AI-Enabled Digital Health

AI-Supported Evidence Assessment for Health Technology Evaluation

Status: Active
Funder: NIHR HTA Innovation Lab (pilot)
Collaborator: Professor Gurdeep Sagoo, Newcastle University

Health technology assessment depends on expert reviewers synthesising large volumes of clinical, cost, and submission evidence under tight timelines. We are developing and evaluating a structured AI harness that supports this work — reading trial reports and submission documents, extracting findings against a fixed protocol, and drafting assessment reports for human review.

Key finding: With all outputs traceable to source documents, and every unverifiable judgement explicitly declared rather than silently passed, the harness produced a complete draft evidence assessment for human review.

Relevant publications:

Pending — pilot in progress.

Enhancing NEWS2 for Earlier Detection of Clinical Deterioration

Status: Active
Collaborator: Dr Chris Plummer, Consultant Cardiologist and Chief Clinical Informatics Officer, Newcastle upon Tyne Hospitals NHS Foundation Trust

The National Early Warning Score (NEWS2) is the standard tool for identifying deteriorating patients in NHS hospitals, but its predictive accuracy beyond 24 hours is limited, particularly in older adults who were underrepresented in its development. Working with Dr Chris Plummer and routinely collected data from Newcastle Hospitals, we are developing machine learning models that incorporate additional demographic and observational variables to improve NEWS2's sensitivity. Our analysis of 9.8 million observations has identified significant opportunities to enhance detection in precisely the populations the original scoring system was not designed around. (Video)

Key finding: Observation-level ML analysis across 9.8 million records demonstrates a 16% improvement in sensitivity (AUROC 0.815 vs 0.777) for predicting clinical deterioration compared to standard NEWS2 scoring.

Relevant publications:

Autonomous AI Follow-Up for Cataract Surgery (Ufonia/Dora)

Status: Completed
Funder: NIHR AI in Health and Care Award
Collaborator: Ufonia Ltd.
Sites: Two NHS Trusts

Cataract surgery is the most common operation in the UK, yet most trusts still require face-to-face post-operative follow-up despite low complication rates. We co-led an evaluation of Dora, an autonomous conversational AI agent delivered by telephone, assessing whether it could safely and accurately perform post-surgical follow-up and identify patients needing further review. Dora's performance was compared against ophthalmic specialists across two NHS trusts.

Key finding: The study found good preliminary evidence of Dora's safety and accuracy for routine follow-up, with generally positive usability ratings. The results also highlighted the importance of maintaining human involvement for patients with concerns or complications. Published in The Lancet eClinicalMedicine.