Bayesian Modeling and Inference for High-Dimensional Disease Mapping and Boundary Detection"

Project Summary/Abstract
This application seeks to advance statistical methods within the Bayesian inferential paradigm for disease map-
ping and spatial boundary analysis. Disease mapping is an epidemiological technique used to describe the
geographic variation of disease and to generate etiological hypotheses about the possible causes for apparent
differences in risk. The last decade has seen an explosion of interest in disease mapping, with recent method-
ological developments in advanced spatial statistics and increasing availability of computerized Geographic In-
formation Systems (GIS) technology. Spatial biostatisticians, data scientists and epidemiologists today routinely
encounter datasets requiring multi- or high-dimensional disease mapping in the presence of spatial-temporal
misalignment, where “dimension” refers to (a) the number of cancer types being studied, (b) the number of spa-
tial units (e.g., census-tracts, counties) in the map, and (c) the number of temporal units (time points) at which
the data are observed. This application offers novel classes of stochastic process-based graphical models with
specific attention to spatially-temporally misaligned data and modeling of multiple cancers. The versatility and
scalability of the proposed framework will allow epidemiologists and public health researchers to account for
information from multiple sources including, but not limited to, environmental factors and climate-related vari-
ables at arbitrary resolutions in spatial-temporal “BIG DATA” settings. The proposal will subsequently develop
a rigorous framework for multivariate boundary detection on maps, where boundaries delineate regions with
significantly different spatial effects.

Grant Number: 5R01GM148761-04
NIH Institute/Center: NIH
Principal Investigator: Sudipto Banerjee