Clinicians turn to systems like electronic health records (EHRs) for a variety of tasks – like clinical documentation – that generate data. But data is only valuable if you know how to use it. Unfortunately, most of these systems don’t inherently know how to manipulate and interpret data so it can be used to solve clinical problems.

That’s where ontologies come into play. Ontologies – or formal, structural representations defining the relationships between different concepts – give clinical systems the ability to support automated reasoning. This function turns them into an active player in the clinical process, not just a passive repository of information.

Building blocks

So, why do we use ontologies? Simply put, ontologies are a way to represent how some piece of information is related to another. But they’re only one piece of the puzzle.

Before we can link one concept to another, we need to be able to define such concepts. That’s where terminologies– lists of concepts and their definitions – are used. Terminologies define not only concepts and can be used to tell us when two expressions are semantically equivalent – like type II diabetes and type 2 diabetes.

Next, we work with taxonomies. Taxonomies group concepts into categories. They help systems infer categorical relationships between two terminologies, like type 1 diabetes is in the same category as type 2 diabetes.

However, taxonomies do not describe how concepts are related. For this, designers use a more comprehensive model – the ontology. 

Like taxonomies, ontologies group related concepts together. These groupings can be hierarchical, with more specific concepts placed under broader ones. This allows for subsumption – or inferring that one concept is a type of another. This important function allows electronic systems to find data using a single, higher-order term rather than multiple, more specific terms, making the process of using and interpreting clinical data much more streamlined.

Practical use

In everyday clinical use, ontologies play a vital role when it comes to querying information in systems like EHRs. For example, imagine a clinician would like to find all of her patients who qualify for a diabetes screening. To do so, she searches the EHR for the term diabetes to retrieve such a list.

This seemingly simple task is actually quite a complex one, and it relies heavily on the use of terminologies, taxonomies, and ontologies. These tools let the system know that different terms for this diagnosis – like diabetes mellitus and type 2 diabetes – as well as terms associated with the diagnosis – like elevated A1C or saxagliptin 5 MG– should all be grouped around the general concept of diabetes.

If the system doesn’t know that all of these terms indicate a diabetic patient, it cannot produce an accurate search result for the clinician needing a list of all her diabetic patients.

Standardized ontologies

One ontology used in healthcare is SNOMED CT®. In SNOMED CT, concepts and their relationships are defined, or modelled.

For example, the relationship between diabetes mellitus and disorder of glucose metabolism is modeled using the is-astatement. SNOMED CT also uses attribute statements like finding site. To use the diabetes example – diabetes mellitus has a finding site relationship to structure of endocrine system. This helps to model non-hierarchical relationships.

Ontologies with attribute relationships can help systems find data that is associated with a concept but not necessarily a sub-type. This nuance is useful for finding patients with diseases that, for example, involve a particular body structure, such as disorder of the endocrine system.

Limits of the system

Ontologies are only effective if they accurately represent a given domain. Concepts that are not fully modeled aren’t as useful for making inferences. But it can be challenging to fit real-world data into a pre-existing structure, especially since ontologies are often designed with a specific use case in mind.

However, ontologies do have the opportunity to dramatically improve information system functions. While we may never be able to fully replicate the “knowledge” of a clinician, there is still plenty room for the industry to leverage more efficient and fit-for-purpose ontologies to drive smarter, easier to use and higher quality information systems.

Want to learn more about how ontological systems like SNOMED CT work? Check out our primer on the system here.