FHIR for Data Science

v0.0.7

Production-Scale FHIR Analytics. Running Anywhere.

The first SQL-native clinical reasoning engine. Translate CQL to DuckDB SQL for blazing fast, auditable population health analytics with zero server infrastructure.

Get Started →🌐 Live Demo

~3.9msSQL mean / patient

ZeroServer infrastructure

2,822Tests passing

FullAudit Evidence

Native FHIR Analytics

FHIR4DS treats FHIR as a first-class citizen in the analytical stack. No pre-flattening, no complex ETL—just standard SQL powered by high-performance DuckDB extensions.

Query FHIR resources using native FHIRPath expressions directly in SQL.

python

import fhir4ds

con = fhir4ds.create_connection()
con.execute("""
  SELECT fhirpath_text(resource, 'Patient.name.family') 
  FROM resources
""")

⚡

Columnar Speed

Process entire populations in a single SQL scan. ~137× faster than traditional engines by utilizing vectorized execution and SIMD.

🌐

Zero Infrastructure

No JVM, no Docker, no servers. Runs locally in notebooks or 100% client-side via WebAssembly for secure SMART on FHIR apps.

🔍

Full Auditability

Every decision includes a human-readable narrative and a logical breadcrumb trail through the logic tree. No more "black box" results.

🔬

Inspectable SQL

Logic compiles to standard, optimisable DuckDB SQL. Debug, index, or export to Parquet—you own the query and the data pipeline.

📐

SQL-on-FHIR v2

Flatten complex resources into analytical tables using portable ViewDefinitions. 100% compliant with the HL7 specification.

🎯

Standards Compliant

100% CQL (1,706 tests)
100% FHIRPath (935 tests)
100% SQL-on-FHIR v2 (134 tests)
47/47 eCQM measures

How It Works

Clinical logic compiles to a single, self-contained DuckDB SQL query. Query raw FHIR JSON directly without pre-flattening or ETL.

CQL Source→Parser→SQL Translator→DuckDB SQL→Results

FHIR4DS vs CQF Clinical Reasoning

Same accuracy target. Radically different performance.

Capability	FHIR4DS	CQF Clinical Reasoning
SQL execution per patient (all 47 measures)	~3.9ms	N/A
SQL execution per patient (12 shared measures)	~6.9ms	~936ms
Speedup (mean, 12 shared measures)	~137× faster	baseline
Zero server infrastructure	✅ DuckDB-WASM	❌ Requires JVM + server
Audit evidence trail	✅ Full narrative	❌ Aggregate counts only
Output is inspectable	✅ Plain SQL	❌ Black-box engine
Columnar scalability	✅ Vectorized	⚠️ Sequential loop
SQL-on-FHIR v2 support	✅ 100% compliance	❌ Not supported

View Full Benchmarking Report →

Ready to scale your FHIR analytics?

Join the modern healthcare data stack. Open-source, standards-compliant, and built for high-performance clinical quality measurement.

Install fhir4ds Explore User Guide ⭐ GitHub