Thoughts from a day at Data + AI

About Data + AI, a two day event (free) from the people at Digital Health. 

I attended the first day to build my understanding of work in this area and see where it might offer opportunities for our network.

I started the day at the keynote on transforming healthcare through data and research.  Dr Claire Bloomfield (of NHS England) started us off with what was one of the recurring themes around moving from a model of data sharing to one of data access. 

This places trust and security at the centre with a view to avoiding some of the issues seen around Care.data.  You can read more about the Powered by Data campaign on the old NHS Digital Data pages. 

Next came Dr Richard Scott from Geonomics England.  He took us through progress since the launch of the 100,000 genomes project in 2013 through to the 2022 publication of Genome UK 2022:2025 implementation plan for England. 

Evidence is that if properly informed people will consent (90% do) to including their data in what becomes a loop between research and clinical practice.  There is an active question about whether we should sequence everyone at birth.  There is a case with over 200 conditions now detectable genomically versus the ten or so we check for at present.

A quick rush across the floor for the next set of keynotes this time on unlocking the future of AI in the NHS.

Dr Katharine Halliday – President of the Royal College of Radiologists gave an interesting overview of how they have been getting involved in what she described as a wild west of products being offered. 

Notable was that during Covid a variety of unusable tools were developed because there was limited or no input from clinicians.  They are aiming to create a registry of AI tools in NHS imaging to help bring some form of order.  There are serious barriers to implementation not least a lack of evidence of effectiveness.  Keeping clinical teams abreast of research on outcomes from AI use might be welcomed.

The other speaker was Dr Stefan Zohren who like many of the speakers had multiple hats across academia, the NHS and spin off companies (Favom in his case).  He gave an illuminating illustration of where AI could support predictive improvement around the case of Expected Date of Delivery in pregnancy. 

The calculations for this are based on data that has error baked in.  Using the multiple data points available around pregnancy they were able to construct a far more accurate prediction.  This supports safety by predicting premature birth (88% of these were caught) and also allows for much more accurate staffing planning for maternity units who can have greater confidence of when they might be busy!  A (very) brief lesson on how transformers work in AI followed.

Post lunch I went to what was badged as a Masterclass: the future of generative AI in health and care. In common with other sessions the time was a bit short for the material available.  I was particularly taken by the talk by Dr Umang Patel – CCIO for Microsoft UK. 

There were important points about the shift that happened around the arrival of patients armed with print offs from the internet and what that might look like in an age of online medical generative chatAIs.  The importance of the difference that users with professional knowledge would make versus the enquiries of non specialists felt important. 

This was nicely illustrated by a chat I had with a company called Facts and Dimensions who had a stand.  They take publicly available datasets and bring them together to support research and analysis.  They are developing an AI interface to allow natural language interrogation rather than needing SQL expertise.  All the data in the world is no use if you do not know what you are looking at and how to extract it. 

Two other points I took from Dr Patel were a lovely quote "We went from being care workers to being information workers" and his view that if you ask your Chief (Clinical) Information Officer for a brief catch up you are likely to find they offer you more time (see full quote here) as they tend to be the ones chasing people rather than the other way around.  Something worth putting to the test!

I went to a number of other talks including on Cogstack (focussed on getting AI really good at reading) and on workforce development where Northamptonshire are building a Digital Skills Academy. 

Overall the sessions proved illuminating on some of the areas we can expect to see progress on.  All the speakers were keen to see us moving beyond the questions of infrastructure and onto safe and appropriate use.  There was also a strong emphasis on improving diversity within the data recognising the dangers of bias from our existing data sets.  As a whistle stop through some of the developing areas of work it was a worthwhile event.