The COVID-19 pandemic disproportionately impacted disadvantaged populations. Human mobility is a critical factor in disease transmission, but its interaction with social vulnerability remains unclear. Existing research has examined the effects of mobility on COVID-19 transmission and the relationship between vulnerability and pandemic outcomes separately. However, few studies have explored how social vulnerability modifies the association between mobility and transmission dynamics. This study aims to address this gap by investigating the heterogeneous effects of mobility on COVID-19 transmissibility across US counties with varying levels of social vulnerability. The complexity of social vulnerability, encompassing social, natural, and engineered systems, necessitates a multidimensional approach. Previous studies have used single or limited socioeconomic variables, neglecting the combined effect of multiple factors. Therefore, a comprehensive measure of COVID-19 vulnerability is needed to understand the heterogeneous mobility-transmission relationship. This study develops a COVID-19 Pandemic Vulnerability Index (CPVI) using principal component analysis (PCA) incorporating socioeconomic, health, and environmental factors to assess this relationship.

A substantial body of literature exists examining the impact of mobility on COVID-19 transmission using aggregated mobile phone data. Studies have demonstrated a strong correlation between mobility and disease spread, both at state and county levels in the US. Mobility reduction measures, like stay-at-home orders, were widely adopted as non-pharmaceutical interventions (NPIs). However, the moderating role of social vulnerability in this relationship has been largely unexplored. While some studies explored the relationship between vulnerability and mobility or vulnerability and pandemic outcomes, they did not address how social vulnerability differentially affects the mobility-transmission association. Existing vulnerability indices often focus on a limited set of sociodemographic variables, neglecting the combined effect of multiple factors. This study aims to improve upon these limitations by constructing a comprehensive CPVI and analyzing its interaction with mobility data.

This study utilized multisource data including daily COVID-19 case counts from USAFacts, county attribute data (demographics, socioeconomics, health, environment) from various sources including the CDC, and population mobility data from SafeGraph. A COVID-19 Pandemic Vulnerability Index (CPVI) was created using principal component analysis (PCA) on a dataset encompassing 17 variables across four dimensions: socioeconomic factors, demographics, health indicators, and environmental factors. The PCA reduced the dimensionality of the data while retaining most of the variance. The CPVI scores were divided into five quintiles representing different vulnerability levels. The effective reproduction number (Rₜ) was calculated for each county using a Bayesian approach based on daily new cases. A fixed effects model was used to estimate the relationship between mobility (intracounty and intercounty), CPVI, and Rₜ, accounting for the interaction between mobility and vulnerability. The analysis focused on the period from June 1 to August 31, 2020, after the initial lockdown measures.

The PCA identified four principal components explaining over 70% of the total variance in the dataset. These components reflected socioeconomic deprivation, health risks, healthcare access, and environmental factors. The CPVI revealed spatial variations in vulnerability across US counties, with highly vulnerable counties concentrated in specific regions. The analysis showed a positive correlation between CPVI and the percentage of days with Rₜ ≥ 1 (indicating high transmissibility). Counties in the top CPVI quintile experienced almost double the percentage of days with Rₜ > 1 (45.02%) compared to the lowest quintile (21.90%). Both intracounty and intercounty mobility were positively associated with Rₜ across all vulnerability levels. However, the strength of this association increased significantly with increasing vulnerability. For intracounty mobility, a 25% change was associated with an 1.81% Rₜ change in the lowest vulnerability quintile and a 15.28% change in the highest quintile. A similar pattern was observed for intercounty mobility, although the effect was less pronounced. The fixed-effects model demonstrated statistically significant positive effects of both IntraM and InterM on Rₜ across all vulnerability levels, with larger effects for higher vulnerability levels. Furthermore, the effect of intracounty mobility on Rₜ was notably greater than the effect of intercounty mobility in the most vulnerable counties.

The findings demonstrate that social vulnerability significantly amplifies the impact of mobility on COVID-19 transmissibility. Highly vulnerable communities experienced a disproportionately greater increase in transmission in response to changes in mobility, particularly intracounty mobility. This highlights the limitations of uniform social distancing policies and the necessity for tailored interventions considering community-specific vulnerabilities. The stronger effect of intracounty mobility compared to intercounty mobility in the most vulnerable counties suggests that within-community contact, possibly at the household level, plays a critical role in transmission in these areas. The observed heterogeneity in the mobility-transmission relationship necessitates localized and targeted public health responses. Reopening strategies should prioritize less vulnerable areas and implement stricter measures in more vulnerable regions.

This study provides compelling evidence for the importance of incorporating social vulnerability into strategies for managing the COVID-19 pandemic and future infectious disease outbreaks. The CPVI developed in this study offers a valuable tool for identifying and targeting vulnerable populations for effective public health interventions. Future research could explore the efficacy of specific interventions tailored to different vulnerability levels and the long-term health and socioeconomic consequences of the pandemic's disparate impact.

The study included only counties with at least one COVID-19 case by June 1, 2020, potentially excluding areas with lower incidence. Time-invariant factors such as pre-existing conditions and the cultural aspects of disease-prevention behaviors were not explicitly modeled, though the use of fixed effects helps control for such factors. The study relied on aggregated mobility data, which may not fully capture individual-level mobility patterns. The analysis focuses on the period after initial lockdown measures, limiting generalization to earlier phases of the pandemic. Finally, the study does not account for vaccination.