University of Minnesota
Flexible Bayesian Methods for Clustering and Subgroup Analyses in Complex Medical Data
Abstract
dc:description.abstractMany modern statistical problems in public health and biomedical research involve subgroup detection, latent structure discovery and information sharing across multiple data sources. This is especially challenging when the primary dataset is too limited to support reliable inference, or if the observations have an inherent temporal, spatial, or network structure. This dissertation develops novel Bayesian methods for subgroup identification, clustering and information transfer in sequential and multi-source data settings. First, we explore how patient-level external data can be leveraged to improve enrichment decisions in pragmatic clinical trial designs, with the goal of identifying subgroups exhibiting heterogeneous treatment effects. Second, we develop a new Bayesian nonparametric clustering prior for the analysis of data arriving sequentially, and with potentially temporal, spatial, network or covariate dependence between observations. The proposed approach directly uses user-defined costs to determine how a newly arriving observation assimilates into existing clusters, while separately controlling cluster-size reinforcement. Finally, we develop a Bayesian transfer learning prior for variable selection in joint survival-longitudinal models. This method enables information transfer from source to target data sources while allowing differences between them, thus improving inference in target-data-limited settings. In summary, these methods offer flexible Bayesian machinery to enable subgroup discovery, structured clustering and adaptive information transfer in complex biomedical datasets.
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Ghosh Dastidar, Souradipto
Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Citation
- Ghosh Dastidar, Souradipto. (2026). Flexible Bayesian Methods for Clustering and Subgroup Analyses in Complex Medical Data. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/283442.