From Messy Clinical Data to Interoperable FHIR: A Python-First Mapping and Validation...
summary
A Flask + pandas pipeline converts uploaded CSV/Excel into valid FHIR bundles using Smart on FHIR, LOINC, SNOMED CT and RxNorm lookups; it stays stateless in memory and yields immediate feedback while avoiding persistent storage. Column mapping uses deterministic dictionary matching (built with offline LLM expansion) for speed and reproducibility, then a second stage uses fuzzy matching via RapidFuzz against coded vocabularies. Built for production, it deduplicates rows, validates bundles, and keeps provenance; it works fast enough for clinicians and supports hundreds of EHR formats.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 1:53Healthcare data reality
- 4:01Evolution of standards
- 5:23Design constraints
- 7:09Stateless architecture
- 8:36Application overview
- 10:47Patient identifiers in FHIR
- 11:30Auto-matching with Python
- 15:52Deterministic vs generative AI
- 16:33Mapping results UI
- 17:36Unmapped data handling
- 19:03Fuzzy matching for codes
- 21:32Final FHIR assembly
- 22:36Pandas transformations
- 24:00Future plans
- 25:04Lessons learned
- 26:09Open sourcing
- 27:14SNOMED and LOINC mapping
- 28:17FHIR provenance
- 31:23Q&A
related talks
36mHealthcare process management in Python: a use case
36mHealthcare process management in Python: a use case
28mProductionize Research Oriented Code By Python
45mPython useRs
32mReproducible Research in Python
01h 01mParallel Python Analyzing Large Data Sets
39mMaking Fake Data Generators for Open Source Healthcare Data Science Projects
29mObject-Oriented Oncology: Making Sense of Complex Patient Journeys