Predictive analytics: responding to a post-COVID-19 world

Problem to be solved

Like most local authorities, Worcestershire County Council has been grappling with how it can provide support to its adult social care users with a constrained budget. Might it be possible to leverage the large quantities of data generated by assistive technology to target and prioritise this support more effectively? Specifically, how might data from remote monitoring devices be harnessed to track service users’ usual behaviour patterns, identify disruption to these patterns and predict corresponding changes to their care needs?

The project

The pre-COVID-19 plan

In 2019, Worcestershire County Council was awarded an NHS Digital Social Care Pathfinders Programme grant to help develop and implement a predictive tool using assistive technology data, adult social care assessments and NHS Secondary Uses Service (SUS) data. 

The AI model analysed data from wearable and stationary sensors - inside and outside of people’s homes - to make predictions about individuals likely to require hospital admission within two days and increased social care packages within five days. Working in partnership with Midlands and Lancashire Commissioning Support Unit (CSU) and PredictX, the council had just reached a point of reasonable confidence in its model when COVID-19 struck.

COVID-19: the challenge and the opportunity

The social distancing and lockdown measures introduced nationally in response to COVID-19 significantly altered people’s circumstances and behaviour patterns overnight. What might have previously been flagged as someone unexpectedly not leaving their house for several days was now the new normal. Whilst inactivity in someone’s kitchen might have previously flagged concerns about their reduced food intake, this lack of cooking might now be accounted for by food parcels delivered to a person shielding.  Given such a dramatically changed context, the council considered whether an AI model that had been trained on what was now potentially unrepresentative data was still valid. And even if it were still valid, would its outputs still be applicable for a care system likely to look very different from before?

The council decided to rapidly re-scope its plans and through this take advantage of COVID-19-induced easing of data sharing restrictions. This mandate to share information more widely within the health and care system offered an unprecedented opportunity to explore what was possible.

The post-COVID-19 plan

Retaining its partnership approach with Midlands and Lancashire CSU and Predict X, the council is now examining how it can combine datasets across adult social care (care packages and contacts with social workers), mental health, maternity and community services, together with publicly available data such as indices of deprivation. Using this wealth of data, they aim to generate predictive analyses which identify local neighbourhoods where demand for social care services will be greater than in others - both in the immediate context of COVID-19, and beyond.

In the short-term, this analysis will provide another tool to help enable more personalised and effective care by helping social workers target community interventions in specific geographical areas. These interventions could include:

• sharing information with specific ‘at-risk’ groups
• putting in place preventative measures to limit the impact of local COVID-19 outbreaks
• acting rapidly to mitigate risks when local outbreaks do occur.

In the long-term, this predictive analysis will help with a health and social care re-balance from reactive management of people’s conditions to proactive prevention or delay in the introduction of long term care needs. 

The nature and emphasis of social work is changing. This predictive analysis could help inform how the council plans strategically for a demand profile that looks very different to what existed before - in terms of geographical demand, individual needs and resource requirements. It could also help the council develop and implement local approaches to market shaping and commissioning.

Lessons learned

Even with data sharing made easier from a legal perspective, there have been practical barriers a-plenty. There is still lots of work to do before information exchange between organisations is commonplace and streamlined.

Whilst common commissioning approaches often involve highly detailed conformance-based specifications, this project has needed something different. Given both the explorative nature of AI projects and the small number of AI companies who operate in the social care sector, the council opted to develop an outcome-based procurement specification, in close collaboration with PredictX as its supplier.

To find out more about this project, contact Nicky Kirkland on email:  nkirkland@worcestershire.gov.uk