Exploring the Advantages of Generative AI for FHIR? Compliant Health Data Systems Part: 1

The advent of modern technology concepts and the emerging growth in artificial intelligence (AI) have unlocked a world of possibilities for improving healthcare data objectives. Here we have members of ASSYST’s Hephaestus Product Development Team; Padmaraj Viswanathan , a Solution Architect and AI/ML Engineer at ASSYST's Green Accelerator Program for Hephaestus, sits down with John Kimberl , an experienced Health Data Solutions Analyst to exchange ideas, explore the possibilities and delve into the potential of Generative AI and its role in facilitating Health Data Interoperability. The exploration of this topic can uncover ideas which support crucial health data operations for smooth data exchange, informed decision-making, and improving patient-centered care and public health initiatives.

John: How can Generative AI help explore data in a health data platform?

Raj: Generative AI can provide several benefits for data exploration in a health data platform.

Generative AI models have the potential to generate synthetic data that closely resembles real patient data. We find this can be quite useful for supporting the development and testing of healthcare information systems, conducting academic research programs, and testing scenarios without the risk of disclosing patients’ data or violating privacy laws. This augmentation technique can help address security concerns and mitigate the risk of re-identification when sharing, analyzing, and testing health data platforms that contain sensitive health information. We believe this can also be useful when conducting pilots on health systems and experimenting with synthetic data, enabling efficiency in the validation process for systems performance in analyzing and visualizing healthcare data and examining trends in public health without using PHI.

We can also explore the use of generative AI for predictive analysis and early detection in healthcare initiatives at the individual patient and population levels. Leveraging pre-existing data sets, generative AI models can be used to create synthetic samples of normal characteristics and behaviors based on average values in a health dataset or aggregation. In addition, we can use generative AI models to execute data cleansing and data quality assessments to assemble complete data sets by filling missing fields based on correlations in similar data sets to complete a more well-rounded statistical model. Once the generative model has been created, we can train the model to detect any outlier data points and flag them as anomalies which can be crucial for diagnosis, early preventative action, or identifying trending directions of public health activity. These statistical models can be altered to reflect progression models based on de-identified historical data for calculating growth in an occurring phenomenon with possible outcomes to support analytical decision-making processes.

Generative AI models can assist in processing and analyzing complex healthcare datasets. Healthcare data is diverse, consisting of various structures and formats and, in many cases, creating large samples of unstructured data. Natural Language Processing models can abstract text information through various documents and datasets to create meaningful sense for decision makers, something that can be very useful with international healthcare and differing languages. With the rise of Self-Service BI capabilities can create excitement for everyday healthcare professionals and non-technical staff to perform their search queries without proficiency in SQL or take advantage of Bots for guided exploration. This can be paired with data visualization tools. By generating representative samples from the data distribution, they can provide insights into the underlying structure and relationships within the data. This can aid researchers in uncovering hidden patterns, identifying subpopulations, and gaining a deeper understanding of the data.

?

John: How can generative AI help with maintaining metadata standards with the most recent version release?

Raj: Generative AI can provide valuable support for maintaining compliant metadata standards in many ways.

The biggest issue with establishing a standard for data interpretation is discovering an effective method to transform large volumes of health data into a standardized format that can be used for enhanced data sharing and to create relational models for interpretation. Healthcare data can come from various sources with different document structures and formats. Harmonizing datasets can be a tricky task that requires a high degree of validity, creating a time consuming manual task that is still prone to human error.

We have discovered that NLP models are helpful in parsing various data sets from diverse sources and creating custom schematic formats that can be mapped to a uniform specification. For example, HL7? is continuously revising and enhancing the FHIR? specification, most recently releasing the R5 format, ultimately creating a need for organizations to stay updated with the most recent version of FHIR?. With AI, these processes can be automated with higher accuracy. We can then train these models to perform text information abstraction and pre-processing from data sets and identify and understand commonalities amongst data elements and their structure within differing standards between desired target specifications and the source schema for metadata mapping to ensure consistency throughout the transformation towards a structured standardization.

This is critical in maintaining the integrity of data sets as they are converted so that they can be accurately parsed to generate meaningful representations. With an integrated Deep Learning AI Model, the healthcare system can immediately flag discrepancies amongst the values in the source metadata schema and the re-structured data set, along with validating the new dataset for compliance with the desired standard. In addition, we can source out the failure during the conversion process and use Chatbots to convey the transformation error in a comprehensive statement to address the issue quickly.

Generative AI models can understand the semantics and context of data standards and metadata elements. This enables them to analyze relationships between different standards, identify overlaps or gaps in coverage, flag discrepancies, and provide insights on enhancing semantic interoperability across different standards and metadata schemas. We believe this can be extremely beneficial to transitioning healthcare systems to a leaner and more interoperable platform that leverages standards such as HL7? FHIR?, CDA, and ICD-10.

John: Can generative AI assist in creating FHIR? documentation or profiles?

Raj: Generative AI models can evolve to generate FHIR? documentation or profiles based on existing specifications. Consistent interaction with validated FHIR? resources enables models to learn patterns and structures across the various specifications and thus extend resources with formal definitions. With further analysis, we can elevate this learning model to create profiles that define the constraints and extensions for a specific resource based on commonly used methods for document structure.

For example, suppose you have a unique requirement within your healthcare data sets and want to ensure they are FHIR? compliant. In that case, you can use AI to generate custom schematic formats with added elements to maintain compliance and support business objectives. Generative AI models can be trained on existing FHIR? documentation and profiles to learn patterns and structures. By understanding the relationships between different elements and their corresponding documentation, AI models can generate initial templates or drafts for specific FHIR? resources or profiles.?

We can also leverage NLP models to interpret FHIR? in its raw state. Being stored in JSON or XML format for web-based sharing, it can be difficult for the average eye to parse a resource or aggregate and provide meaningful context to the data. With Natural Language Generation, we can reverse the process of NLP and use AI to identify the text and theme of the data set to provide material insight into a consumable document. This also extends to validation processes. By applying Machine Learning tactics to your AI and exposing it to various FHIR? resources to absorb the rules and constraints defined by the standard, your AI model can perform data cleansing or data quality assessments to evaluate whether a generated resource is adhering to the standard. Thus, AI will assist in identifying issues or inconsistencies in created datasets or transformations, enabling quality assurance and consistency across documentation. We are excited to see this validation process come to fruition as we witness more healthcare providers migrate their non FHIR? compliant data to standardized FHIR? and cloud platforms.

To be continued in Part 2...

Hephaestus (https://www.assyst.net/hephaestus/) is an ASSYST innovation from our Green Accelerator Program, designed, developed, and implemented for Healthcare Providers, Government Health Agencies, Health IT Companies, and Health Insurance Companies. Hephaestus is an eloquent, lightweight, low-code application platform that provides customers with the component architecture required to implement the FHIR? specification.

As a Health IT Systems Integrator and Platform Solutions provider and Gold Member of Health Level Seven International (HL7?), ASSYST helps US government health and human services agencies, research, and nonprofit organizations to deliver high-quality health and wellness services. Our customers include the Centers for Medicare and Medicaid Services (CMS), Food and Drug Administration (FDA), Health Research and Services Administration (HRSA), Program Support Center (PSC), and many State and Local Government agencies. ASSYST has been an industry leader in data interoperability, trusted data exchange, compliance, metadata, taxonomy, open data, and dissemination initiatives for over 25 years. In addition to HL7? FHIR?, ASSYST delivers solutions complying with data exchange standards, including XML, GJXDM, NIEM, XBRL, EDI, X12, and SDMX. ASSYST is an HL7? Gold Level member organization.