AI Can Bridge the Health Data Interoperability Gap

In a bustling hospital, a healthcare professional is tasked with uploading data from electronic medical records (EMR) into a public health reporting system. This data is vital for enabling the swift detection of disease outbreaks and ensuring the rapid responses necessary to contain them. Unfortunately, the hospital’s EMR data formats and structure don’t match those required by the reporting system, and making the conversion from one to the other is a complex, error-prone undertaking that requires significant time, concentration, and expertise. Colleagues at other healthcare facilities face the same hurdle, and the result, in practice, is that the data in question is uploaded only occasionally, if at all.  

This is a serious problem. In public health, complete, accurate, and timely data is essential for effective monitoring and response. Without it, an outbreak that may otherwise have been quickly detected and contained could spiral unchecked into the next global pandemic. To head off that alarming possibility, healthcare and public health’s data interoperability issue must be addressed as quickly as possible. Large language model (LLM)-based AI tools appear well-suited to taking on this challenge and are ready to be explored as the basis for a near-term solution. 

In a world where health organizations are constantly balancing limited resources while also accomplishing tactical, data-dependent tasks, difficult choices become the norm. Whether it’s regarding time, funding, or staffing, every decision must be strategic. This too often means that even important initiatives—such as interoperability—end up falling by the wayside. After all, when faced with the immediate need to care for patients, it’s understandable that complex public health data conversion tasks might take a backseat.

For healthcare and public health to truly thrive, comprehensive and reliable data is key. Seamless exchange of health information across the system is essential for coordinated care, effective monitoring, clinical decision support, and rapid response to public health challenges. Accurate, complete, and timely data not only supports these functions—it is critical to understanding the needs of the nation’s most marginalized communities and effectively deploying resources to achieve equitable health outcomes.

Much progress has been made on the path to interoperability through valuable efforts in both the private and public sectors. However, resource limitations often create gaps in data quality, resulting in incomplete or inaccurate information and hindering integration across the health sector.

Now imagine that same healthcare professional tasked with uploading EMR data. This time, when they upload the data, there is nothing more they need to do. No restructuring, no cleaning, no error codes. They simply take the data in whatever format it’s in and send it to the public health system, where an AI-powered semantic layer standardizes it to the necessary format and ingests it into the receiving system.  

This example illustrates how the power of AI can be harnessed to support interoperability efforts. The concept is not far-fetched. AI’s ability to grasp and generate complex human language, ideas, and creativity is well-documented. AI models can take abstract thoughts and transform them into vivid images, detailed narratives, or even functional code. Applying this capability to health data makes sense. If AI can correctly interpret intricate expressions in human language—capturing nuances, context, and intent—it stands to reason that it will also be able to accurately unravel the complexities within diverse health data formats.

AI-assisted data transformation and standardization could help bridge difficult gaps, empowering health organizations to drive better outcomes in their communities. Importantly, such an approach would reduce the burden of data sharing, thereby encouraging greater participation in interoperability by entities that are resource constrained.

There are already robust examples from literature that exhibit the potential of AI to transform health data:

• One study assessed the potential for LLMs to take clinical narratives in heterogeneous structures and format them into the Fast Healthcare Interoperability Resources (FHIR) standard. The report achieved exceptional accuracy, showcasing LLMs’ ability to bridge discrepancies in data formatting and enhance interoperability. 
• Another study demonstrated how natural language processing technology can extract unstructured EMR data and map patients’ chief complaints to International Classification of Diseases codes, making the data easier to analyze and use. Despite its high level of success (88%) in mapping to the correct code, the study posited that with a greater sample of data, semi-supervised learning and language models could further enhance accuracy. 

The promise of AI for improving interoperability lies in its ability to reduce the manual burden that has long hindered data sharing in healthcare. By automating data transformation and standardization, AI can free up resources, allowing health officials to focus on the more immediate aspects of their missions. This is not to say that this approach replaces current efforts towards interoperability, where data is collected, stored, and reported in aligned formats and structures. Nor will it solve all interoperability challenges. It’s not a panacea, but the approach does help overcome a set of challenges that would significantly enhance surveillance, monitoring, and response systems in the very near term. It may also allow data teams to increase focus on making progress toward aligning data standards.

The road to making AI-powered interoperability a reality will not be without its own challenges. Using AI for data exchange raises concerns regarding sharing and protecting personal health information and preserving the overall accuracy of the data exchanged. While previous use cases exhibit AI that can be accurate over 90% of the time, this level of accuracy typically involves time and resources to achieve. All of this demonstrates the need for technology systems to be connected to robust workflows and processes designed to mitigate risk and identify errors early on. As with all health IT systems, any AI-centered approach will need to address privacy, data sharing, and data use concerns to be developed and deployed in a way that enhances trust across healthcare and public health systems.

Healthcare and public health systems share the same core goal: helping individuals achieve optimal health outcomes. AI-enhanced interoperability can help expedite data exchange of non-standard data during public health response. It can also enhance standards-based data exchange to better inform understanding of the needs of the nation’s most marginalized communities. These insights will enable the right resources to be deployed to those in the field who are working to achieve equitable health outcomes.