How can unstructured data from administrative care records support adult social care policy and operations?

Social care faces significant funding challenges. Administrative care records provide a rich source of evidence about the characteristics and support received by social care service users. It is now possible, using machine learning, to extract structured information from this data.

This project trained a large language model to classify social care case notes as being indicative of loneliness or social isolation. This allows us to quantify the prevalence of loneliness among social care users, and to understand its impact.

The findings have implications for how local authorities commission and target services.

Using advanced AI-powered natural Large Language Models, we were able identify signs of loneliness and social isolation in care notes with 97% accuracy. 

The resultant findings are striking: 

• 44% of older people were identified as lonely or isolated during their first care assessment. 

• Lonely individuals entered care homes around 7-9 months earlier than others with similar needs.

• Loneliness was the strongest predictor of use of day centres – services designed to promote social inclusion. 

• There was also a strong link between loneliness and cognitive decline, such as memory problems. 

Beyond its findings on loneliness, the research offers a blueprint for future care data analysis. It shows how machine learning can extract valuable, structured information from vast volumes of text – saving time for social care staff, supporting better decisions, and improving the overall effectiveness of care systems. 

The open-source tool used in the study is available for other councils, researchers, and developers to explore, adapt, and apply.

The findings are being shared with local policymakers, social care providers, and the Department of Health and Social Care, with the goal of informing future decisions on social care funding and service delivery. 

London School of Economics