Build the WASM demo
cd web/wasm-demo && npm run build

# 4. ⚠️ Deploy to website static directory (required — website uses a pre-built snapshot)
cd ..  # back to repo root
rm -rf web/website/static/wasm-app
cp -r web/wasm-demo/dist/. web/website/static/wasm-app/

# 5. Verify — all 11 tests must pass
cd web/wasm-demo && npx playwright test tests/e2e/playground.spec.ts tests/e2e/web-component.spec.ts

Why step 4 matters: The website's static/wasm-app/ is a pre-built snapshot served by Docusaurus. It is NOT auto-updated when web/wasm-demo/ is rebuilt. Skipping step 4 causes the standalone demo to work while the website CQL playground fails with a Pyodide error.

Why duckdb is excluded from Pyodide: duckdb has no pure Python wheel on PyPI. The Pyodide worker uses deps=False and manually installs only the required pure-Python dependencies. Do not add native C extension packages as hard imports in fhir4ds.cql or any module imported by the WASM worker. See AGENTS.md and web/wasm-demo/AGENTS.md for full details.

Benchmarks

Performance benchmarking tools and results are located in the benchmarks/ directory. See benchmarks/AGENTS.md for details.

License

This project is dual-licensed:

1. Open Source: GNU Affero General Public License v3 (AGPL-3.0). See LICENSE for details.
2. Commercial: A proprietary license for enterprise use, embedding in closed-source products, and high-performance C++ extensions.

For commercial licensing inquiries, please contact contact@fhir4ds.com.