A Standardized Clinical Data Harmonization Pipeline for Scalable AI Application Deployment (FHIR-DHP): Validation and...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Data Science

Background:

• Digitalization generates vast healthcare data, promising AI-driven clinical insights and improved patient care.
• Challenges in AI adoption include non-standardized data, poor interoperability, and limited stakeholder collaboration.
• Existing standards like Fast Healthcare Interoperability Resources (FHIR) have limited AI usability.

Purpose of the Study:

• To develop a data harmonization pipeline (DHP) for clinical datasets utilizing the FHIR standard.
• To enhance the usability of healthcare data for artificial intelligence applications.

Main Methods:

• Developed a FHIR-based data harmonization pipeline (DHP).
• Validated the DHP's performance and usability with data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database.
• Detailed the 5-step DHP workflow: data querying, FHIR mapping, syntactic validation, patient-model database transfer, and AI-friendly export.

Main Results:

• Presented the FHIR-DHP workflow for transforming raw hospital records into harmonized, AI-ready data.
• Demonstrated the pipeline's application using clinical diagnoses records.
• The DHP successfully converts heterogeneous data into a standardized, AI-compatible format.

Conclusions:

• The FHIR-DHP facilitates scalable, needs-driven data modeling for large, diverse clinical datasets.
• This approach is crucial for enhancing cooperation, interoperability, and patient care quality in clinical practice and research.
• The DHP represents a significant advancement in leveraging healthcare data for AI-driven medical applications.