B1235
Title: Individual-level prediction of Alzheimer's disease progression: Tackling the TADPOLE challenge
Authors: Steven Hill - MRC Biostatistics Unit, University of Cambridge (United Kingdom) [presenting]
James Howlett - MRC Biostatistics Unit - University of Cambridge (United Kingdom)
Steven Kiddle - University of Cambridge (United Kingdom)
Sach Mukherjee - German Center for Neurodegenerative Diseases (Germany)
Anais Rouanet - MRC Biostatistics Unit (United Kingdom)
Bernd Taschler - German Center for Neurodegenerative Diseases (Germany)
Brian Tom - MRC Biostatistics Unit (United Kingdom)
Simon White - University of Cambridge (United Kingdom)
Abstract: Alzheimer's disease is an increasing burden on public healthcare systems. It is thought that treatments are most likely to be effective in the very early stages of the disease process. However, identifying individuals in these early stages is challenging. The Alzheimer's Disease Prediction Of Longitudinal Evolution (TADPOLE) Challenge aimed to investigate the ability of methods to predict the future progression of individuals at risk of Alzheimer's disease. Training and test data for this open community challenge was from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and the task was to predict three outcomes: clinical diagnosis, a cognitive test score and a brain imaging biomarker. We used several approaches to tackle this problem, including multi-state models, latent-class mixed models and high-dimensional regression. Overall, our methods performed well and were top-ranking in the cross-sectional prediction subcategory. An overview of the challenge, our approaches and their performance will be provided, and some insights from our participation in the challenge will be drawn.
