Please note: this event has passed
Abstract: In 2016 there were 47 million estimated dementia sufferers worldwide, with a forecasted increase to over 131 million by 2050. The cost of dementia was estimated to 818 billion USD in 2016 and is expected to increase to 2 trillion USD by 2030. For comparison, dementia has currently a health and social care cost higher than cancer, stroke and chronic heart disease, taken together. Diagnosing dementia is problematic because there is currently no standardized dementia test, and on the other hand it involves a complex process which is both expensive and invasive, based, among others, on using MRI methods and on the detection of well-established biomarkers such as amyloid-beta, p-tau and t-tau in the cerebrospinal fluid (CSF). Moreover, the diagnosing procedure is a highly specific task based on the different sub-types of dementia – the most common being Alzheimer’s disease which accounts for 60% to 80% of dementia cases. Mild cognitive impairment which implies a significantly higher risk of developing dementia, can either become cognitively stable or return to a healthy cognitive state. Overall, such complexities (partly) contributed to a large proportion of people with dementia to go undiagnosed. Moreover, the prevalence of undetected dementia is high globally, according to a recent research based on reviewing 23 selected studies, which concluded that the pooled rate of undetected dementia was around 60% of dementia cases. However, current thinking suggests that about a third of dementia cases could be prevented given early at-risk identification and proactive interventions. In this context, there is an increasing body of research applying powerful methods based also on machine learning and statistical learning for analysing, from usual to more complex clinical data, and developing accurate prediction models aimed to contribute to improving diagnosis rates, to detecting risk of dementia several years in advance, as well as to dementia biomarker discovery.
Current and recent research conducted by Dr Daniel Stamate and his Data Science group in Computing Department at Goldsmiths College (University of London) within different external collaborations, regards various approaches to predicting dementia with state-of-the art machine learning and statistical learning methods. One such direction, in collaboration with the University of Manchester and King’s College London, concerns the application of deep learning to predicting dementia and mild cognitive impairment on the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort, as well as developing machine learning classification approaches for addressing the problem of deterioration in Alzheimer’s type of dementia whose results were presented in a recent seminar in this series. Another direction of research was developed in a collaboration with Copenhagen, KCL, Oxford, and other centres in the EMIF-AD Consortium, and regards a machine learning study based on state-of-the-art algorithms (including XGBoost) for diagnosing Alzheimer-type dementia in blood – a study analysing data from the European Medical Information Framework for Alzheimer disease (EMIF-AD) biomarker discovery cohort, which demonstrated that plasma metabolites have the potential to match the (AUC) performance of well-established AD CSF biomarkers such as amyloid-beta, p-tau and t-tau. Another research direction, developed by Dr Stamate’s group in collaboration with the University of Manchester, regards a new study of predicting risk of dementia with machine learning and statistical methods, using routine primary care records from the Clinical Practice Research Datalink (CPRD).
This talk will present some results from another, ongoing study that Dr Stamate’s group does in collaboration with Prof Daniel Stahl (KCL) and Dr Olesya Ajnakina (UCL, KCL) on Predicting Risk of Dementia with Survival Machine Learning and Statistical Methods: Results on the English Longitudinal Study of Ageing Cohort (ELSA). Three recent survival method extensions based on the machine learning algorithms of XGBoost, Random Forest and Elastic Net were applied to train, optimise, and validate predictive models based on the ELSA cohort. The three survival machine learning models are compared with the largely applied conventional statistical Cox proportional hazard model, proving their superior predictive capability and stability on the ELSA data, as demonstrated by computationally intensive procedures such as nested cross-validation and Monte Carlo validation (subsuming also reiterated model optimisation and evaluation). A currently ongoing exploration is looking into the extension of these results by applying new survival neural networks methods, as well as by developing interpretable AI predictive models, which will make the object of other presentations.
Bio: Dr Daniel Stamate is a machine learning scientist and statistician, and a Senior Lecturer (Associate Professor) in Data Science at University of London – Goldsmiths College’s Computing Department, where he and his group in the Data Science & Soft Computing Lab develop research based on state-of-the-art machine learning and statistical learning methods, as well as their applications in health, in particular in predicting dementia. Dr Stamate is affiliated also with the University of Manchester, where he is an Honorary Senior Lecturer, developing research in predicting risk of dementia with machine learning and statistical methods using routine primary care records, in the Division of Population Health, Health Services Research & Primary Care, School of Health Sciences, where he collaborates with Prof David Reeves’ team. Dr Stamate’s research lab in London has developed several projects in collaborations with different research groups in UK and abroad, including a long collaboration with Prof Daniel Stahl’s group of Precision Medicine and Statistical Learning in the Department of Biostatitics and Health Informatics at King’s College London, and also with different research teams in the medical schools of UCL, Imperial College, Oxford, Maastricht, Yale, and other research centres. Dr Stamate has various collaborations also with industry, including with Sherwin-Williams, Santander, Hitachi, and is a regular contributor at FIMA Europe. At Goldsmiths he founded and led one of the first Data Science MSc programmes in UK (started in 2014), which was recently replicated and offered as an online programme at the University of London. Daniel has a background in Computer Science and Mathematics, having specialised in Computational Statistics in his BSc & MSc from the University of Iasi, and got his PhD in Computer Science from the University of Paris-Sud – currently Paris-Saclay University.
Links:
[1] Survival ML Dementia ELSA paper: https://doi.org/10.1007/978-3-031-08341-9_35
Daniel Stamate, Henry Musto, Olesya Ajnakina, Daniel Stahl: Predicting Risk of Dementia with Survival Machine Learning and Statistical Methods: Results on the English Longitudinal Study of Ageing Cohort. 18th IFIP International Conference on Artificial Intelligence Applications and Innovations - AIAI 2022, IFIP Advances in Information and Communication Technology, vol 652, Springer, 2022
[2] Daniel Stamate’s research homepage: https://personalpages.manchester.ac.uk/staff/daniel.stamate/