Clinical Predictive Modeling Development and Deployment through FHIR Web Services.

Area of Science:

• Health Informatics
• Computational Medicine
• Software Engineering

Background:

• Clinical predictive modeling faces challenges in both developing and deploying models.
• Integrating electronic health records (EHRs) with predictive models is complex.
• Standardized data formats and interoperability are crucial for clinical applications.

Purpose of the Study:

• To demonstrate a software architecture for developing and deploying clinical predictive models.
• To leverage the Health Level 7 (HL7) Fast Healthcare Interoperability Resources (FHIR) standard for clinical modeling.
• To enable efficient model development using EHR data and deployment for patient scoring.

Main Methods:

• Developed a web service architecture integrating OMOP Common Data Model (CDM) databases with HL7 FHIR resources.
• Utilized MIMIC2 ICU and a synthetic outpatient dataset, transformed into OMOP CDM, for model development.
• Deployed predictive models as FHIR resources capable of receiving patient data, performing predictions, and returning scores.

Main Results:

• The developed system successfully enabled clinical predictive model development from EHR data.
• Deployed models integrated with FHIR resources provided patient-specific prediction scores.
• Evaluated web services demonstrated a practical and reasonably fast performance, with a one-second response time per patient prediction.

Conclusions:

• The proposed software architecture effectively addresses challenges in clinical predictive model development and deployment.
• Utilizing HL7 FHIR standard with OMOP CDM offers a viable approach for integrating predictive models into clinical workflows.
• The system's efficiency suggests its potential for real-time clinical decision support.