This project uses routinely collected paediatric observation data to identify early indicators of clinical deterioration in hospitalised children. Using data from Great Ormond Street Hospital, it will analyse physiological measurements and clinical signs, such as heart rate, oxygen saturation, and respiratory effort, to determine which observations most accurately predict adverse events, including unplanned intensive care admissions and paediatric arrest calls.

The study will apply advanced statistical and machine learning methods to evaluate the predictive value of individual and combined observations and to better understand patterns associated with deterioration. A key aim is to reassess and improve existing Paediatric Early Warning Scores (PEWS), which currently rely on simplified, manually calculated thresholds and have recognised limitations.

By developing and comparing data-driven prediction models, the project seeks to generate evidence for more accurate and interpretable paediatric early warning systems, while also improving understanding of the clinical significance of commonly recorded observations.

This project explores whether large language models (LLMs) can be used to identify and classify acute respiratory tract infections (ARIs) in adults presenting to the emergency department, and whether their performance is comparable to traditional expert case review. ARIs are the most common infectious presentations in UK emergency departments, but determining their underlying cause remains challenging due to the lack of a robust diagnostic gold standard.

Using both structured and unstructured electronic health record data from UCLH, the study will assess the feasibility of LLM-based approaches for diagnosing ARIs and assigning likely aetiological categories. It will then compare LLM-generated classifications with expert clinical review to evaluate agreement and determine whether LLM-based phenotyping could be used reliably in future research.

The overall aim is to develop more efficient, scalable, and consistent methods for phenotyping ARIs, supporting future studies investigating biomarkers and improving diagnostic accuracy.

This project investigates the challenges care home staff face when using hospital discharge summaries and aims to develop co-produced solutions to improve communication and patient care. Discharge summaries are a key mechanism for transferring information from hospitals to care homes, but they are often written using complex medical terminology, making them difficult for care home staff to understand. This can affect patient safety and care quality, increase workload within care homes, and lead to additional demands on NHS services through increased primary care contacts and avoidable readmissions.

The study comprises four workstreams that will examine current experiences of discharge summaries, co-produce guidance for writing more accessible summaries, and develop a large language model (LLM) capable of generating lay summaries. As part of the initial work, a retrospective analysis of discharge summaries from three NHS trusts will assess factors such as readability, comprehensibility, clarity of required actions, and the literacy, numeracy, and digital skills needed to use the information effectively.

This study aims to improve decision-making around critical care admission for patients with haematological cancers, such as leukaemia and lymphoma, who are at high risk of serious complications during treatment. While therapies can be curative, they often result in severe illness requiring intensive monitoring and support. Compared with the general hospital population, these patients have a substantially higher likelihood of needing critical care, yet not all patients benefit from intensive care interventions.

The research will investigate whether earlier identification of deterioration and earlier admission to critical care could improve outcomes. Current clinical tools based on medical history, blood tests, and physiological observations have limited ability to predict which patients are most likely to require intensive care and when intervention would be most beneficial.

Using anonymised data from haematology cancer patients treated at UCLH between 2019 and 2026, the study will examine associations between early indicators of deterioration, critical care admission, and patient outcomes. Leveraging UCLH's large population of critically ill haematology patients, the project aims to develop evidence that supports more informed and timely ICU referral decisions, ultimately improving care pathways and outcomes for patients with haematological malignancies.