Identifying At-Risk Communities and Key Vulnerability Indicators in the COVID-19 Pandemic
The COVID-19 pandemic’s impact on health in the United States has been devastating, with over 81 million COVID-19 cases reported to the CDC (as of May 2022), over 4.6 million patients hospitalized, and over 990,000 deaths [CDC 2022]. However, the impact of COVID-19 is much greater than just its case count. The pandemic also caused housing emergencies, food shortages, and population-level mental health crises. As Americans have endured the past 30+ months, it has become clear that some communities have been, and continue to be, disproportionately impacted by the effects of the pandemic. This disparity is the result of environmental and historical differences that impact health, including economic status, education, race, ethnicity, infrastructure, and more. For example, research has found that predominantly Black counties have higher COVID-19 infection and mortality rates compared to counties that are predominantly White [Yancy 2020]. Investigating these differences is key to examining how health disparities occur. At this time, the CDC uses their Social Vulnerability Index (which was created before the pandemic started), along with COVID-19 case and death rates, to understand community health needs. However, this is a limited view of the vast impact COVID-19 has had and will have on the country. This impact can be better described with a new model of vulnerability, like the Community Vulnerability Index for COVID-19 (CVI). This model, developed by Community Insight & Impact, examines recent and existing health risks across several categories of lasting impact due to COVID-19.
The CVI is a set of three metrics that describe a community’s risk to different adverse impacts of the COVID-19 pandemic at the county level. The three metrics are COVID-19 case severity (Severity), risk of economic harm (Economic Harm), and need for mobile health resources (Mobile Health). The Severity metric measures the risk of hospitalization due to COVID-19 present in a county. Economic Harm measures a county’s risk of severe, negative economic impact due to COVID-19. Finally, Mobile Health measures the area’s need for alternative healthcare delivery access, including mobile health clinics, telehealth services, and health app solutions. Each metric score consists of a combination of quantitative variables, each weighted differently according to an extensive literature review (learn more about our methodology here).
Variable weighting is used when working with complex equations involving many variables. When weighting, variables are assigned comparatively smaller or larger impacts on the outcome being studied, in order to accurately predict community risk. Variable weighting for the CVI was informed by public health, social science, and statistical research studies. A proxy study was conducted to further validate these weights and determine if they needed to be adjusted, by creating a machine learning model trained on the raw dataset (which contains both metric variables and non-metric variables) that then predicted a proxy outcome for each CVI metric. The machine learning model-derived importances for each variable were then compared to the initial weights given to the variables. This highlighted any differences.
Proxy variables are used when the outcomes described by our metrics cannot be measured directly. A related variable is selected and measured in place of the selected metric. For this analysis, the proxy outcome for each metric was chosen based on relevant literature supporting each selection as a strong indicator of each intended measurement. For the COVID-19 case severity (Severity) metric, COVID-19 Hospitalization Rate per 100,000 populations was used as a proxy outcome. The hospitalizations were measured cumulatively from March 2020 to April 2021. The proxy outcome used for the risk of economic harm (Economic Harm) was cumulative unemployment initial claims between January and November 2020 per 100 people in the 2019 labor force, which describes individual economic status rates. Lastly, for the need for mobile health resources (Mobile Health) metric, a proxy outcome of Ratio of Hospitals per 100,000 population was used. This was used because counties with a low number of hospitals have a higher need for mobile healthcare services.
Several discoveries made during the proxy analysis were interesting and sometimes unexpected. For the Severity metric, the most powerful predictor of a person’s COVID-19 infection being severe enough to require hospital admittance was the existing COVID-19 case rate in a community. This means that the most predictive feature for determining hospitalizations in a community is the prevalence of COVID-19, regardless of the presence of comorbidities, like hypertension, diabetes, etc. This indicates that keeping case counts low overall is the most effective objective in reducing overall COVID case severity, rather than specifically focusing on at-risk people. Even more surprising and interesting was the strong predictive value of the percent of children enrolled in free and reduced lunch, which was uniquely predictive compared to other poverty indicators. Interestingly, this indicates that the ‘percent of children enrolled in free and reduced lunches’ variable captures information that is missed by other commonly used measures of poverty and is more useful for understanding community risk. The Economic Harm metric held several impactful variables, including the percentage of individuals in a community with severe housing cost burden — spending over half of their income on housing. Households that have to spend large proportions of income on housing may be prevented from accruing emergency funds or accessing healthcare, and therefore are at greater risk of financial risk from the pandemic. In addition, the unemployment rate was found to be the most important feature in the Economic Harm metric, underscoring the importance that pre-existing economic difficulty has on communities experiencing disparate impacts of crises. The Mobile Health metric was most predicted by variables describing access to social and physical infrastructure, such as access to exercise opportunities and social association rate. This suggests that communities without these structural resources have additional need for alternative healthcare access.
Many of the predictive variables in this model assess how a community’s pre-existing resources can impact their experience of the pandemic, even if those resources are outside of the obvious realms of health access. The metrics the CVI presents can measure both the span and magnitude of the COVID-19 pandemic’s impact on communities across the United States by using socioeconomic data. Our recently published paper both establishes the validity of the CVI’s predictions, and raises several variables as candidates for further research. The CVI can address the needs of our most at-risk communities by providing possible causal pathways through which societal resources and health access affect them. The model’s findings are useful for several organizations and governmental bodies as they seek to support their communities during the pandemic.
This article is a synopsis of CII’s recently published paper, Identifying At-Risk Communities and Key Vulnerability Indicators in the COVID-19 Pandemic, which details the creation of the CVI. We hope that this adds insight into how an inclusive, open-source data source like CVI is constructed. You can read the paper in full at: https://www.medrxiv.org/content/10.1101/2021.09.19.21263805v1.full
Written by Maisie Conrad & Annina Christensen
