The healthcare industry is overloaded by free biomedical text. By some estimates, 80% of healthcare data is unstructured, including clinical notes of diverse types, medical imaging reports, medical publications, and more. This health data has the promise and potential to positively transform care, but it is often underutilized and incompatible across health organizations. Free text and siloed health data may be the two most fundamental challenges in our global quest to improve operational efficiencies and enhance patient outcomes.
Today, we are excited to share a breakthrough solution to simultaneously accelerate unstructured data insights and supercharge interoperability between health organizations: Text Analytics for health structuring to FHIR. Now in public preview, this feature of Text Analytics for health enables health organizations to transform unstructured clinical documents into FHIR resource bundles, to deliver better health insights to more people, faster.
Text Analytics for health is a generally available natural language processing (NLP) service within Microsoft Azure Cognitive Services that was purpose-built to extract information from biomedical and clinical free text documents. In our newest release of this service, Microsoft will become the first cloud provider to allow customers to formalize their NLP output as bundles of interconnected hierarchical FHIR resources, in adherence with the US Core standards.
Unlocking clinical insights and analytics from unstructured data
When you think of free text, you may imagine scribbles on your doctor’s notepad or the seemingly endless forms you fill out in the waiting room. In reality, unstructured text is intertwined throughout the patient journey, in doctor’s notes, free text in electronic health records, pathology reports, medical encounter transcripts, medical imaging reports, clinical trials protocols and more. Separately, these data points are just pieces of text; together, they hold the power to transform healthcare.
Text Analytics for health enables researchers, data analysts, software developers, and medical professionals to extract information from unstructured protected health data and unlock a wide range of secure scenarios—like producing analytics on historical medical data and creating prediction models, matching patients to clinical trials, viewing population-wide trends, or assisting in clinical quality reviews. Trained on a diverse range of medical data, this service is capable of processing a broad range of data types and tasks. The service provides advanced capabilities for extracting insights, including:
- Named-entity recognition: Identify medical terms in text and determine boundaries and classification into domain-specific entities.
- Entity linking: Associate medical entities with common ontology concepts from standard clinical coding systems.
- Relation extraction: Infer and extract semantic relationships and dependencies between different entities.
- Assertion detection: Return modifiers for Certainty, Conditionality, and Association.
This powerful combination of artificial intelligence and NLP is already seeing exciting results across the patient journey. Tel Aviv Sourasky Medical Center uses Text Analytics for health in its Genetics, Oncology, and Pathology Institutes to extract genes and variants from tumor genomic sequencing reports. The results are used in a genetics decision support application, saving oncologists, pathologists, and geneticists many hours of work by cross-referencing results with knowledgebases for pathogenicity and evidence for hereditary mutations. This application helps oncologists decide whether further genetic counseling is needed.
“Using Text Analytics for health, we are able to capture tumor genetic test results, cross-reference them with up-to-date scientific databases, and share real-time genomic intelligence with physicians” says Prof. Hagit Baris Feldman, Director of the Genetics Institute and Genomic Center at Tel Aviv Sourasky Medical Center. “This AI-powered service accelerates time to treatment and helps our oncologists target the patients who need lifesaving genetic counseling.”
Catalyzing interoperability by structuring clinical notes into FHIR resources
The 21st Century Cures Act was passed in an effort to promote innovation in healthcare and requires, among other things, support of health information technology interoperability. The key to achieving full interoperability is representation of health data in the FHIR format.
While FHIR is rapidly being adopted across the globe, there are a number of gaps in how to represent unstructured information in clinical narratives. Specifically, unstructured text typically cannot be easily ingested, normalized, structured, and analyzed by healthcare stakeholders in resources generated per US Core FHIR Guidelines. Organizations adhere to the FHIR standard in the pursuit of interoperability but lose the deep context and relationships from the clinical narrative.
With the addition of Structuring to FHIR in public preview, Text Analytics for health customers and partners will now be able to structure diverse types of clinical documents to FHIR. The resources generated are grouped together into a single bundle that represents the clinical note, where all FHIR resources connect to the patient resource and reference to the exact point in the original text where the resource was generated. The FHIR output is structured in a hierarchy where resources such as observations and conditions maintain relationships with the resources representing the document sections within the bundle they came from. This allows for deeper patient insights, population-level insights, and decision support that maintains the context of the clinical narrative.
The feature supports all US Core Clinical Notes, from consultation notes to procedure notes to imaging narratives to pathology reports and beyond. Codes go beyond mandatory US-core terminologies and can be linked to 100+ coding terminologies.
Clinical document structuring ingests and normalizes data, extracts insights, and outputs FHIR resources bundled to represent the clinical document.
Foundation 29, a non-profit organization dedicated to diagnosing rare diseases for physicians and patients already uses the new Text Analytics for health structuring to FHIR feature. The organization has built Dx29, a collaborative diagnosis support platform to control genotype and phenotype analysis through clinical observation. The first step in identifying these rare diseases is symptom identification from medical notes in any language. Using Microsoft Cognitive Services and Text Analytics for health structuring to FHIR, physicians can extract information from clinical notes and cross-reference symptoms with rare disease data sets for analysis. Dx29 closes the information gap by bringing deep phenotyping results to support the physician with relevant insights and help the physician decide on the correct diagnosis and treatment.
“Text Analytics for health increased the quality of our services to the rare disease patients” says Pablo Botas, PhD, CEO at Foundation29. “We started using named entity recognition to help users build patient profiles for Dx29, detecting signs and symptoms and other entities. In the context of genetic testing, it allows automation with a quality on par to that of standard workflows, as demonstrated in a project developed in conjunction with the University of Utah health Science Center. Adding FHIR structuring broadens the possibilities because it makes the information in medical documents actionable, not only for patients and caregivers, but also for engagement with research, industry and, especially, regulatory entities.”
Leveraging Text Analytics for health in downstream cloud-enabled implementations
Unlocking the power of health data with Microsoft Cloud for Healthcare and its expanding pipeline of services like Text Analytics for health allows caregivers to gain a holistic view of the patient with insights and actionable next steps for more informed, personalized care management.
The new structuring to FHIR capability within Text Analytics for health is a critical first step to extracting insights from unstructured clinical data. Using this service, unstructured text such as clinical summaries, imaging reports, and pathology reports can be converted into FHIR resources. These FHIR resources can be ingested directly into Azure Health Data Services, which allows for this data to be persisted alongside structured clinical and imaging data. This makes downstream analytics and connections to other parts of the Microsoft ecosystem seamless. Now generally available, Azure Health Data Services is a unified solution that helps protect and combine health data in the cloud and generates healthcare insights with analytics that can help improve health outcomes.
Organizations can also now ingest unstructured data for metrics and visualization using the tooling of choice, such as PowerBI or Synapse. For example, Azure Health Data Services offer a PowerBI connector for FHIR that queries the FHIR server directly and creates analytics that can be leveraged to extract clinical insights at the patient or population level. Customers are in control of how to use output from Text Analytics for health structuring to FHIR, whether for interoperability with other health systems, or for further processing/analyzing of the data.
Getting started with Text Analytics for health structuring to FHIR
Healthcare organizations can now use Text Analytics for health to seamlessly unlock interoperable insights from unstructured biomedical data, including clinical notes, medical publications, electronic health records, clinical trial protocols, and more.
With our newest release of Text Analytics for Health, we are also excited to introduce tiered pricing to make processing of unstructured clinical documents at scale more affordable. Users of Text Analytics for health will be able to use the new structuring to FHIR feature at no additional charge.
To get started or learn more:
- Sign up to access the Text Analytics for health structuring to FHIR public preview
- Review Text Analytics for health documentation
- Learn more about Microsoft Cloud for Healthcare
