Abstract: Global health experts warn we are currently facing two pandemics--the coronavirus disease 2019 (COVID-19) and an accompanying infodemic of misinformation. Belief in misinformation is associated with lower compliance with pandemic mitigation recommendations, and has already had fatal consequences. For this project, we focused on misinformation linking 5G to COVID-19. While claims of 5G health hazards have circulated among conspiracy theorists for years, they have seen a spike in popularity amid the COVID-19 pandemic. To understand the dynamics of this misinformation propagation, we randomly sampled 10% of tweets made between January 1 and April 15, 2020 containing the 1-gram “5G” (case insensitive) for a total of 629,657 tweets. Using two complementary natural language processing approaches, we found evidence to suggest not only an increase in 5G-related content, but also the evolution of 5G narratives over the course of the pandemic. Broadly, we note a transition from “5G” occurring primarily in technology-related tweets to being increasingly associated with COVID-19. Specifically, training word embeddings on this corpus reveals a shift in “5G” being most closely associated with terms like “innovation” and “data” in February, to being associated with “corona” and “conspiracy” in April. Temporal networks of 1-grams similarly demonstrate how the most central 1-grams in 5G-related tweets shifted from “2020,” “huawei,” and “#ai,” to “people,” “conspiracy,” and “coronavirus.” Using community detection on these semantic networks, we discover an infrastructure of 5G narrative categories, including conspiracy-related tweets focusing on US politics, as well as narratives echoing older conspiracy theories (e.g., chemtrails). Together, our findings suggest that misinformation does not simply appear out of nowhere, but evolves as part of an ongoing, larger narrative. Understanding the dynamics of distinct narratives within the COVID-19 infodemic can inform targeted interventions to inoculate against misinformation more precisely in an effort to support compliance with health guidance.
Abstract: Striking racial disparities have emerged as COVID-19 infection rates climb throughout the United States. For example, in the metropolitan area of Chicago, Black people represent nearly half of COVID-19 cases (and 70% of deaths) despite comprising 31% of the population. We have observed a troubling tendency for COVID-19 racial disparities to be dismissed in national discussions as simply due to higher rates of pre-existing conditions or "unhealthy lifestyles" rather than structural factors (e.g., access to healthcare). We aim to provide an evidence-based counter to this harmful and essentializing narrative through two studies.
Study 1 uses Natural Language Processing (NLP) on COVID-19 news articles to explore the extent to which news outlets discuss racial disparities in COVID-19 outcomes in terms of endogenous differences (vs. social/structural factors). Study 2 uses Agent-Based Modeling (ABM) to test whether social forces alone can produce the racial disparities in COVID-19 cases that we observe in Chicago, even when equating agents' baseline health. We constructed a three-layer ABM to simulate the effects of racial segregation and socioeconomic status on COVID-19 cases. This ABM uses demographic and segregation data from Chicago to create a racially-segregated network (Layer 1), which we overlay with an SIR model (Layer 2) and an economic game (Layer 3). We compare three models to a null model (Model 0; no segregation; equal punishments): Model 1 (racial segregation only), Model 2 (punishments weighted by racial groups' mean SES), and Model 3 (racial segregation + weighted punishments). We find that Model 3 best fits the Chicago data, suggesting that segregation and income inequality by race are sufficient to produce the Chicago racial disparities in COVID-19 cases even when equating agents’ baseline health. Given these findings, we caution against interpreting racial disparities in COVID-19 cases and outcomes through the lens of endogenous vulnerability.
Abstract:
Background: Non-pharmacological interventions (NPIs) such as physical distancing and mobility restrictions were implemented worldwide to mitigate the viral infection and disease (SARS-CoV2/COVID-19) spreading. In Mexico City, physical distancing was driven by the closure of non-essential economic entities including schools and universities. However, a total lockdown was not implemented. The Mexican government promoted physical distance and stay-at-home (cooperators) measures, particularly for mild symptoms (auto-isolation). Several reasons including economic needs prevent the total compliance of these requests (non-cooperators).
Aim of the study: To develop and implement an agent-based SIR model on a network to estimate the effect of cooperators on the spread of SARS-CoV2 to prevent the saturation of local health systems.
Methods: We simulated local NPI dynamics by extracting and processing mobility and contagion data from public sources. To estimate the fraction of cooperators and non-cooperators (>25% and <25% mobility reduction between prior- and post-COVID-19 mobility, respectively) for NPIs, we compared prior- (Origin-Destination mobility survey, 2017) and post-COVID-19 mobility data (Google community mobility reports and mobile data from the UNDP/Grandata lab, 2020) from 10 regions of Mexico City. Agents are heterogeneous with different demographic characteristics including recovery time, number of meetings, and total connections (degree) in the interaction network. Recovery time and number of meetings were selected from Gaussian and exponential distributions, respectively; their degree by the interaction network. We simulated 25 fixed networks with random starting conditions for different cooperation probability values (10 - 100%).
Results: The infected fraction of the population, the number of simultaneous infections, and the duration of the epidemic decreased as the mean cooperation probability in the population increased, reaching a lower and earlier epidemic peak when a total lockdown was simulated (100% cooperators). On one hand, the infected fraction of the population decreases linearly at a rate of 0.4 (10% reduction per every 25% increase in the cooperant fraction) between 0 and 65% cooperators, after which it barely decreases. On the other hand, the number of simultaneous infections decreased in a non-linear manner with 21% of cooperators reducing the total number of simultaneous infections by half.
Conclusions: NPIs decreased the spread of a viral epidemic among the population to avoid saturation of healthcare systems. Even though some features of the model are kept fixed, we developed a scalable and modifiable framework that can be tuned to represent the mobility and demographic dynamics of populations coping with epidemics —real-world data.
Our panel of world-renowned judges brought extensive expertise from diverse fields related to Network Epidemiology:
Laurent Hebert-Dufresne, University of Vermont
YY Ahn, Indiana University
Sam Scarpino, Northeastern University
Ginestra Bianconi, Queen Mary University of London
Mirta Galesic, Santa Fe Institute
Ben Althouse, Institute for Disease Modeling
The awards are sponsored by UMD's COMBINE program in Network Biology. The first place team will receive $2000 USD; the second place team will receive $1000 USD. Award amounts will be distributed equally among group members. We will be in touch with the prize winners to process awards.
Fine print: Awards will be made in accordance with US law and University of Maryland policy. When possible, awards will be processed as honorarium payments. For foreigners who are in the US on visas that do not allow them to receive such an honorarium, the award will be made as a travel/conference award, whereby we will reimburse expenses related to scientific travel or conference attendance, up to the award amount. Travel/conference awards require pre-approval and must be used within 12 months of the award date.