Will

Will is working on creating an AI-powered EHR system. He has a background in General Practice and has been designing and building PMS systems for the last five years. He also works with a large PHO, assessing new Health IT tools.

When Will got in contact with me via email, he told me that he too, was working on a "Totally radical PMS system which seeks to redefine how systems of this kind work". Naturally, I was super interested and had some guesses at what it would be, but I never expected what it actually was.

"The system I'm working with is phenomenally clever AI. It writes its own software basically. You tell it in English what you want it to do, and it does it."

It's being built by an international co-operation, and they've invested a "few million" in it so far, so they are very serious about it. To first understand why it's worth this much investment, we need to look at what problem it would solve.

Why are EHR systems hard to build?

Will pointed out that fundamentally EHR systems are not that complicated - their functionality isn't that different to the services that Google can provide very effectively for free. However, he claimed what made them complicated was disagreements in workflow.

"Oh I can tell you why they're hard to build, that's easy - It's because General Practitioners Argue all the time. And they all think their way is the right way to do things, and they won't listen to anybody else."

There are many different ways to record a blood pressure. Some record it as "bp 120/70", or "bp 120:70", or even just "120/70". Different systems have different ways to represent information, and healthcare professionals have their own preferred ways. Some systems have specific places to record blood pressure, in some you just put it in the text notes.

The reality is that both existing software, and medical professionals are inconsistent in how they record healthcare information. The AI system will hopefully take unstructured healthcare information, and structure it. It doesn't matter how you represent a blood pressure - it will recognise it as one, as it recognises context, and store it consistently

The potential impact of a technology like this is huge. It would allow GP's to work in whatever way they choose, while still maintaining data consistency. A large issue in healthcare is 'Interoperability' - basically a measure of how will one system can interface with another. Shifting a health record from one system to another is a huge challenge because both systems will structure the information in different ways. An AI system could provide an automatic way to format a record for a new system. It also would make it much easier to do large-scale statistical analysis of records - as the data is already consistent, and cleaned.

Their vision of the future is appointments where the GP doesn't interact with a computer screen - the software listens with a microphone, understands what the doctor is saying, and writes the notes for them automatically.

Challenges

Of course, building a technology like this has challenges. Apart from the technical issues, convincing the world that it's safe is another thing. However, Will seemed optimistic, claiming their system is truly three generations ahead of anything on the market.

What makes a good EHR system

We also discussed what qualities EHR systems should have, and he had a few points to make.

Centralised Data Repository

EHR systems are defined by a single record which all parties access - a centralised repository of healthcare information. However, this creates two issues - how do you store the data, and how do you access the data. The format which you store the information in needs to be specific and consistent enough that it can fit all use cases. The biggest issue is trusting the person who stores the data - they have a tendency to view it as 'their data'.

Extensibility

Extensibility is necessary for an EHR system as requirements will change over time. The PHO he works for has been wanting to move away from MedTech 32. They assessed many different options, but in the end decided on Profile for Windows. He described profile as 'rubbish' - but it was his second choice, because of how extensible it is. It simply presents the data, in an HTML based shell. In 15 minutes he could start changing the interface and making it fit their needs. It's not complicated to make it feel good, look good.

He thought healthcare systems should be designed as composable components, which provide high-level instructions which can be configured as needed.

Key Takeaways