GPS-based personalized estimation of longitudinal environmental exposures and its impact on health

Abstract

This thesis aims to develop a GPS-based methodology for estimating personalized, longitudinal environmental exposures and to investigate their potential associations with health outcomes, particularly changes in microbiota composition. The primary objectives include constructing accurate spatiotemporal exposure maps by integrating national environmental databases with GPS data from personal wearable devices, and evaluating whether variations in exposure to environmental factors particularly urban/rural exposure settings are associated with alterations in microbiota at the phylum level.

To achieve this, the thesis introduces a novel methodology that performs environmental exposure estimation across five distinct layers: urban/rural classification, forest environment detection, proximity to water systems, radiation levels, and PM2.5 air pollution. A hybrid classification approach combining population distribution data and land cover datasets is employed to define urban boundaries with high spatial resolution. Forest and waterbody exposures are determined using vector-based spatial containment methods, while dynamic environmental layers radiation and air quality, are linked to GPS trajectories through time-aware spatial indexing and nearest-neighbor matching. Data from 24 participants monitored over a two-month period were analyzed, with GPS recorded every 5 seconds and microbiota samples collected biweekly.

The study evaluates the temporal accumulation of exposure within each environmental layer, revealing both overall trends and subject-specific variations. Analyses, including correlation and linear regression, were conducted to examine the potential associations between urban/rural exposure and microbiota composition. Results suggest that regional environmental factors, particularly urban versus rural environments, may influence microbial diversity. However, due to data noise and the dominance of certain phyla, the correlation outcomes are sensitive and warrant more detailed investigation.

The proposed methodology offers a flexible, scalable, and computationally efficient framework for high-resolution environmental exposure estimation and these findings underscore the significance of personalized exposure modeling.