End-to-End CCDA to FHIR Conversion Pipeline: Utilizing NLP and Hybrid Approach
End-to-End CCDA to FHIR Conversion Pipeline Using NLP - Contra
Summary
Hasshya Moorthy developed an end-to-end interoperability pipeline that converts CCDA (Clinical Document Architecture) XML files into standardized FHIR resources. The system employs a hybrid approach combining deterministic extraction, NLP-assisted ontology mapping, and semantic normalization.
Details
This project addresses the challenge of clinical data being stored in heterogeneous formats within healthcare systems by building a conversion pipeline from CCDA (Clinical Document Architecture) XML to FHIR (Fast Healthcare Interoperability Resources) compliant resources. The system utilizes streaming XML parsing for efficient handling of large clinical datasets and extracts both structured and unstructured medical data. A key component is the integrated NLP pipeline, which uses spaCy/SciSpaCy to perform advanced tasks such as clinical entity recognition, contextual linking, and negation detection. Concepts are mapped to major healthcare standards including SNOMED CT, LOINC, and RxNorm. Furthermore, generated FHIR R4 resources and JSON bundles undergo validation, incorporating a human-in-the-loop review workflow and a Confidence Scoring System (S-MCS) to ensure high reliability and practicality. The technology stack is built on Python, leveraging libraries like lxml and spaCy, providing a robust data conversion solution compliant with the FHIR R4 standard. This represents an advanced case study directly relevant to improving interoperability of electronic medical records and clinical documentation in Japanese healthcare settings.
Original content copyright by respective publishers