How AI driven Norfolk project is transforming patient care

A pioneering project is using artificial intelligence (AI) to identify residents most at risk of falling, shifting the focus from reacting to a more proactive and preventative approach. Norfolk County Council in the UK has introduced the AI‑powered Proactive Intervention & Prevention programme to flag the residents most at risk from falling, enabling earlier and more targeted support.

Falls are a significant issue for older people. In Norfolk, a third of people over 65 fall every year, deeply impacting a person’s confidence, mobility and wellbeing, and costing health and social care over £4,000 per fall. Norfolk County Council (NCC) have developed the Proactive Intervention & Prevention programme to identify and help people most at risk. The purpose of the project is to alleviate the growing pressure on health and social care systems.

Nick Clinch (pictured, right), Director of Transformation at NCC, said: “This was an opportunity for us to work in a more preventative way than ever before to proactively support residents earlier, which helps to maintain independence and prevent, reduce or delay the need for more formal care.”

Harnessing AI for prevention

The success of the programme has hinged on its use of AI. Although there is data available that could be useful in identifying the Norfolk residents who would benefit from targeted, preventative support, there sheer amount of data makes it practically impossible for a person to extract any useful insights in a realistic amount of time.

Nick continued: “We have been able to develop an approach using machine learning (natural language processing and predictive modelling) to gain meaningful insight from our data and identifying people at risk of escalating needs.”

 Machine learning works by automatically assessing thousands of relationships between risks to identify which combinations of risks are the best predictors of escalating need. There are over 100 risk factors associated with falling, including previous instances of falls, recent hospital admissions, certain health conditions and whether there are recorded issues with a property’s suitability. The approach the Proactive Intervention & Prevention programme takes looks at the various risk combinations and how often they are mentioned and then predict whether the resident is at risk of a fall.

Pilots

There have been two pilots, the first used only Adult Social Care records and focused on residents with a predicted falls risk of 90% and over.  The second was done in collaboration with South Norfolk District Council, and incorporated risks obtained from data sets along with social care records. This enabled them to identify additional people who, for instance, had assisted bin collections or had received support from the Handy Person service etc and may not be known to Adult Social Care.

Nick said: “What we found with the first pilot was that by only using Social Care records, we were only able to reach those who were already known to us and knew there were many more people at risk of a fall who had never been in touch with us before. The expanded second pilot enabled us to analyse whether the types of conversations and challenges faced by residents in each cohort differed, what support was being asked for and what skills did those carrying out the connector calls need.” 

Lessons learned

To maximise the effectiveness of the AI, the programme had to ensure that the data used was relevant and accurate. Nick emphasised that using data for data’s sake is not enough, the ‘so what’ is more important. Approaching it from the angle of what do you want to achieve, how will you do this, what skills and resources are needed and how will you measure it is essential to success. Data accuracy is also vital. Nick also stressed how important it is that those responsible for inputting data are informed of the wider uses of the data they are recording.

For organisations considering similar projects, Nick offered several recommendations:

• Data capture - You must fully consider what do you want to achieve and where do you get your data from.  If you want to, for instance, look at hoarding, where would you look?  Most records would be after someone has been identified. Take it back further, what are the things in people’s lives that may lead to it and then look at where that information may be found.
• Comparable projects - To be aware of what other similar work is being done in the wider space.  There may already be data sources who can filter much of this information already.
• Different approaches - Consider a variety of approaches for different cohorts of people – one size most definitely doesn’t fit all.
• Capture success metrics - Be clear on how you will measure success and develop mechanisms to capture this. Qualitative and quantitative data are equally important, but you need to be able to interpret it all.