Impact of spatial aggregation on detection of spatiotemporal disease clusters: analysis of SARS-CoV-2 infections in 3-D high-density settings

Introduction High-density congregate housing, including cruise ships, hotels, residence halls and correctional facilities are epidemiologically important, and key aspects of pathogen transmission have been elucidated in these environments. A range of methods have been developed to detect unusual clusters of infections in these settings; however use of explicitly 3-D (x,y,z) spatial data has received little attention. In this study, we use data collected during the COVID-19 pandemic to assess the fine-scale spatial epidemiology and the clustering of confirmed cases to better understand impacts of spatial resolution and aggregation on spatio-temporal cluster detection. Methods Data for this analysis combined the results from mandatory weekly viral testing during the 2020-2021 academic year with high-resolution spatial data from university students residing in high-rise residence halls at the University of Massachusetts, Amherst campus. These data were analyzed for statistically-significant clustering of SARS-CoV-2 cases in three-dimensional space as well as time, within and between the high-density buildings on campus. Two sets of analyses were conducted. The first used a Space-Time Permutation Model, which scans for areas with a greater than expected number of cases (SaTScan). To assess the impact of data aggregation, analysis was done at several levels of spatial resolution. Additionally, we performed sensitivity analyses using a purely temporal surveillance algorithm, CDC’s Early Aberration Reporting System-EARS. Results and conclusions Analysis with SaTScan at the room- and floor-level identified multiple statistically significant clusters within one residence hall. Analyses with these same cases aggregated at the floor-level were found to be as sensitive, but far less computationally intensive, than room-level analysis. Analysis at both of these spatial scales was more sensitive than analysis aggregated at the street address-level. Two events exceeding alert thresholds were detected in the purely temporal analysis; one of which was also detected in spatio-temporal analyses. These results expand our understanding of spatio-temporal scan metrics to include 3-D analysis, and optimizing choice of spatial scales. These results have broad applicability in epidemiology in assessing the ability of spatio-temporal methods for public health surveillance, with potential expansion to ecological studies incorporating vertical movement.