Driving interoperability standards in healthcare with AI

• Modern healthcare systems are increasingly producing large volumes of patient data
• Data often comes from different sources and in different formats, making it difficult to build a comprehensive patient overview
• Interoperability standards and advances in artificial intelligence are helping to make data standardization in healthcare more accessible to clinicians in a streamlined and user-friendly format

Digital innovations such as electronic health records, artificial intelligence (AI), and machine learning are driving demand for interoperability standards in healthcare and changing the way healthcare systems operate across the globe.

Digital technologies enhance efficiency, accessibility, and patient care.¹ However, digital healthcare requires a high level of interoperability so that systems can communicate and exchange data effectively across a range of platforms.²

At this year’s Healthcare Information and Management Systems Society (HIMSS) Global Health Conference & Exhibition, Dr. Vishakha Sharma, Senior Principal Data Scientist at Roche Information Solutions, discussed how interoperability standards in healthcare and generative AI can be harmonized to optimize healthcare.

Data standardization in healthcare is critical to operations, and the amount of data produced in healthcare systems is estimated to be increasing at a rate of 47% per year.³ Dr. Sharma outlines how patient journeys can result in an extremely complex data picture. “Even just for one patient, the data is coming from different sources. It is structured and unstructured. At the same time, it is sitting across different silos and different systems,” says Dr. Sharma. Different silos and systems can result in a lack of access to unified information, which could result in oversights. Clinicians must take into account a range of information, such as patient demographics, previous diagnosis, lab results, vitals, and any biomarker testing, before making any treatment decision, and this process is made even more complex by data presented in differing formats.⁴

Much has been spoken about how AI could benefit healthcare, and Dr. Sharma believes that the future of healthcare is tied to digital transformation. She noted that, “It is possible to build an AI solution to streamline data from different modalities, and present clinicians with a very comprehensive view of the patient so they can make informed decisions about therapy.” The challenge in building this type of solution lies in the disparate data standards currently in use.

Dr. Sharma explains that by data standardization in healthcare, we mean “methods, protocols, terminologies, and specifications for data collection, exchange, storage, and retrieval of information.” Standards are necessary so that it is clear what kind of data should be collected, how it is going to be saved, and how that information is going to be exchanged across different systems.

This need for standards when sharing healthcare data between systems led to the development of Fast Healthcare Interoperability Resources (FHIR) by Health Level Seven International (HL7®) – the global authority on standards for interoperability in healthcare technology.⁴ So great is the importance of these standards, HL7 has signed a Project Collaboration Agreement with the World Health Organization, which describes interoperability standards as ‘critical for consistent representation of data and information in health’.⁵ Other standards come from Digital Imaging and Communications in Medicine (DICOM), Integrating the Healthcare Enterprise (IHE), and the International Organization for Standardization (ISO).

Dr. Sharma explains how these standards are put into practice for integration purposes: “We model them in a way in which we can create conceptual relationships, even if the data is missing or unstructured in source systems, and make sure that there is a linking that happens from the start to the end for the entire continuum of that patient.” It is also important to use standard terminologies such as the International Classification of Diseases codes to ensure the data is harmonized and understood by the end user.⁶

To get to the stage where data can be standardized, Dr. Sharma outlines several challenges that must be considered:

1. Data access: Data is often sitting in different silos, so thought must be given as to how to make sure there is access to the required data at the point of decision-making by the clinician.
2. Data quality: Sometimes the information entered is not correct, or sometimes the information is not entered at all. For example, a patient might have pathologic staging, but be missing the clinical staging from radiology reports, which would provide additional information to the clinician.
3. Heterogeneity: From hospital to hospital, across countries and regions, data is collected and exists in different formats. At the same time, information might be sitting in a different data source from other systems, making it difficult to bring together a full picture.
4. Large volumes: Data is often in large volumes, so thought must be given as to how to handle these large volumes while maintaining data interoperability standards.
5. High dimensionality: Healthcare data is high-dimensional data, meaning one specific patient sample could be sparse compared to the full data set, making statistical conclusions difficult.
6. Incentives for researchers: Community-based initiatives led to current data standards, so for any new resource or new standards to be adopted, consideration must be given to researchers and if they are being incentivized in a way that promotes this crowd-based approach, and can be shared widely.
7. Keeping up with research: Scientific knowledge is increasing every day, and it is important to make sure that this is all in par with the community standards, and any new research that is coming.

Of the many challenges with data standardization, Dr. Sharma believes that the main goal should be to capture data in a structured format using agreed-upon coding. This is because unstructured data is not straightforward to analyze and requires extensive preprocessing, which poses a big challenge for researchers, as up to 80% of current data in healthcare is unstructured.⁷

This is where a combination of generative AI and human input can be utilized, explains Dr. Sharma, whose team has leveraged natural language processing and machine learning to create an AI tool to extract data from unstructured reports, which is especially helpful for multi-disciplinary team meetings. “All the complex cancer patient cases are discussed in these meetings, and to prepare you have to collect a lot of information from different systems to put in front of the oncologist to make a decision about a patient case,” Dr. Sharma explains, “There is significant time that goes into compiling and synthesizing that information about the treatment history, and the data is coming from different fragmented systems and it is unstructured.”

Dr. Sharma says that her team is working on automating case summarization using generative AI: “For the tumor board meetings, what we are doing is going through all the data that is needed, and with the help of generative AI, we are summarizing the case reports.” This AI-generated summary can be customized based on the user or on clinician preference to optimize workflows and save time.

Although technology is evolving rapidly, Dr. Sharma points out that human involvement is still critically important at every step of the process. She confirmed that, “From the design, to development, to deployment, to monitoring. For each phase of any AI solution that we want to build, it’s important that there is a human in the loop who can make sure that, from the beginning, the information is interpreted in the right way.” The combination of digital technology with specialized clinical knowledge promises to enhance efficiency, accessibility, and the quality of healthcare for clinicians and patients alike.⁸