AI system aims to uncover Alzheimer’s patients earlier and tackle diagnostic delays

DementAI prototype developed by SAS partner Katalyze Data aims to surface hidden Alzheimer’s cases earlier and reduce pressure on overstretched clinical teams

Thousands of people living with Alzheimer’s in the UK remain undiagnosed, often until the disease has significantly progressed. A new British-built AI system, DementAI, is designed to change that.

Created by UK data and AI consultancy Katalyze Data as part of the SAS Hackathon 2025, DementAI supports earlier identification of Alzheimer’s disease, the most common form of dementia, by surfacing risk signals already present within existing clinical records.

The team estimates that the tool could help identify the condition up to around two years earlier than existing routes, based on retrospective analysis of patient records, while also reducing time and resource pressures across parts of the diagnostic pathway. By bringing forward the point of identification, the approach could help ensure patients are assessed while care and intervention options remain more effective.

Dementia is the UK’s leading cause of death, and a significant proportion of people estimated to be living with the condition still have no formal diagnosis. The result is a hidden backlog, where warning signs exist in plain sight for sufferers whose cases may not yet have triggered timely specialist assessment.

DementAI has been built as a working, end-to-end prototype that connects stages clinicians typically handle separately – from analysing medical record data to deploying models within decision pathways – with an emphasis on usability and clinician-in-the-loop design. Rather than introducing new screening burdens, the system works with data health providers already hold, helping to turn fragmented information into actionable insight.

The prototype draws on medical records, brain activity and unstructured clinical information, using synthetic data to support development and testing where appropriate. By combining structured and unstructured signals, the tool detects subtle patterns of decline that may otherwise be difficult to piece together during short consultations.

Recognised as the Health Care & Life Sciences category winner at the SAS Hackathon 2025, DementAI goes beyond point prediction to embed analytics within a governed decision workflow. Instead of generating isolated risk scores, the system provides clinicians with a transparent, auditable rationale for why a patient has been flagged.

By identifying patients earlier, the team believes DementAI could improve the experience for families while helping clinical teams manage growing demand with finite resources. The project is now seeking engagement with healthcare organisations to further evaluate the model’s impact in operational settings.

“We are in a race against time when it comes to dementia,” said Tamás Bosznay, Principal Consultant at Katalyze Data. “Early identification can make a meaningful difference to how patients and families experience the condition. But without better ways of finding people sooner, those opportunities can be lost.

“We didn’t build DementAI just to make predictions; we built it to buy patients time. By surfacing the signals already hiding in plain sight within clinical records, the system is designed to help ensure that when care teams are ready to act, the right patients are identified earlier and more consistently.”

Alongside performance, DementAI was designed with governance embedded across the lifecycle of data and models, including explainability, audit trails, ongoing monitoring and bias reporting.

“Synthetic data, agentic AI concepts and governance are not ‘nice-to-haves’ in sensitive settings like healthcare – they are what make innovation usable at scale,” said Dr Iain Brown, Global Head of AI & Data Science at SAS. “DementAI shows how artificial intelligence can be applied in a way that is both ambitious and responsible.”

The tool blends classic machine learning with large language model (LLM) elements and synthetic data, using technologies including SAS Viya, SAS Data Maker and SAS Viya Workbench alongside open source tools such as Python and R.

The team is seeking engagement with NHS Trusts to explore pilot deployments that could validate these outcomes and support efforts to reduce diagnostic delays.