Unlocking the potential of data and AI - meet Matthew Stammers

Dr Matthew Stammers is not your typical gastroenterologist. Alongside busy clinical commitments, he strives to improve life for health professionals and patients through the latest digital technology.

Matthew is part of the Research Leaders Programme (RLP) at University Hospital Southampton. This autumn, he is stepping into a new Data and AI Lead role for the trust’s SETT Centre.

He shares his hopes for data and artificial intelligence (AI) at UHS.

How did you first become interested in using clinical data and AI?

My best friend has inflammatory bowel disease (IBD), and I have personally suffered from irritable bowel syndrome (IBS). This initially spurred me to become a gastroenterologist. 

However, the problem with traditional medicine is that it doesn’t scale. Because I wanted to help more people, I began to use data and software in 2016. This was first as a hobby and then as part of my job, and increasingly, it has become my profession. In some senses, I have become what could be termed a ‘computational gastroenterologist’. Half of my week, I see patients face-to-face. In the other half, I work on the problem of helping them at scale.

Over the years, I have learned much about computer science from online courses and informal learning. I completed Datacamp’s ‘Data Scientist with Python’ course in 2018. However, since 2022, I have studied it formally as part of my DM (doctor’s PhD). I still have about two years to go to complete this, but the NHS’s need is now, so I will have to complete this alongside my new role.

What is the SETT Centre?

 The SETT Centre is helping clinicians and patients through research.

Specifically, its Data and AI arm, which I now lead, aims to find, develop and safely scale data and AI innovations from ‘discovery’ (University) into ‘deployment’ (Healthcare Practice) to benefit patients, clinicians, and healthcare institutions.

The centre is led by Professor Chris Kipps, who I greatly respect, and managed by Rachel Chappell, without whom we could never have come this far.

What are the potential advantages of using data and AI in healthcare?

Data and AI scale are unlike traditional healthcare. This means that more patients can be cared for better, at a lower cost to the taxpayer, through the use of computers – but it has to be safe!

Logistics, administration, and decision-making are all areas that can be improved through the use of applied clinical AI. We are starting to see that come to fruition now in the visual areas of clinical practice, such as radiology, histopathology, and dermatology.

Here are some areas that SETT: data and AI is currently investigating:

• Age-related macular degeneration (AMD) is the most common form of blinding eye disease in the developed world. We are investigating the federated deployment of an AI model to pre-screen patients with AMD for potential future recruitment into cutting-edge trials. Dry AMD currently has no treatment at all.
• Extracting structured data from unstructured text using internally hosted large language models and other algorithms to create a research-grade data warehouse.
• Deploying an algorithm that can reasonably accurately predict patient length of stay across the trust to help ensure we have enough beds and can plan operations more effectively.

These only scratch the surface of what is possible, but we must start small and build up incrementally to get to that future safely. 

What are the potential challenges?

Many potential challenges are reasonably specific to healthcare, which will take longer to overcome than in other domains. As a result, typically, healthcare lags over industries by at least 15 years.

Firstly, humans are complex, not just complicated; therefore, interaction with machines/computers has unintended consequences. As a result, the bar for safety in clinical AI has to be at least as high as that for autonomous vehicles, but with many more unknowns to deal with.

Secondly, medicine is complex and only partially understood, meaning that AI’s understanding may be as flawed as our own. As a result of the above, the data itself is often challenging to prepare, nearly always requiring clinical input to get it right, and this is time-consuming at a time when clinical resources are already over-stretched.

Other potential challenges include physical hardware constraints, high cloud-compute costs, lack of data access, lack of code/knowledge sharing, hidden bias and information governance, ethical risks, and regulatory barriers. Additionally, most surveys suggest that around 95% of clinical staff are sceptical about clinical AI. These factors will all take time to address, and as a result, I take a very long view of this journey, which will take at least 20 years to play out and will require the utmost transparency and integrity to reach.

What do you hope to achieve in the SETT Data and AI theme?

I want to gently start changing the culture of UHS to make AI/machine learning part of life at the trust in a way that benefits patients and clinicians alike. This necessarily means being careful, deliberate and level-headed.

Initially, this will be in small ways, but these will grow into more comprehensive solutions over time. To achieve it, we must avoid corporate capture and become a beacon of open-source sharing. We must become a place of both learned and lived experience that attracts many talented people.

I will be happy if we can achieve co-development, maintain transparency, and deliver tangible benefits to patients and clinicians, which will scale across the NHS and internationally.

If I learned anything over the last three years at the trust, it is that legacy is left in people, not in things. The trust's values are: 1. Patients first, 2. Working together, and 3. Always improving. If I can lead a team embodying those values, then we should eventually get to the correct destination.

You are invited to reach out to Matthew if you have any ideas or want to discuss projects. You can reach him at matthew.stammers@uhs.nhs.uk.