A1020
Title: Bayesian dynamic clustering factor models: Estimating subgroups and transitions
Authors: Tsering Dolkar - Virginia Tech (United States)
Marco Ferreira - Virginia Tech (United States)
Hwasoo Shin - Virginia Tech (United States)
Allison Tegge - Virginia Tech (United States) [presenting]
Abstract: With the increased recognition of heterogeneity in clinical cohorts, there is an increased need to develop algorithms to identify subgroups. Motivated by this need, and with an application to longitudinal health data as a case study, novel Bayesian dynamic clustering factor models are proposed. It is assumed that participants are assigned to one of several clusters at each time point. Each cluster corresponds to a health state. In addition, the transitions are modeled among the different clusters using a hidden Markov model. A Markov chain Monte Carlo algorithm is developed to explore the posterior distribution of the cluster means and the cluster transition probabilities. Finally, model selection is developed to concomitantly choose the number of clusters and the number of factors.
