One of the most concerning aspects of the ongoing COVID-19 pandemics is that it disproportionately affects people from Black and African American backgrounds, creating a so-called "COVID-19 infection gap", i.e., a marked difference between the percentage of African Americans citizens in a community and the percentage of African Americans infected by (or died of) COVID-19. The abnormal impact of COVID-19 on these ethnic groups seem to be almost uncorrelated with other risk factors, including co-morbidity, poverty, level of education, access to healthcare and response to cures. Here we introduce the concept of dynamic segregation, that is the extent to which a given group of people is internally clustered or exposed to other groups, as a result of mobility and commuting habits [1]. By analysing census and mobility data on major US cities, we found that the weekly excess COVID-19 incidence and
mortality in African American communities is significantly associated with their dynamic segregation. The results confirm that knowing where people commute to, rather than where they live, is much more relevant for disease containment.
We quantify the dynamic segregation of a certain group in a urban area by means of the typical time needed by individuals of that group to get in touch with individuals of other groups when they move around the city. In our model, a city is represented by a graph G where nodes are census tracts and each edge indicates a relation between two areas, namely either physical adjacency or the existence of commuting flows between them. Each node is assigned to a class, according to the ethnicity distribution in the corresponding area. Then, we consider a random walk on the graph G, and we look at the statistics of Class Mean First Passage Times (CMFPT) and Class Coverage Times (CCT). The former is the expected number of steps needed to a walker starting on a node of a certain class $\alpha$ to end up for the first time on a node of class $\beta$, while the latter is related to the time needed to a random walk to
visit a certain fraction of all the classes present in the system. The underlying idea is that a random walk through the graph preserves most of the information about correlations and heterogeneity of node classes [2]. Consequently, if a system is dynamically segregated, the statistics of CMFPT and CCT will be substantially different from those observed on a null-model graph having exactly the same set of nodes and edges, but where a node is assigned a class at
random from the underlying ethnicity distribution. Starting from the statistics of CMFPT and CCT at the level of each city, we defined three indices of dynamic segregation, namely dynamic clustering (C), dynamic exposure (E), and dynamic isolation (I), and we associated to each state in the US the weighted average of each of those indices across the largest metropolitan areas of the state.
In Fig. 1 A-C we show the correlation between the infection gap and our segregation measures as the pandemic progresses. Indeed, there exists a quite strong correlation between dynamic segregation and the disproportionate number of infected in African American communities. In particular, the dynamic clustering of African Americans in a state correlates positively and quite strongly with the infection gap observed in that state in the first weeks of the data set, both on the adjacency and commuting networks. The dynamic segregation also provides significantly higher correlations than other state-of-the-art indicators. Interestingly, the combination of dynamic segregation and use of public transport seems to explain the persistence of infection gap throughout the early phases of the pandemic (Fig. 1 D-F).
Our results suggest that before lock-downs are put in place, the commuting patterns of African Americans increase their exposure to the virus. After lock-downs are enforced, instead, they are more likely to pass the virus over to other African Americans, as a result of the high levels of clustering and isolation of these communities, as measured in the adjacency graphs of census tracts. In general, the states where African Americans are more exposed due to long-distance trips are also those where they are more clustered due to short-range mobility. The existence of such positive correlations provides an interesting hint for policy makers: by mitigating the clustering of dynamically segregated ethnicities one would be able to also reduce their disproportionate exposure to the spread of diseases.
[1] A. Bassolas, S. Sousa, V. Nicosia, Disproportionate incidence of COVID-19 in African Americans correlates with dynamic segrega- tion. arXiv preprint arXiv:2007.04130 (2020).
[2] V. Nicosia, M. D. Domenico, V. Latora, Characteristic exponents of complex networks. EPL (Europhysics Letters) 106, 58005, (2014).