Understanding adult social care using Large Language Models with administrative records

Rickman, Sam (2024) Understanding adult social care using Large Language Models with administrative records. PhD thesis, London School of Economics and Political Science.

Text - Submitted Version Restricted to Repository staff only until 20 March 2026. Download (7MB)

Abstract

This thesis explores Large Language Models (LLMs) for addressing two related challenges in adult social care: the scarcity of information about care recipients with the highest needs, and the significant administrative workload on practitioners documenting service delivery. Using the administrative social care records for 3,046 older adults who were receiving care in a London local authority between 2016 and 2020, the thesis evaluates LLMs for improving access to unstructured data in care records, and for reducing the administrative burden through automation. The research demonstrates that LLMs can extract valuable information from rich, free text social care case notes — an important development, as survey data often fails to capture those with the highest needs. LLMs can generate a structured indicator of loneliness among older adults, outperforming traditional natural language processing methods. Using this indicator in a survival analysis finds loneliness is a significant predictor of care home entry. These results suggest that, when combined with existing statistical methods, LLMs can improve our understanding of the factors influencing service use in social care. In addressing LLMs for reducing the administrative workload, the thesis evaluates gender bias in LLMs used for automating documentation tasks, such as generating or summarising case notes. This reveals meaningful differences in gender bias among state-of-the-art language models, emphasising the need for rigorous evaluation when integrating LLMs into social care practice. As one of the first studies to explore LLMs in social care, this research concludes that LLMs are useful for improving access to information in administrative records that is otherwise unavailable. However, this requires a substantial investment of time and expertise. LLMs hold promise for reducing administrative burdens in social care, but their implementation must be approached cautiously. Ethical and practical considerations around accuracy and bias are essential to ensure these innovative models are used effectively and equitably in social care.

Item Type:	Thesis (PhD)
Additional Information:	© 2024 Sam Rickman
Library of Congress subject classification:	H Social Sciences > HV Social pathology. Social and public welfare. Criminology Q Science > QA Mathematics > QA76 Computer software
Sets:	Departments > Health Policy
Supervisor:	Fernández, José-Luis and Malley, Juliette
URI:	http://etheses.lse.ac.uk/id/eprint/4836

Actions (login required)

Record administration - authorised staff only