Spatiotemporal impacts of human activities and socio-demographics during the COVID-19 outbreak in the US

As of June 2022, the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the aetiological agent of coronavirus disease 2019 (COVID-19), has infected 84 million people and caused more than 1 million deaths [1] in the US. The main routes of SARS-CoV-2 entry and transmission are “contact”, “droplet”, and “airborne” [2]. In light of the severe consequences from the COVID-19 outbreak, different public authorities quickly responded to the outbreak through various strategies, including the declaration of emergency, travel restrictions, city lock-down, and enforcing social distancing [3, 4]. If properly followed and executed, these measures serve as the crucial first steps to limit physical contact and mitigate the extent of the outbreak before a vaccine is available. Nevertheless, under similar mitigation measures, significant differences are observed in the number of reported infections and the mortality rate across the US [5]. This motivates us to explore the underlying factors that result in the heterogeneous disease dynamics for assisting the disease mitigation policies in the remaining phase of the COVID-19 and better preparing against future risks of unknown infectious diseases.

As mentioned in the WHO study for the 2009 H1N1 pandemic [6], in addition to the pathological variables, the extent of the disease outbreak may be attributed to various non-epidemiological factors, including mobility level, social-demographics, pre-existing conditions of the population [7], quality of health services, travel patterns, social network [8‐11], ecological factors [6, 12], etc. But our knowledge of the precise impacts of these factors is very limited, primarily due to the lack of data that may enable the nexus between disease dynamics and the possible contributing factors. With recent advances in ubiquitous computing and epidemiology and the wide adoption of smartphones in the past decade, we are now able to monitor human activities at a fine spatiotemporal level and overlay such dynamics with high-resolution trajectories of disease outbreaks. This, together with the available data on socioeconomic, demographics, and historical daily commuting patterns, provides an unprecedented opportunity to scrutinize the impacts of non-epidemiological factors and comprehensively evaluate how these factors drive the fate of the disease outbreak across the US.

Existing studies have related social-demographic characteristics and human activity with the spread of the COVID-19. The social-demographic structure of the population is demonstrated to have a significant effect on the fatality rate. An early study in China [13] suggested that people with an age greater than 80 years older have the highest fatality rate of 14.8%, and similar findings were obtained from studies in other countries [14, 15]. In addition, studies [16, 17] revealed the existence of racial disparities among the Whites, the blacks, the Asians, and the Hispanics in the COVID-19 outbreak. In particular, nearly 20% of the US counties had a disproportionate black population [18], and they accounted for 52% of the confirmed cases and 58% of the deaths nationally. Except for demographic factors, the social and economic factors are also found to affect the fate of the COVID-19 outbreak. The study [19] suggested that households with the lowest income level are six times less likely to be able to work from home and three times less likely to be able to self-isolate in the UK during the COVID-19. Besides, Stojkoski et al. [20] mentioned that the high-income population is more resilient to being infected by the COVID-19. Finally, extensive efforts have shown that human activities and mobility dynamics are dominating factors that facilitate the spread of infectious diseases [21‐24]. Studies suggested that information propagation and commercial activity patterns co-affect the epidemic propagation [25‐27]. Nevertheless, Lima et al. [28] provided evidence that restricting mobility may not eliminate the diseases. And Bajardi et al. [29] recommended that stricter regimes of travel reduction would have led to a delayed outbreak of two weeks based on the study of the 2009 H1N1 pandemic.

The aforementioned studies highlighted the significant roles played by mobility-related and social-demographic factors in the disease spreading process. Nevertheless, few studies examined the collective impacts of non-epidemiological factors on the spatiotemporal dynamics of infectious disease. In addition, the impacts of these influencing factors were primarily assumed to be stationary over time and space in the existing literature. The lack of consideration of these aspects will fail to reveal the interdependencies among modeling determinants and may result in biased model estimations.

To address the issues, the study aims to introduce a quantitative approach, named mobility-augmented geographically and temporally weighted regression model (M-GTWR), to investigate the heterogeneous effects of non-epidemiological factors on the spreading dynamics of the COVID-19. By relating pre-pandemic inter-county traffic data with the spatial adjacency, the M-GTWR quantifies the spatiotemporal effects of the social-demographic characteristics and human activity on the weekly average daily confirmed cases in the US. Our results suggest that counties with a high percentage of black population, a high household income level, a low education level, and a high unemployment rate are associated with more weekly average daily confirmed cases. Moreover, the impact of human activity is found to differ spatially. Grocery and pharmacy activities only show positive and statistically significant effects on the COVID-19 cases in rural counties, and the effects of the public transit activities are tightly related to the work from home policy and reopening strategies.

The estimated results of the M-GTWR model are obtained in Table 4. The M-GTWR model addresses the nonstationarity issue for the fundamentally heterogeneous and dynamic disease propagation and provides more efficient estimates by assuming the effects of the influencing factors are spatiotemporal heterogeneous. Compared with the global OLS model, the M-GTWR model improves AICc and adjusted \(R^{2}\) from 59779.59 and 0.54 to 26309.13 and 0.94.

In this study, we developed an M-GTWR model to investigate the effects of non-epidemiological factors on disease propagation. Specifically, we show that the proposed M-GTWR model is superior to the state-of-the-art benchmarks in capturing the spatiotemporal heterogeneity of disease dynamics during the COVID-19 outbreak. Our results find that the older, the black, and the Latino are more vulnerable to the COVID-19 than other population groups. The reason may be attributed to either physical weakness or low-risk awareness. The highly educated population is more likely to comply with the restrictions during the COVID-19 outbreak. For the commuting time, its median elasticity shows that a 1% increase in the commuting time to work results in a 0.22% increase in the weekly average daily confirmed cases. Finally, the change in human activity patterns also presents a mixed impact on disease dynamics. In particular, the scale of the impacts is found to be closely related to the activity intensity and activity types. The grocery and pharmacy activity is found to be significant in low population density areas. And activities associated with public transit usage lead to a positive impact on the weekly average daily confirmed cases. This indicates the major role played by the public transit during COVID-19 and implies the need to restrict public transit usage, especially in high-transit demand areas. These insights address the spatiotemporal effects of the non-epidemiological factors on the COVID-19 propagation.

Several implications for the high population density areas (e.g., New York City, counties in California, Washington, Arizona, Virginia, Minnesota, and Florida): Several implications in our study that are important for the low population density counties (e.g., counties in Arizona and counties in Massachusetts): The study explores the spatiotemporal effects of non-epidemiological factors on the COVID-19 propagation and addresses the heterogeneous effects of demographic characteristics and daily activity on disease propagation. However, there are some limitations in the study. First, the efficiency of the intervention strategies (e.g., wearing face masks, maintaining social distance, and handwashing) for mitigating the spreading of COVID-19 lack of exploration due to the limited data source. More importantly, since these strategies are at a great cost to the economy, the optimal control strategies to balance public health and freedom of movement, the economy, and society deserve further investigation. Second, although we estimated the effects of the several types of activities, we do not differentiate the risk level of detailed activities due to the data limitation (e.g., we estimated the effects of the recreation activities, but the exposure risk of the bar and book store might be different). The understanding of the exposure risk of detailed activities provides directional suggestions for the policy-makers in conducting control strategies for COVID-19 prevention. Third, the findings rely on the analysis of the aggregated county-level dataset. However, the lack of exploration of the microscopic behavior-related analysis would increase the uncertainty of underlying reasons. Thus, future studies should be more tailored to the demographics and socioeconomic of the particular location and groups. Besides, we used the sampling of US counties to construct the model. The applicability of insights remains to be tested for the rest of counties in the US and other countries. The model parameters can also be adjusted using the data from other locations.

By establishing a quantitative framework for identifying influencing factors of COVID-19 dynamics in the US, the study first concludes that the proposed M-GTWR achieves a substantial improvement over other benchmark methods in addressing the spatiotemporal nonstationarity issues in the disease dynamic data. Then, we obtain several key results from the study. High population density and the availability of public infrastructures will facilitate the spread of the disease. A 1% increase in population density and public road mileage leads to 0.63% and 1.03% more daily cases on average, respectively. Besides, the effects of socio-demographic attributes and the travel-related attributes differ significantly over time and the underlying location. Moreover, the effectiveness of limiting human contact through reduced human activity levels is found to vary significantly over space and time. The grocery and pharmacy activity is positively related to daily cases in about 30% of studied counties in the 12th week of 2020. This number decreases to 10% in the 48th week of 2020. This reveals that the general preventative non-pharmaceutical measures, such as work from home policy and travel restrictions, are unlikely to be universally effective over all subareas of a country. The insights derived in this study will provide important guidance for efficient resource allocation strategies (e.g., the distribution of medical resources) and non-pharmaceutical interventions for future disease mitigations and interventions.