B1465
Title: Regression analysis using compositional balances: Case study for chronic kidney disease and environmental toxins
Authors: Jennifer McKinley - Queen's University Belfast (United Kingdom) [presenting]
Ute Mueller - Edith Cowan University (Australia)
Pete Atkinson - Lancaster University (United Kingdom)
Damian Fogarty - Belfast Health Trust (United Kingdom)
Abstract: Investigating the importance of multi-element interactions of environmental toxins in understanding the occurrence of clusters of chronic diseases, such as chronic kidney disease, is of global concern. Environmental toxins (air, soil and waterborne) often comprise soil or water geochemical data, which are compositional in nature in that they convey relative information that should be extracted by treating log-ratio or equivalently transformed data. For regression analysis involving compositional, the concept of balances between two groups of parts of a composition provides an interpretable approach to identify components whose relative abundances may be associated with a response variable. Several approaches to select balances are available, which constitute data- or knowledge-driven approaches within a compositionally-compliant context. However, many questions remain on how to interpret compositional balances including the impact of the ordering of elements in a compositional balance within a data-driven or knowledge informed knowledge-driven approach. Moreover, regression models assume independence between the observations, an assumption that may not be valid for spatial data. The aim is to explore different compositional balance approaches and the impact of spatial dependence, using a case study on chronic kidney disease and its relation with air, soil and waterborne environmental toxins.
