Why do we focus so heavily on structured medical data?

Two years ago, we started with a simple but powerful conviction: care providers should spend less time behind their screens, and more time with their patients. This was the first step in the right direction. But not the only one.

The real problem: medical data is unstructured!

Cavell listened during consultations and automatically generated medical reports based on what was said. The care provider conducted the conversation, Cavell wrote, in free text. Afterwards, the physician only needed to review and validate. No more typing while the patient talks, no more catching up in the evening.

In our survey of more than 110 care providers, we saw clear results. Four out of five physicians indicated that their contact with the patient improved. They describe their work as “faster”, “less burdensome” and “administratively lighter”.

But the survey among these users also revealed something else. Converting a consultation into free text is only half the work.

More than 80% of all medical data consists of unstructured free text. Clinical notes, discharge letters, consultation reports; all written in plain language, stored in documents that machines cannot simply read or process.

This causes a cascade of problems. Medical data must be:

• Recorded during consultations, after surgeries, during nursing rounds
• Processed when preparing prescriptions, certificates, lab requests or discharge letters
• Consulted when preparing for a consultation, creating a treatment plan or reviewing a medical history

As long as all that information is in free text, every step must be carried out manually by a person who reads, interprets and re-enters the data. Slow, error-prone and difficult to scale.

Structured medical data means that clinical information is captured using international standards, so that systems can automatically read, process and exchange that information.

What is structured medical data then?

There are many important internationally accepted standards. Standards for coding medical concepts, such as SNOMED CT for example.

SNOMED CT is a large medical dictionary of more than 350,000 clinical concepts, each with a unique code. “Diabetes mellitus type 1” gets code 46635009, “ex-smoker” becomes 8517006. Every diagnosis, symptom, procedure or observation has a precise, unambiguous identifier.

In addition, there are standards that allow these medical codes to be structured, such as FHIR, so that we can add relationships, narratives and more to these medical codes.

FHIR is a framework for structuring medical data into interconnected resources: Patient, Condition, Observation, CarePlan, Medication, and more.

When these standards are combined, a simple note like “55-year-old patient with stabbing headache, probable diagnosis migraine, brain CT scheduled” can be converted into a set of structured, machine-readable data points, searchable, reusable and exchangeable across systems.

The benefits of this are clearly significant:

• Easily searchable: find all patients with a specific diagnosis in seconds
• Automatically processable: trigger workflows, alerts and prescriptions without manual intervention
• Interoperable: share data seamlessly between hospitals, general practitioners and specialists
• Ready for research and AI: build population analyses and clinical AI models on consistent, clean data

Why doesn’t this happen automatically yet?

These internationally recognized standards have existed for decades. The question can therefore be asked: if the benefits are so clear, why don’t we do this?

The answer is very simple!

“Entering structured data manually is a nightmare!”, testifies virtually every care provider in the world.

SNOMED CT has 350,000 codes. Correct coding requires specific training that most physicians and nurses have never received. And even if they have, selecting the correct code during a busy consultation adds administrative burden, exactly the kind of friction that kills adoption. There are tools that try to solve this. Structured forms allow care providers to enter coded data via fixed input fields, but that typically brings more administration, not less. Regular AI scribes do reduce the administrative burden for the care provider, but often generate little to no structured medical data.

The conclusion is clear: structured data will only be captured at scale if it happens automatically, intuitively and without extra burden on the care provider, during the care moment itself.

And that is exactly why we go a step further with Cavell!

Cavell automatically generates structured and coded medical data from a consultation, without any extra actions from the care provider. But it is not just about generating that data. It is about what happens with it afterwards.

From free text to coded data points. When a physician says for example “patient has had stabbing headache on the right side for several weeks, most likely migraine without aura”, Cavell processes that not as text for the conclusion, but as meaning. The system recognizes the diagnosis, links the correct SNOMED code to it, and automatically registers it in the right place within the EHR. All formatted according to the FHIR standard. All ready to validate.

Structured data tailored to the EHR. The structured data is directly aligned with the structure of the EHR in which the care provider works. There are many different electronic health records on the market in Belgium and Europe. Cavell adapts the data model per context (per specialty, per software environment) so that the necessary information ends up exactly where the EHR expects it. No manual transfer, no loss of structure upon import.

What this makes possible. Once the structured data is validated and in the EHR, an entirely new layer of possibilities opens up. Automatic triggers for follow-up, population analyses across patient groups, quality indicators that calculate themselves, and more.

It should be clear that there are many problems in healthcare today. Many of those problems can be traced back to poor structured medical data. Tools like Cavell, which can generate truly valuable structured medical data without placing additional burden on our healthcare staff, will be the catalyst for more value-driven care.