Accelerating Healthcare Data Interoperability with Azure Text Analytics for Health FHIR Structuring

As the healthcare industry embraces digital transformation, the challenge of managing vast amounts of unstructured data, such as clinical notes, lab reports, and medical transcripts, has become increasingly prominent. We are excited to announce the general availability of the FHIR (Fast Healthcare Interoperability Resources) structuring capability in Azure Text Analytics for Health, enabling healthcare organizations to seamlessly convert free-text clinical data into standardized FHIR-compliant formats.

In preview, this feature enabled organizations to leverage Azure's advanced AI to parse and structure unstructured medical text data, extracting valuable information and converting it to a FHIR format. Now, with its general availability, the FHIR structuring capability is ready for production-scale deployments, helping healthcare providers, researchers, and health-tech innovators drive better interoperability, streamlined workflows, and enhanced patient outcomes.

Why FHIR Structuring Matters in Healthcare

FHIR has become the standard for exchanging healthcare information electronically, providing a framework for easier, faster, and more secure data sharing across systems and organizations. However, the majority of clinical information is still locked in unstructured formats, making it challenging to standardize and share data.

 Azure Text Analytics for Health addresses this challenge by using machine learning to analyze medical text and identify important clinical information such as symptoms, diagnoses, medications, and more. With the FHIR structuring capability, the service can now translate this extracted data into FHIR-compliant resources, ensuring data consistency across various health IT systems. This enables healthcare organizations to focus on data insights and patient care without the technical burden of manually structuring data.

How FHIR Structuring in Text Analytics for Health Works

Azure Text Analytics for Health employs a powerful combination of NLP and healthcare-specific machine learning models to identify and extract medical entities from clinical text, detect relations between entities, detect assertions and link the extracted entities to clinical ontologies. The service also leverages document and section recognition models in order to map the extracted entities to FHIR resources, allowing them to be used in various applications, such as electronic health records (EHRs), health information exchanges (HIEs), and research databases.

Sample of an excerpt from FHIR resource:

FHIR’s core building blocks are known as FHIR resources, which represent discrete pieces of healthcare information, such as a patient, practitioner, procedure, or medication. Each FHIR resource includes standard data properties to ensure ease of use and interoperability across systems. These resources are designed with a modular and extensible structure, allowing developers to adapt them to various healthcare scenarios while maintaining consistent standards. They are expressed in formats like JSON or XML, enabling seamless integration with modern web technologies. FHIR also employs a robust referencing mechanism, linking resources to one another to reflect real-world relationships, such as associating a patient with their clinical observations or encounters. This structured yet flexible approach supports diverse healthcare workflows and facilitates improved data exchange between healthcare providers, applications, and systems.

Extensibility and Data Traceability:

FHIR supports an extensibility framework, which Azure Text Analytics for Health utilizes to enhance the usability of its output. For example, the service stores the offset and length of derived data within the original unstructured text, allowing downstream systems to trace the source of the structured data.

Referencing Between Resources:

FHIR resources often reference each other to create a more comprehensive clinical picture. For example, a Procedure resource might reference a Patient resource to show which patient underwent a specific procedure. These references help connect pieces of information, offering a more holistic view of a patient’s healthcare interactions.

FHIR Resource Bundles:

When multiple FHIR resources are packaged together, they form a FHIR Resource Bundle. This container is essential for clinical data integrity and interoperability, as it holds a collection of interconnected resources. When calling Text Analytics for Health FHIR capability, FHIR bundles are included in the service’s output, enabling easy data transfer and integration with other systems. For instance, Azure’s FHIR Importer function can process these bundles, importing them directly into Azure Health Data Services.

Sample of an API call that includes FHIR output:

 Getting Started

 Healthcare organizations looking to enhance their data workflows and streamline interoperability can now start implementing FHIR structuring in Azure Text Analytics for Health. The service is readily available through Azure, and organizations can access detailed documentation on the FHIR structuring capability - https://learn.microsoft.com/en-us/azure/ai-services/language-service/text-analytics-for-health/concepts/fhir.