The Propel programme has been increadibly valuable for us. We learnt much about the NHS, met some great people and crucially disovered an amazing use case for our NLP models, which we hadn’t anticipated before joining the programme: Transforming free-form text (letters & secure emails) into structured patient data AKA “coding”.
The problem
In an ideal world, the NHS and Local Governtment would operate fully integrated IT systems, across primary, secondary, tertiary and social care. Despite the best efforts of NHS Digital, we’re still a long way from this. In the meantime, a lot of data is transferred through secure emails and letters.
For example: following a referral, a consultant will often send a letter or secure message to the patients GP, explaing the outcome of the investigation or treatment plan. This information contained within this letter need to be entered into the patient’s records at the GP practice.
The Light bulb moment
We realised that we can use the same underlying AI that we use for self-service to automate this process. I say “we realised”, but this isn’t strictly true, the potential was actually identified by the GPs, and secondary care specialists on the Propel programme. I explained how we use AI to make sense of unstructured text, and a GP asked “could we use this for our coding?“. We decided to look into this…
How it works
Currenly we use our AI models to answer user’s questions - ask a question, point Viko at some data and we’ll generate an answer - it’s reactive. However, there’s nothing that prevents us from taking a pro-active approach:
We can point Viko at the contents of a letter or email and ask “which procedures were performed?”
We can then apply a process known as entity linking to transform the answer into structured data e.g. a SNOMED-CT, RxNorm or LOIN code.
Finally we can submit this structured data into a clinical IT system e.g EMIS or TPP.
Status and next steps
We’re now working with several GP practices and PCNs to understand in detail how they work, and how we can best support them. We’ve developed a basic POC, which was well received. We’ll continue iterating until we have a suitable primary care MVP. Alongside this we’ve embarked upon the regulatory journey that will enable us to supply to the NHS at scale.
The next step will be pilot deployments with the primary care organisations we’re working with. This will allow us to gather the evidence needed to support scale-up.
In the long term we want to go beyond primary care, applying the same technology underlying to the secondary and social care sectors. 🚀