Developing and validating ‘Sia’: a responsible AI platform for dementia prediction

Challenge: Nearly 1 million people in the UK live with dementia; it devastates families and costs £42 billion annually. Dementia causes damage to the brain 10+ years before symptoms appear, and early signs are often missed or misdiagnosed. Earlier detection would improve quality of life and support better care for People with Dementia (PwD). Yet only 65% of NHS patients with early dementia are diagnosed (Darzi (1)) and current diagnostic tools lack sensitivity. Solution: Sia is an end-to-end software prototype developed by Prodromic for early stratification and staging of dementia. The logic of the innovation is simple: while individual markers (e.g., medical history, blood test, brain scan, memory test) provide limited insight into a patient’s disease, there are useful signals hidden in the complexity of health data that—when combined—predict a person’s health trajectory. Sia implements a suite of multimodal machine learning tools to automatically (a) extract and (b) fuse diverse data types to (c) stratify PwD based on their (d) future disease trajectory. It separates “clinically indistinguishable” patients early and delivers a prognostic score accompanied by a simple ‘traffic light’ label (stable, slow-, rapid-progressive). Sia is validated across 5 peer-reviewed studies: its predictions are robust and generalise across contexts (research cohorts, memory clinics, clinical trials) and geographies (US, Europe, Asia). Competitive advantage: Sia is co-designed with stakeholders (including PwD, carers) and works with routinely-collected NHS data (leveraging harmonisation, imputation, and bias monitoring tools) to facilitate widespread adoption. It is three times more sensitive than standard of care, and its predictions are validated against longitudinal data over 5 years. Sia is interpretable-by-design: it is not ‘black box’ (clinicians can understand how it makes predictions) and it delivers robust, stable outputs (not generative AI), to build trust. Workplan: This project develops Sia from a validated prototype to a scalable product that is feasibility-tested and technically ready for the NHS. We will test Sia’s predictions in large-scale, diverse healthcare data to validate its readiness for real-world adoption. This includes data from different clinical settings (primary, secondary care), including patients from diverse backgrounds (ethnicity, education, deprivation). We will register Sia as a medical device (Class 1) and develop its NHS adoption strategy. We will conduct an early-stage feasibility and clinical utility study to refine Sia’s usability, clinical pathway fit and capture health economic benefits over standard care. We have embedded multi-stakeholder perspectives throughout, with strong PPIE to ensure that Sia meets patient and carer needs and priorities. Impact: By reducing misdiagnosis and improving detection and prognosis, Sia has potential for significant impact on PwD, families and health systems. Sia provides more certainty about what will happen to improve care planning. It can reduce the healthcare burden: avoiding unnecessary invasive testing, and focusing NHS resources (clinical time, new disease modifying treatments) for patient benefit. Sia provides a validated approach to standardise dementia diagnosis using non-invasive, low-cost tools, reducing inequalities. This project advances Sia’s impact, building a responsible innovation solution for the NHS to ultimately benefit the lives of people with dementia.