Harnessing Geospatial Artificial Intelligence (GeoAI) for Environmental Epidemiology: A Narrative Review

There have been two major recent shifts in geospatial data types and analytic methods. First, novel methods for statistical prediction, combining geospatial analysis with machine learning and artificial intelligence (GeoAI), allow for scalable geospatial exposure assessment within large population health databases (e.g. cohorts, administrative claims). Second, the widespread adoption of smartphones and wearables with global positioning systems and other sensors has allowed for passive data collection from people, and when combined with geographic information systems, enables exposure assessment at finer spatial scales and temporal resolution than ever before. Illustrative examples include refining models for predicting outdoor air pollution exposure, characterizing populations susceptible to water pollution, and use of deep learning to classify Street View image-derived measures of greenspace. While these tools and approaches may facilitate more rapid, higher quality objective exposure measures, they pose challenges with respect to participant privacy, representativeness of collected data, and curation of high quality validation sets for training of GeoAI algorithms. GeoAI approaches are beginning to be used for environmental exposure assessment and behavioral outcome ascertainment with higher spatial and temporal precision than before. Epidemiologists should continue to apply critical assessment of measurement accuracy and design validity when incorporating these new tools into their work.