T he coronavirus disease (COVID-19) pandemic disproportionally affects socially disadvantaged populations because of economic, social, and demographic factors, as well as their health conditions and practices (1). Identifying vulnerable communities and effectively allocating ameliorating resources to them are necessary if policy makers are to manage the effects of COVID-19. Community vulnerability indexes (CVIs) have been increasingly used to assess community social vulnerability to a pandemic using community-level socioeconomic and demographic data (2-7). In the United States, greater CVI and vulnerability in domains of minority status, household composition, housing, transportation, and disability at the county level were signifi cantly associated with greater risk of COVID-19 diagnosis (3,4). We aimed to construct a CVI more socioculturally adapted to metropolises in Asia to explain the impact of COVID-19 across more microgeographic units (i.e., districts) within a highly urbanized city, Hong Kong, China. We also analyzed the extent that CVI was correlated with the evolution of the COVID-19 pandemic in Hong Kong.
The StudyHong Kong has long been regarded as an epicenter for many infectious diseases and is predisposed to severe COVID-19 impact because of its dense and rapidly aging population (8,9). Geographically, Hong Kong comprises 3 main regions, New Territories, Kowloon, and Hong Kong Island, which are further subdivided into 18 administrative districts (10). As of August 31, 2020, Hong Kong had experienced 3 waves of COVID-19 (Appendix Figure 1, https:// wwwnc.cdc.gov/EID/article/27/7/20-4076-App1. pdf), reporting 4,811 COVID-19 cases, including 89 deaths; 76.5% of cases occurred in wave 3 (11).Following methods used by the Surgo Foundation (6), we fi rst defi ned 5 domains that contributed to an overall CVI: socioeconomic status, household composition, housing condition, healthcare system, and epidemiologic factors. We included 22 indicators in the 5 domains for calculating domain CVI and overall CVI (Table 1). We fi rst ranked each indicator by district, with a higher rank indicating greater vulnerability. Then, we calculated the percentile rank of each district over each indicator using the formula of percentile rank = (rank -1)/(n -1), where n refers to total geographic units (n = 18). A higher percentile rank indicates greater relative CVI of the district over the specifi c indicator. We then summed the percentile ranks over all indicators within each domain, reranked them to calculate domain CVIs, and summed the percentile ranks of all domains to calculate an overall CVI for each district. We assumed equal weights for indicators within domains and for the 5 domains within the overall CVI because of a lack of available evidence informing a more optimized weight scheme. Finally, we categorized all districts into very high (>80%), high (60%-80%), moderate (40%-60%), low (20%-40%), and very low (<20%) vulnerability on the basis of their domain and overall