When using LLMs in healthcare, semantic interoperability is key

As you build a strategy for leveraging large language models (LLMs) to support healthcare initiatives, don’t overlook the critical role of semantic interoperability.
The cover of an insight brief titled: When using LLMs in healthcare, semantic interoperability is key. The cover has a purple dragon on it (a nod to the movie "How to train your dragon".

As the healthcare industry seeks novel ways to gain greater value from both unstructured and structured clinical data, many are curious how large language models (LLMs) – like ChatGPT, for example – could help. However, as the excitement builds over how to best use augmented intelligence (AI), stakeholders must understand how these tools work and what their limitations may be. In the case of successfully using LLMs to support clinical uses like quality improvement and population health initiatives, one thing is certain though – prioritizing semantic interoperability is key.

So, what is semantic interoperability? Put simply, it’s the ability to transmit clinical concepts while preserving clinical intent. If data is semantically interoperable, it can be transmitted throughout the vast healthcare ecosystem, beyond the electronic health record (EHR), without losing its original meaning.

In our latest insight brief, IMO’s Senior Director of Clinical Informatics, Amol Bhalla, MD, and co-author Amy Loriaux, PhD, discuss the potential inherent in LLMs, NLP, and other AI applications in healthcare. But there is a catch – LLMs must be trained on robust clinical terminology mapped accurately to standard code sets to maintain semantic interoperability and optimize LLMs for real insights.

Keep scrolling for an excerpt from the brief that details how semantic interoperability helps retain clinical meaning – or, to read the full brief, click the button below.

Preserving meaning with semantic interoperability

Semantic interoperability is an important piece of foundational interoperability – the two-fold activity that preserves both format and meaning by aligning data both syntactically and semantically. Specifically, syntactically-aligned data is fitted to an underlying data model, or schema, which structures the data in a way that supports a specific use case, such as transmitting data from one system to another. On the other hand, semantically-aligned data ensures that the clinical data stored within a data model can be transmitted without losing its meaning and intent. The semantics of the data can influence the syntax, and vice versa, and both are needed for effective interoperability.

To facilitate and maintain clinical intent, clinicians often employ precise nomenclature and terminology to refer to similar but different disease states. For example, they may use double receptor-positive metastatic breast cancer, estrogen and progesterone positive breast cancer, or hormone receptor breast cancer – HR+/HER2. And while it may be easy for a clinician to understand these minor yet critical differences, these semantic nuances can be very challenging for a computer.

As a result, a tremendous amount of effort is necessary to improve that programming, allowing the computer to understand and associate activities over time. To overcome this challenge, (at least in part), structured clinical terminology, comprised of codified terms from a common clinical vocabulary, can be employed to accurately represent clinical concepts like diseases or labs. Common examples include ICD-10-CM, SNOMED CT®, and LOINC®.

Not only does the use of these terminologies – and their corresponding codes – maintain the clinical intent for the end user, it also enables information systems to exchange data in a manner that empowers the receiving system to perform automated reasoning. In healthcare, semantic interoperability is considered crucial for various initiatives such as quality improvement programs, regulatory reporting, and managing population health. It also has a significant impact on emerging healthcare models and organizations, such as health information exchanges, accountable care organizations, and patient-centered medical homes. Without semantic interoperability, the precise meaning of the clinical terms used, and therefore the original clinical intent, can be degraded or lost altogether, leading to unreliable or even invalid insights.

To understand the role of semantic interoperability when leveraging NLP, the promise and limitations of LLMs, and the importance of training these models on structured clinical terminology, download, How to train your LLM: And why semantic interoperability is the key to success.

SNOMED and SNOMED CT® are registered trademarks of SNOMED International.

Ideas are meant for sharing.

Sign up today and have Ideas delivered straight to your inbox.

Latest Ideas​

A lack of specificity in clinical documentation can negatively impact reimbursement, billing, and more. Read on to learn how to protect your
By leveraging IMO Health technology, CyncHealth has scaled data standardization and enriched patient diagnoses in their data warehouses.
Let's explore how better HCC capture can improve payer-provider collaboration, leading to enhanced data quality and more efficient workflows in healthcare.