The Leeds Institute for Data Analytics is pleased to present the following seminar in our series showcasing data analytics.
The seminar will be held in Seminar Room 9.60, Level 9, Worsley building.
In this talk we will introduce some of the key issues and challenges in modelling disease progression when utilising many different sources of data. We will discuss the different ways that data can be collected, how this data can be used to build realistic models of disease, and some of the challenges posed by data quality and heterogeneity, such as the need to electronically phenotype. Resulting models will assist in the prediction of progression so that interventions can be made early enough to slow the advance of degenerative diseases such as glaucoma, type 2 diabetes and dementia.
We will discuss a range of methods and approaches, from some more traditional epidemiological and descriptive methods to capture specific markers of disease severity, to pattern mining of healthcare events in electronic health records. We will also explore some novel approaches developed at Brunel to integrate different types of data into models that can simultaneously identify key stages in progression and subcohorts of patients, exhibiting variations in how their symptoms progress. This is achieved through a combination of probabilistic approaches with latent variables.
Finally, we will discuss some of the issues with the over-use of latent variables and the general risks of relying on black-box approaches to modelling data with respect to the ethics of AI models in healthcare and in particular how we ensure healthcare workers and the wider public retain trust in them.
About the speakers
Allan Tucker is the head of the Intelligent Data Analysis group at Brunel University. After his first degree in cognitive science, Allan became interested in using AI models of time-series data to try and understand the underlying processes that generate brain function, leading to his PhD on “The Automatic Explanation of Multivariate Time Series”. His current research interests include applications of machine learning for modelling high dimensional gene expression data and exploring dynamics of fish population in the Northern Atlantic. His expertise has been sought on multiple committees (NHS, Wellcome, REFORM) on the implications of AI in health and medical research.
Maxine Mackintosh is a PhD student in data science at University College London. Her research focuses on mining medical records to identify new predictors of dementia. In addition, she is the co-founder of One HealthTech – a community which champions and supports underrepresented groups in health innovation, particularly women, to be future leaders in healthcare. Her professional work has led her to the Royal Society, Roche, L’Oreal, Department for International Development, and NHS England. She is part of a number of communities and committees including the World Economic Forum’s Global Shapers, the British Computer Society and the Digital Health Forum.
15.30-17.00: Progression beyond regression; using heterogeneous data in disease trajectories
17:00-18:00: Networking reception with drinks and nibbles hosted in the LIDA staff room
To book please email Hayley Irving with your name, occupation and faculty/organisation.