Title: Nonparametric clustering approach for longitudinal cognitive measurements, baseline imaging and genetic data
Authors: Brian Tom - MRC Biostatistics Unit (United Kingdom) [presenting]
Anais Rouanet - MRC Biostatistics Unit (United Kingdom)
Abstract: Dementia is one of the most challenging global health problems of the 21st century, affecting over 47 million people globally with numbers expected to rise substantially over the next thirty years. At present there is a high failure rate for treatments tested for Alzheimers dementia. This may be due to treatments being tested on those who already have irreparable brain damage. Identifying persons early in disease through use of biomarkers may increase the likelihood that treatments will be more effective in slowing or arresting further progression of the disease. Clustering approaches provide a powerful means of profiling at-risk populations over time. We have developed a Bayesian Dirichlet process mixture model linking non-parametrically a longitudinal outcome and baseline variables which allows clustering structure within a heterogeneous population to be uncovered. It flexibly models the longitudinal outcome through cluster-specific Gaussian process priors and allows the number of clusters to vary through the use of a Dirichlet process prior. The methodology is applied to the ADNI cohort to identify typical profiles of subjects at high risk of dementia using longitudinal cognitive measurements, baseline socio-demographics, imaging and genetic data. Four clusters of subjects, including two with steep cognitive decline profiles, were obtained.