key: cord-0798386-qcbfyf1g
authors: Neelon, Brian; Mutiso, Fedelis; Mueller, Noel T.; Pearce, John L.; Benjamin-Neelon, Sara E.
title: Spatial and temporal trends in social vulnerability and COVID-19 incidence and death rates in the United States
date: 2021-03-24
journal: PLoS One
DOI: 10.1371/journal.pone.0248702
sha: 07c34efd1b2d5b9341530ec3167a2a05c9ad44a5
doc_id: 798386
cord_uid: qcbfyf1g

BACKGROUND: Socially vulnerable communities may be at higher risk for COVID-19 outbreaks in the US. However, no prior studies examined temporal trends and differential effects of social vulnerability on COVID-19 incidence and death rates. Therefore, we examined temporal trends among counties with high and low social vulnerability to quantify disparities in trends over time. METHODS: We conducted a longitudinal analysis examining COVID-19 incidence and death rates from March 15 to December 31, 2020, for each US county using data from USAFacts. We classified counties using the Social Vulnerability Index (SVI), a percentile-based measure from the Centers for Disease Control and Prevention, with higher values indicating more vulnerability. Using a Bayesian hierarchical negative binomial model, we estimated daily risk ratios (RRs) comparing counties in the first (lower) and fourth (upper) SVI quartiles, adjusting for rurality, percentage in poor or fair health, percentage female, percentage of smokers, county average daily fine particulate matter (PM(2.5)), percentage of primary care physicians per 100,000 residents, daily temperature and precipitation, and proportion tested for COVID-19. RESULTS: At the outset of the pandemic, the most vulnerable counties had, on average, fewer cases per 100,000 than least vulnerable SVI quartile. However, on March 28, we observed a crossover effect in which the most vulnerable counties experienced higher COVID-19 incidence rates compared to the least vulnerable counties (RR = 1.05, 95% PI: 0.98, 1.12). Vulnerable counties had higher death rates starting on May 21 (RR = 1.08, 95% PI: 1.00,1.16). However, by October, this trend reversed and the most vulnerable counties had lower death rates compared to least vulnerable counties. CONCLUSIONS: The impact of COVID-19 is not static but can migrate from less vulnerable counties to more vulnerable counties and back again over time.

a1111111111 a1111111111 a1111111111 a1111111111 a1111111111

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the cause of coronavirus disease 2019 (COVID- 19) , has created a global public health crisis since its onset in late 2019. As of late February 2021, there have been over 27 million confirmed COVID-19 cases and nearly a half a million deaths in the United States (US) alone [1] . Overwhelming evidence has found that the pandemic has disproportionately affected people of color, older individuals, and those of lower socioeconomic status [2] [3] [4] [5] [6] [7] [8] [9] [10] . Several studies have shown that African Americans have higher incidence and death rates than non-Hispanic whites [6, 8, 9, 11] . Two studies also reported that COVID-19 infection rates are greater in US counties and in states with high Latinx populations and monolingual Spanish speakers [4, 7] . Further, older age has been associated with an increased risk of death among those infected with COVID-19 [5, 12, 13] . Underlying health conditions and comorbidities may partially explain these associations [5] , but do not fully account for the disproportionate burden. Recent studies suggest that social determinants of health and community contextual factors contribute to these disparities, and that socially vulnerable communities are at highest risk for COVID-19 outbreaks [6, [14] [15] [16] .

Protecting vulnerable populations is critically important during the COVID-19 pandemic, as these groups are generally at higher risk for adverse health outcomes [17, 18] . Hurst et al. define vulnerability as an identifiably elevated risk of incurring greater wrong or harm [19] . One type of vulnerability-social vulnerability-has been used by the Centers for Disease Control and Prevention (CDC) to identify communities most at risk when faced with adverse events that may impact health, such as natural disasters or disease outbreaks. The CDC developed the social vulnerability index (SVI) to assist federal, state, and local governments in targeting and mobilizing resources for at-risk counties in response to adverse events.

Recent studies have demonstrated the importance of considering social vulnerability in both COVID-19 cases and deaths, although the findings have been somewhat inconsistent [20] [21] [22] [23] . Karaye et al. examined associations between the SVI and cumulative COVID-19 cases on May 12, 2020 [20] . They found that SVI total score was associated with increased rates of COVID-19. However, the authors found no association when they examined six states with high testing rates. Khazanchi and colleagues conducted an analysis of COVID-19 cases and deaths through April 19, 2020 , and found that those living in the most vulnerable counties (highest SVI) had greater risk of infection and death [22] . Nayak et al. examined associations between the SVI and COVID-19 incidence and case fatalities through April 4, 2020, and found a significant association between social vulnerability and case fatality but not incident cases [21] . Finally, Wang and colleagues found that social vulnerability and COVID-19 incident cases and deaths had spatially varying associations; however, the authors did not adjust for potential confounders [23] . Notably, all studies were cross-sectional and conducted at different time points early in the pandemic, which might contribute to the inconsistent findings. In fact, to date, no prior studies have examined longitudinal trends in social vulnerability and COVID-19 incidence and death rates in an effort to determine how these relationships change over time. Therefore, the purpose of this study was to examine temporal trends among counties with high and low social vulnerability and to quantify disparities in these trends over time.

We conducted a retrospective longitudinal analysis examining COVID-19 incidence and death rates from March 15, 2020 to December 31, 2020 for each of the 3,142 US county and county equivalents based on their unique Federal Information Processing Series (FIPS) codes [24, 25] . Specifically, we modeled the temporal trend in daily incidence and death rates for each county and assessed differential risks by county-level social vulnerability. We hypothesized that highly vulnerable counties would have higher incidence and death rates compared to less vulnerable counties and that this disparity would widen over time. The Institutional Review Boards at the Medical University of South Carolina and Johns Hopkins Bloomberg School of Public Health deemed this research exempt from review.

We obtained daily COVID-19 incident case and death data from USAFacts [26] and the Johns Hopkins Center for Systems Science and Engineering [27] . Because national data are only available at the county level, we used county as the geographic unit for our analyses. Moreover, because Johns Hopkins aggregates data for some counties (e.g., the five boroughs of New York) [28] , we opted to use the USAFacts data in our primary analysis, and conducted a sensitivity analysis using Johns Hopkins data. For both data sources, we downloaded daily incident case and death counts from March 15 to December 31, 2020. We obtained county population data from the 2019 population datafile compiled by the US Census Bureau [29] . We present data sources for all variables in Table 1 .

We used publicly available data from the CDC's Agency for Toxic Substances and Disease Registry to classify counties using SVI [30] . The SVI is a percentile-based measure of social vulnerability, or the resilience of communities to address stressors to health related to external hazards (e.g., natural disasters or disease outbreaks) [34] . The Geospatial Research, Analysis & Services Program within the Agency for Toxic Substances and Disease Registry created the SVI database to help public health officials identify communities that will most likely need support and resources during and after a hazardous event like a pandemic [30] . The overall index and each theme are scored from 0 to 1, with higher scores indicating greater vulnerability [30, 34] . The index was constructed using data from 15 variables from the US Census Bureau. A percentile rank was calculated for each of these variables and grouped among four themes of SVI that measure various aspects of vulnerability-these include Socioeconomic Status, Household Composition, Race/Ethnicity/Language, and Housing/Transportation [30, 34] .

The Socioeconomic Status theme is composed of percentile rank data for the following variables: percentage below poverty, percentage unemployed, per capita income, and percentage with no high school diploma. For Household Composition, the variables include percentage age 65 years and older, percentage age 17 years or younger, percentage age 5 years or older with a disability, and percentage of single-parent households. The Race/Ethnicity/Language theme encompasses percentage minority and percentage who speaks English "less than well". Finally, the Housing/Transportation theme includes data for the percentage of multiunit structures, percentage of mobile homes, percentage crowding, percentage having no vehicle, and percentage of group quarters.

For our analyses, we downloaded the 2018 county-level SVI data (the most recent available) for all 3,142 counties. One county was missing SVI data; for this county, we imputed SVI data using the national average.

We adjusted for several variables that could help explain the differential impact of COVID-19 on upper and lower SVI counties. These variables were chosen a priori based on previously reported associations with COVID-19 incidence and deaths [20] [21] [22] [35] [36] [37] [38] . As noted above, we provide data sources for all variables in Table 1 . These included the percentage of each county designated as rural, the percentage of female residents, the percentage of adult smokers in the county, the number of primary care physicians per 100,000 in each county, average daily temperature (degrees Fahrenheit), and average daily precipitation (inches). We also controlled for the percentage of residents in poor or fair health, a validated measure of the overall health and comorbidity burden of the population [32] . We additionally controlled for population density, defined as the number of residents per square mile [33] and the average particulate matter of diameter � 2.5 micrometers (PM 2.5 ) [32] , a measure of fine particulate air pollution that can compromise respiratory function. Finally, we controlled for the daily proportion of COVID-19 Viral (RT-PCR) tests performed in each state (county-level data are not currently available). We converted the number tested to a proportion by dividing the number of tests by the state population size, obtained from the US Census Bureau's population estimate dataset [39] . Each of the above variables was included in the models examining the impact of overall SVI on cases and deaths as well as the SVI themes. However, because each SVI theme includes only the SVI variables specific to that theme, we included additional adjustment variables in the analyses involving specific SVI themes. For the Socioeconomic Status theme, we included the percentage of county residents aged 65 years and older, the percentage of non-Hispanic Black residents, and the percentage of Hispanic residents. For the Household Composition theme, we adjusted for the percentage below the federal poverty line as well as percentage of NHB and Hispanic residents; we did not adjust for age, as this is included as part of the theme. For Race/ Ethnicity/Language, we adjusted for age � 65 and over and percent poverty. For the Housing/ Transportation theme, we adjusted for age � 65, percent poverty, percent NHB and percent Hispanic. Finally, as a sensitivity analysis, we fit unadjusted models for overall SVI and the four themes.

We fit Bayesian hierarchical negative binomial models with daily incident cases and daily deaths for each county as the outcomes. The models included penalized cubic Bsplines for both the fixed and random (i.e., county-specific) temporal effects, with knots placed approximately every two weeks over the study period (20 knots total). The models also included county population as an offset on the log scale to convert the case and death counts to population-adjusted rates. The models further adjusted for the variables described above.

To avoid overfitting the temporal splines, we assigned ridging priors to the fixed and county-specific spline coefficients-i.e., independent, mean-zero normal distributions with shared inverse gamma variances [40] . We assigned weakly informative normal priors to the corresponding regression parameters. We assigned a gamma prior to the negative binomial dispersion parameter. We developed an efficient data-augmented Gibbs sampler to aid posterior computation [41, 42] . For both the incidence case and death rate models, we ran the Gibbs sampler for 2,500 iterations with a burn-in 500 to ensure convergence.

To report results, we compared counties in the top or upper SVI quartile (most vulnerable) to those in bottom or lower SVI quartile (least vulnerable). For both quartiles, we graphed the posterior mean incidence and death rate trends for the reference covariate group along with the corresponding 95% posterior intervals (PIs). We also reported adjusted risk ratios (RRs) and 95% PIs comparing the upper and lower quartiles on each day for the overall SVI and its themes.

As sensitivity analyses, we ran unadjusted models as well as models based on Johns Hopkins case and death data. We conducted all analyses using R software version 3.6 (R Core Team 2019, R: A language and environment for statistical computing, R Foundation for Statistical Computing, Vienna, Austria).

The final analytic sample comprised 917,464 observations (3,142 counties x 292 study days). There were 786 counties in each of the upper and lower SVI quartiles. Table 2 presents county demographics overall and by upper and lower SVI quartile. The two quartiles were similar with respect to gender, age, smoking, and the total number of RT-PCR tests performed during the study period. The quartiles differed with respect to percent poverty, race/ethnicity composition and temperature, with more vulnerable counties having higher average daily temperature, suggesting that many of these counties are located in the southern US. Fig 1A presents the per capita incidence trends (expressed as cases per 100,000) for the upper (most vulnerable) and lower (least vulnerable) quartiles of SVI from the unadjusted analysis. For counties in the upper quartile, the average incidence increased steadily from March 15 (0.10 cases per 100,000; 95% PI: 0.09, 0.12) to May 12 (9.36 cases per 100,000; 95% PI: 9.08, 9.61). The incidence leveled off in mid-May before a precipitous increase through July 30 (31.10 cases per 100,000; 95% PI: 30.52, 31.68). The incident cases receded in late summer and early fall, before a final uptick from mid-September to December 31 (57.21 cases per 100,000; 95% PI: 54.94, 59.43). By comparison, the least vulnerable quartile exhibited a gradual increase in incidence through the summer and fall, achieving a maximum incidence rate of 60.23 cases Fig 1B presents the posterior mean adjusted RRs comparing the upper and lower quartiles on each day. On March 15, the RR for incident cases was 0.95 (95% PI: 0.82, 1.07), suggesting that most vulnerable counties had, on average, fewer cases per 100,000 than less vulnerable 

United States Laboratory Testing

Assessing Differential Impacts of COVID-19 on Black Communities

COVID-19 and African Americans

Risk for COVID-19 infection and death among Latinos in the United States: Examining heterogeneity in transmission dynamics

Predictors of mortality for patients with COVID-19 pneumonia caused by SARS-CoV-2: a prospective cohort study

Social Vulnerability and Racial Inequality in COVID-19 Deaths in Chicago. Health education & behavior: the official publication of the Society for Public Health Education

COVID-19 Pandemic: Disparate Health Impact on the Hispanic/Latinx Population in the United States. The Journal of Infectious Diseases

Disparities in COVID-19 Testing and Positivity in New York City. American journal of preventive medicine

Disparities in the Population at Risk of Severe Illness From COVID-19 by Race/Ethnicity and Income. American journal of preventive medicine

Vulnerable Populations: Weathering the Pandemic Storm

The coronavirus is infecting and killing black Americans at an alarmingly high rate

Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study

COVID-19 in older people: a rapid clinical review

Examining Social Determinants of Health, Stigma, and COVID-19 Disparities

Importance of collecting data on socioeconomic determinants from the early stage of the COVID-19 outbreak onwards

The Structural and Social Determinants of the Racial/Ethnic Disparities in the U.S. COVID-19 Pandemic: What's Our Role? Am J Respir Crit Care Med

What Now? Epidemiology in the Wake of a Pandemic. American journal of epidemiology

Focusing on Vulnerable Populations During COVID-19

Vulnerability in research and health care; describing the elephant in the room?

The Impact of Social Vulnerability on COVID-19 in the U.S.: An Analysis of Spatially Varying Relationships. American journal of preventive medicine

Impact of Social Vulnerability on COVID-19 Incidence and Outcomes in the United States

County-Level Association of Social Vulnerability with COVID-19 Cases and Deaths in the USA

The spatial association of social vulnerability with COVID-19 prevalence in the contiguous United States

Remote Sensing, and Geospatial Data: How many counties are in the United States?

US Coronavirus Cases and Deaths

COVID-19 case counts 2020

FAQ-COVID-19 UNITED STATES CASES BY COUNTY 2020

United States Census Bureau. State Population Totals

Centers for Disease Control and Prevention/Agency for Toxic Substances and Disease Registry/ Geospatial Research A, and Services Program

Centers for Disease Control and Prevention County Population Estimates

County Health Rankings & Roadmaps: Rankings Data & Documentation 2020

National Oceanic and Atmospheric Administration

Measuring Community Vulnerability to Natural and Anthropogenic Hazards: The Centers for Disease Control and Prevention's Social Vulnerability Index

Exposure to air pollution and COVID-19 mortality in the United States: A nationwide cross-sectional study

SARS-CoV-2 and COVID-19: A review of reviews considering implications for public health policy and practice

Association of cardiovascular disease and 10 other pre-existing comorbidities with COVID-19 mortality: A systematic review and metaanalysis

Associations Between Primary Care Provider Shortage Areas and County-Level COVID-19 Infection and Mortality Rates in the USA

National Population Totals and Components of Change

High dimensional structured additive regression models: Bayesian regularization, smoothing and predictive performance

Fully Bayesian inference for neural models with negative-binomial spiking

Bayesian negative binomial regression for differential expression with confounding factors

Current smoking, former smoking, and adverse outcome among hospitalized COVID-19 patients: a systematic review and metaanalysis

A nicotinic hypothesis for Covid-19 with preventive and therapeutic implications

The Three Steps Needed to End the COVID-19 Pandemic: Bold Public Health Leadership, Rapid Innovations, and Courageous Political Will

State-level impact of social distancing and testing on COVID-19 in the United States

The LZIP: A Bayesian latent factor model for correlated zero-inflated counts

Spatiotemporal hurdle models for zero-inflated count data: Exploring trends in emergency department visits

Bayesian Zero-Inflated Negative Binomial Regression Based on Polya-Gamma Mixtures

Prisons and custodial settings are part of a comprehensive response to COVID-19

American Indian Reservations and COVID-19: Correlates of Early Infection Rates in the Pandemic

COVID-19 in nursing homes

Nursing homes and COVID-19: We can and should do better

This Time Must Be Different: Disparities During the COVID-19 Pandemic

The COVID-19 Pandemic: a Call to Action to Identify and Address Racial and Ethnic Disparities

Early State Vaccination Data Raise Warning Flags for Racial Equity