Maximising the impact of social prescribing through a case-finding tool using machine learning

Evidence suggests that social prescribing can lead to a range of positive health and wellbeing outcomes by supporting individuals to take greater control of their own health.

However, social prescribing resources are limited and need to be targeted on those patients who would benefit most leading to a positive impact for both individuals and the healthcare system. Arden & GEM’s Advanced Analytics Unit has developed a case-finding tool, based on a machine learning model, to enable social prescribers to proactively target patients in advance of a GP referral.

The challenge

• 20% of patients consult their GP for problems which are primarily social rather than medical concerns.
• Social prescribers aim to help patients by referring them to a range of local voluntary, community, faith and social enterprise groups (VCFSE) which can help alleviate some of their social, emotional and practical problems. However, low numbers of social prescribers make it difficult to be proactive in identifying patients who could benefit.
• By offering a holistic approach to healthcare, social prescribing has shown promise in helping to reduce pressure on NHS resources and improve patient outcomes. Schemes have shown average reductions of 28% in use of GP services, 24% in A&E attendances and statistically significant drops in referrals to hospital.

Our solution

Bedfordshire, Luton and Milton Keynes (BLMK) Integrated Care System (ICS) approached the Advanced Analytics Unit (AAU) to design and develop a dashboard capable of identifying patients who may benefit from social prescribing and enabling users to measure and monitor service uptake.

A key feature of the dashboard is a case-finding tool, which utilises a machine learning model to effectively highlight and triage patients most likely to be suitable for referral to a social prescribing programme, without having to wait for a GP referral.

Social prescribing is a key part of the system’s wider ambitions to make personalised care business as usual. Providing the best tools and analytics possible to the workforce supports this ambition and enables them to make a difference.

What we did

AAU built a tool designed to assist social prescribing and primary care staff from several providers, including the VCFSE sector, in BLMK.

By leveraging and linking data from multiple primary and secondary care sources, within the data warehouse, near real-time information about the entire population of BLMK was available. The data was then used to create a Tableau dashboard showing key performance and prevalence data helping teams, practices, PCNs and the ICB to understand and monitor social prescribing uptake across different cohorts. These metrics also enable social prescribers to examine uptake and ensure equitable access across population groups.

Utilising machine learning
The team developed a machine learning model to calculate the likelihood that an individual would be referred to social prescribing based on historical patterns in social prescribing referrals. This score is then presented within a case-finding tool in order to identify and prioritise new patients who may benefit from a referral. The scores are shown alongside event-level information for health, social, economic, demographic and care utilisation to support decision making.

The tool has been iteratively developed with user testing to include:

• Count data on social prescribing referrals made and accepted at both practice and patient level
• Alignment of the case-finding tool to the clinical risk factors frequently used by social prescribers
• A patient view which includes information on demographics, biometrics and address details.

Current and potential benefits

• Enables social prescribers to case-find and reidentify patients who would benefit most from social prescribing and the types of patients most likely to accept social prescribing referrals.
• Gives system stakeholders the ability to identify trends and patterns in social prescribing referrals and referral declines.
• The tool is being used to find manageable caseloads, equipped with comprehensive near-real time information about patients and their wider determinants of health in a single view.
• KPI support in monitoring and quantifying the impact of social prescribing, evidencing the benefits and easing the burden of healthcare utilisation in primary and secondary care settings.
• Utilising a machine learning model generates a single likelihood score that can be used for prioritisation and targeting, reducing the need for social prescribers to trawl through multiple, disparate records.