At HTN: AI and Data, we were joined by members of the NHS England Blueprinting Team for a discussion on the blueprinting programme, the NHS AI Lab Skunkworks programme, and a project from Gloucestershire Hospitals NHS Trust in partner with PolyGeist.
Joining us were Andrew Freeman (programme manager of the blueprinting programme at NHS England); Oludare Akinlolu (former data and technology lead at the NHS AI Lab, NHS Transformation Directorate); Sarah Hammond (deputy CIO at Gloucestershire Hospitals NHS Foundation Trust); and Bradley Pearce, (director and co-founder of Polygeist).
The blueprinting programme
Andrew began by defining a blueprint is the step-by-step process that a trust or organisation goes through when deploying a particular product, system or process. It’s not the same for everyone, he pointed out; every organisation is different. “We like to think is that blueprints can offer a degree of support from a minimal level.”
As of last month, there were 3,346 users registered on the platform. There are different levels of blueprints available, including full blueprints outlining the full detail of the project supported by relevant artefacts; blueprints on a page (BPOAP) which provide key insights, benefits and an outline of the main activities and project journeys, and technical annexes, which provide a more detailed guide in implementing technical solutions and roll-out. At present the programme has published 204 blueprints and just under 3,000 artefacts which range from issue logs to business cases to project plans.
“The ultimate aim here is for us to start to work across ICSs and across the wider health and care system,” Andrew scared. “We are working with social care, so there will be new social care blueprints emerging soon and we have already got several primary care blueprints because that’s another new area in which we’re expanding.”
NHS AI Lab programmes
Joining the conversation next was Oludare Akinlolu. Oludare shared the work taking place at the NHS AI Lab.
Starting with the current structure of the lab, Oludare explained how the NHS AI Lab is divided into five functional programmes; AI imaging, the AI in Health and Care Award, AI regulation, AI ethics, and the AI Lab Skunkworks.
AI imaging looks after anything to do with imaging, Oludare said, including supporting the development of imaging tech. The AI in Health and Care Award tests and evaluates some of the most promising AI technologies s. AI regulation convenes the key regulators to look at how we provide develop a robust and streamlined regulatory regime for AI products. , whilst AI ethics looks after the “ethical part of the life cycle of the solution, to make sure it is for good, useful purposes, and that it is ethical in terms of adoption across the whole system.”
Finally, looking at the AI Lab Skunkworks team, Oludare explained that his team helped organisations to test innovative ideas to see if they can scale or not, providing a research and development functionality for trusts to test functions before deciding their approach.
AI Lab Skunkworks
The AI Skunkworks programme ran from 2020 to 2023 with a vision of “creating an environment for organisations in health and care to test the possibilities of AI capabilities through practical experience, to potentially challenge what best meets their needs or fits their purpose.
The AI Skunkworks team achieved this vision through three practical channels: experimentation, capability, and fostering community. They invited organisations to pitch potential problems and select the ones deemed most suitable for AI solutions to solve; provide scientific capability and outsource resources to equip organisations to explore a problem, and foster a community of AI practitioners by sharing knowledge. “Every project we have done in terms of proof of concept, we have always published our code as an open source for anyone and everyone,” Oludare shared.
Oludare then explained that AI falls into two broad families: narrow AI and general AI.
“Narrow AI is what we focus on in our Skunkworks team,” he said. “It is a goal-oriented system that performs one single task and does it extremely well – for example, facial recognition.” The AI solutions currently being developed by technologies companies and within the NHS for health and care are all narrow AI.
General AI, meanwhile, is where the system is human-like in terms of intelligence, can perform simple tasks simultaneously and is more aware of its own decisions, the consequences and the impacts on its environment. This type of AI currently doesn’t exist.
On the common issues around AI, Oludare said: “There are a lot of concerns around job security, and a lot of hype around AI replacing people or taking over the job market. But it’s meant to augment people, not to replace them.”
Other concerns include trust in AI’s decision-making capabilities. “We are engaging with patients and clinicians around this,” Oludare noted. “There is a clinical power with AI, but will clinicians have the confidence and trust to use it? Then from the patient’s side, will they trust the diagnosis that AI has given them? We are doing a lot in that space to build confidence here.”
Ultimately, Oludare said, it is about identifying areas where AI could speed up work processes. “It’s making things faster and better in such a way that it’s not going to replace anyone, but it’s taking the mundane parts out of their work. That means that the clinician can focus on caring for the patient.”
Implementing AI in Gloucestershire
Sarah Hammond from Gloucestershire Hospitals NHS Foundation Trust joined the discussion at this point to share a case study from the trust’s AI work.
Starting off their AI journey, she said, there was a lot of debate around what would make the most impact. “We got it down to a shortlist in terms of some of the big problems, both for Gloucestershire, but more importantly for the wider NHS,” she said.
Concerns around length of stay made the shortlist. “Length of stay would impact on hospital flow; if we had more beds available, more patients can get through the system in a more timely way. Our feeling was that it would impact on the big problem,” Sarah said.
On moving forward to apply for support from AI Skunkworks, Sarah noted: “What was important about our pitch was that we had senior clinical involvement from the beginning, and we had stats around the impact on patients.”
Gloucestershire was partnered with a company called PolyGeist, which Sarah called “a wonderful experience from start to finish. People were very generous, both PolyGeist and NHSE colleagues. They were really generous with their knowledge. Along the way we learned a lot, and actually it was a really easy process, although the subject was quite complicated.”
Next Bradley Pearce came into the conversation to explain more about PolyGeist and how they helped Gloucestershire on their AI journey.
“We are primarily a defence company. We produce operational software for counter-terrorism, intelligence, law enforcement etcetera. This project was so successful that we’ve now branched into the healthcare market,” he shared.
The PolyGeist team wanted to predict a patient’s length of stay in absolute terms, and turn that prediction into a stratified risk score. “Scores were between one and give – one being patients that are probably not going to get admitted, all the way to five, where we are fairly certain that this patient is going to stay for 21 or more days,” Bradley explained. “There were many factors that contributed to that risk score, and we were able to build a reliable model that made those predictions.”
They took data from the trust and transferred it into a large spreadsheet that could be processed by their new model. “The system was crunching for a long time, to train a model to produce predictions,” Bradley noted. “We did that in isolation in our laboratories, and then we met up on a weekly basis to discuss the progress.”
Within 12 weeks, the system was not only operational but also integrated into the trust’s IT systems, producing results that could be seen within the EPR.
Bradley explained how it works: “Patients come in, we look at their record that is picked up by the systems at Gloucestershire, and we send that information to a machine learning system that is updated regularly. The features we identify are extracted, the model provides a prediction, and that in turn provides a risk score.” For example, if a patient got a risk score of four out a five, a patient is likely to stay between 10 and 15 days, and are at considerable risk of slipping over into 21 days, and the clinical staff can take action accordingly.
Bradley commented on the financial savings to be found for the trust if the system so much as affects a single day of bed occupancy, and added: “More importantly than that, patient outcomes are saved. Releasing people from hospital beds is a form of medicine.”
One of the key takeaways from the project from PolyGeist’s perspective was the importance of having a team inside the hospital that understands the clinical workflow and how data can drive decision-making and change behaviour. “That gave us a real ability to target particular areas in the patients’ stay where impact could be made, and build that into the technology,” Bradley said.
Reflecting on Gloucestershire’s lessons learnt, Sarah highlighted how AI can make a difference in the hospital. “It’s not a concept that people use behind closed doors and do something clever with, it can really have a difference.”
She added: “Although our project was successful, we didn’t know that until we got quite far into the 12 weeks. So it’s okay to fail. If you’ve got a good idea, it’s still okay to take it forward.” To underline her point, she commented that her team thought that they knew what the answers would be in terms of the main drivers for the algorithm, but they did not.
“Use the experience to gain knowledge for you and your team, and if appropriate and it’s related to clinical, make sure you’ve got a clinical sponsor, that’s really important,” she advised. “It’s so worth the time commitment. We made this commitment because we felt it was really important. And I’d also say, enjoy the journey, because it’s an exciting process.”
Many thanks to the team for joining us.