The study, led by Dr Jie Lian and colleagues from the Nuffield Department of Women's and Reproductive Health and the Department of Psychiatry, applied advanced artificial intelligence techniques to routine healthcare data to analyse patterns in how these conditions develop and progress.
The growing public health challenge
Neurodegenerative diseases such as Alzheimer’s and Parkinson’s represent a significant and growing societal challenge. Alzheimer’s disease is rapidly becoming one of the most costly and debilitating conditions of the twenty-first century.
Parkinson’s disease, the second most common neurodegenerative disorder, affects approximately 2–3% of adults over the age of 65.
More than 3 billion people were living with a neurological condition in 2021, resulting in an estimated 443 million healthy life years lost due to illness, disability or premature death.
As ageing is a key determinant in the development of neurodegenerative diseases and populations worldwide continue to age, the overall burden of these conditions is expected to rise. This makes advancing understanding and developing effective interventions increasingly important.
Addressing heterogeneity in neurodegenerative disease
The long-standing problem is that neurodegenerative diseases such as Alzheimer’s and Parkinson’s are highly heterogeneous, meaning that patients with the same diagnosis can experience very different symptoms, rates of progression and overall outcomes.
This variability presents major challenges for prognosis, clinical trial design and the development of targeted treatments, meaning that diagnostic categories do not always capture the underlying biological and clinical diversity within these conditions.
The research team sought to determine whether large-scale electronic health records could be used to identify more precise disease subtypes based on real-world clinical trajectories.
How the AI model worked
The researchers analysed longitudinal data from the Clinical Practice Research Datalink and UK Biobank. Using a transformer-based artificial intelligence model, they examined years of medical history recorded before diagnosis. Rather than focusing on a single symptom or test result, the model identified patterns across diagnoses, comorbidities and disease progression over time.
From these patterns, five reproducible subtypes were identified for each disease. Each subtype was characterised by distinct combinations of co-existing conditions, progression patterns and genetic risk.
Examples included:
- A vascular-associated subtype, marked by high rates of hypertension.
- A metabolic–inflammatory subtype, linked to diabetes, obesity and renal disease, and associated with more aggressive progression despite lower genetic risk.
- A genetically susceptible but clinically resilient subtype, with high inherited risk but relatively fewer additional health conditions.
Notably, several patterns, particularly vascular and metabolic profiles, were observed across both Alzheimer’s and Parkinson’s disease, suggesting shared systemic contributors to neurodegeneration.
Our study shows that Alzheimer’s and Parkinson’s disease are not single entities, but comprise distinct clinical subtypes identifiable from routine health data. Recognising this heterogeneity may open the door to earlier intervention and more tailored treatment pathways.- Dr Jie Lian
Implications for earlier and more targeted care
While further validation and prospective studies are needed, the findings highlight the potential of large-scale health data and artificial intelligence to advance precision medicine in complex neurological conditions.
Collaborators
The study was conducted by researchers from the Deep Medicine research group at the Nuffield Department of Women’s & Reproductive Health and the Department of Psychiatry, University of Oxford.
The research team included Jie Lian, Zhengxian Fan, Ben Omega Petrazzini, Wei Fan, Shishir Rao, Qianqian Yang, Guyu Zeng, Nouman Ahmed, Fatemeh Tabassi Mofrad, Malgorzata Wamil, Kazem Rahimi.
Publication
Nature Aging (2026).