Cluster Analysis on Longitudinal Cognitive Function Trajectories: A Poisson Mixture Model Approach

Cognitive Performance Scale (CPS) is a seven-category scale that indicates the cognitive impairment of a person. Motivated by other researchers, we aim to find distinct CPS change trajectories among the data provided from an unsupervised learning approach. One of the well-known issues of cluster analysis is the validation of the clustering results. Therefore, we also intend to simulate a dataset which is similar to the real one in order to validate the results. We designed a method to define the longitudinal trajectories by the transition probability matrix. The restricted Poisson mixture model was proposed in order to obtain the distinct templates and desired quantities to conduct the simulation. Applications of two different clustering methods were implemented on the simulated data. Results suggested that the longitudinal clustering method (Leffondre et al., 2004) is more appropriate.M.Sc