Unsupervised Modeling of Patient-Level Disease Dynamics

To provide insight into patient-level disease dynamics from data collected at irregular time intervals, this work extends applications of semi-parametric clustering for temporal mining. In the semi-parametric clustering framework, Markovian models provide useful parametric assumptions for modeling temporal dynamics, and a non-parametric method is used to cluster the temporal abstractions instead of operating on the original data. Our contribution extends abstraction to continuous-time Markov models and the clustering component to the non-parametric Bayesian setting, which does not require the number of clusters to be indicated a priori.