Production engineering patterns for healthcare data pipelines.
FHIR R4 & HL7 v2 parsing · clinical ETL · terminology resolution · HIPAA de-identification & patient matching.
🌐 Visit the site → www.clinical-data.org
clinical-data.org is a deep, production-focused engineering reference for the teams who move clinical data for a living — health-tech engineers, clinical data scientists, ETL developers, and compliance teams. Every page is written for real systems: deterministic patterns, concrete failure modes, explicit type contracts, audit-ready logging, and runnable Python.
It answers the questions that break pipelines in production: how to tokenize a vendor-mangled HL7 v2 stream without dropping messages, how to resolve SNOMED CT to ICD-10 through a version-pinned decision table, how to keep a FHIR bulk export idempotent, how to de-identify a longitudinal cohort without fragmenting a patient, and how to match records across systems that share no common identifier.
The standards contract end to end — HL7 v2 segment grammar and MLLP transport, FHIR R4 resource modelling, REST vs Bulk Data export, terminology-server integration, US Core conformance, SNOMED/LOINC mapping, and transaction-bundle referential integrity.
Turning raw clinical telemetry into trusted assets — Python-native parsing, type coercion, async batch processing, idempotent loads, data-quality validation frameworks, and orchestration with Airflow, Prefect, Spark, and dbt.
The identity boundary — Safe Harbor vs Expert Determination, removing the 18 identifiers, date-shifting, k-anonymity, deterministic and probabilistic (Fellegi–Sunter) patient matching, master patient index strategies, and tamper-evident PHI audit logging.
- Runnable, not hand-wavy. Every implementation section ships correct, idiomatic Python you can lift into a service and test in isolation.
- Standards-accurate. FHIR R4 (4.0.1), US Core, HL7 v2 encoding rules, UCUM/LOINC/SNOMED CT, and the HIPAA Privacy and Security Rules are treated as first-class facts, not afterthoughts.
- Compliance built in. PHI handling, minimum-necessary, audit trails, and re-identification risk are engineered into every layer rather than bolted on.
- Deeply interlinked. Hand-authored diagrams, spec tables, and a tight cross-reference graph make the whole domain navigable in one or two clicks.
Static site built with Eleventy, hand-authored inline SVG diagrams, structured
data (JSON-LD TechArticle, BreadcrumbList, FAQPage), a PWA service worker, and full WCAG 2 AA
accessibility. Deployed on Cloudflare.
Spotted an error, or want a topic covered? Open an issue — corrections to the clinical, standards, or compliance details are especially welcome.