VIRTUAL Lisbon Livable Cities

Abstract

Nearly three years after onset of the COVID-19 pandemic in New York City (NYC), cumulative hospitalization and mortality rates were not uniformly distributed across the city. To help understand potential drivers of this observed geospatial disparity, we applied geographically weighted Poisson regression (GWPR), which differs from conventional regression modeling by allowing associations between these outcomes and community-level predictors to be location-dependent.
Cumulative COVID-19 hospitalization and mortality rates in n=177 NYC modified ZIP code tabulation areas as of December 31, 2022 were obtained from the NYC Department of Health and Mental Hygiene, while socioeconomic and demographic predictors were queried from the 2018 American Community Survey. We experimented with both non-multiscale models (GWPR) where one common adaptive bandwidth is applied for all predictors, and multiscale models (MGWPR) where the adaptive bandwidth is allowed to vary among predictors. Both GWPR and MGWPR models yielded better diagnostics than the more conventional global models that assume spatially stationary associations, with the non-multiscale GWPR model performing the best. Several predictors acted as both risk and protective factors for both outcomes, depending on location, including the percentage of non-Hispanic whites, foreign born citizens, male and having had at least one vaccination, along with mean commute time. Other predictors showed more geographically consistent effects. For mortality, the percentage of residents without health insurance acted solely as a risk factor. Similarly, for hospitalizations, the percentage of residents with a disability acted solely as a risk factor. The percentage of residents >24y with a bachelor’s degree or higher acted solely as a protective factor against both outcomes. These results highlight potential areas for city-wide policies to reduce disparities and the overall burden of future epidemics.