key: cord-0848596-yeb4ke3j
authors: Fefferman, Nina; Chen, Chien-fei; Bonilla, Gregory; Nelson, Hannah; Kuo, Cheng-Pin
title: How limitations in energy access, poverty, and socioeconomic disparities compromise health interventions for outbreaks in urban settings
date: 2021-11-01
journal: iScience
DOI: 10.1016/j.isci.2021.103389
sha: c2d3753ea43a63496c15b322e2c58689ac86f521
doc_id: 848596
cord_uid: yeb4ke3j

Low-income households (LIHs) have experienced increased poverty and inaccess to healthcare services during the COVID-19 pandemic, limiting their ability to adhere to health protective behaviors. We use epidemiological models to show how a households’ inability to adopt social distancing due to constraints in utility and healthcare expenditure can drastically impact the course of disease outbreak in five urban U.S. counties. LIHs suffer greater burdens of disease and death than higher income households, while functioning as a consistent source of virus exposure for the entire community due to socioeconomic barriers to following public health guidelines. These impacts worsened when social distancing policy could not be imposed. Health interventions combining social distancing and LIH resource protection strategies (e.g., utility and healthcare access) were the most effective in limiting virus spread for all income levels. Policies need to address the multidimensionality of energy, housing, and healthcare access for future disaster management.

The broader impacts of energy insecurity and poverty on household environments are related to mental and physical health (Snell, Bevan and Thomson, 2015; Hernández, 2016; Mayer and Smith, 2019; Boateng et al., 2020) , which is a serious situation during the COVID-19 pandemic (Castán Broto and Kirshner, 2020; Chen et al., 2020; Graff and Carley, 2020; Memmott et al., 2021) ; for example, anxiety, stress, and depression are associated with living in poor housing conditions (Harrington et al., 2005; Liddell and Morris, 2010; Hernández, Phillips and Siegel, 2016) . Energy insecurity is defined as households' inability to meet basic energy needs (Hernández, 2016; Memmott et al., 2021) , while energy poverty is generally measured by the inability to pay for utility bills (Drehobl, Ross and Ayala, 2020) . Certain socioeconomic groups, including low-income, senior, non-white, and renter households, and those without a fulltime job, spend significantly higher percentages of their income on energy costs than other groups. Households that experience intense energy poverty often have to make a fundamental trade-off between health-supportive resources like medicine or healthcare to afford utilities, such as water, gas, and electricity (Sovacool and Dworkin, 2015; Xu and Chen, 2019; Drehobl, Ross and Ayala, 2020; Graff and Carley, 2020) . This situation has adverse effects on individual and family health, especially for households at or below the poverty line, where marginal increases in household expenditure on necessities can compromise the ability to seek basic medical care (Jessel, Sawyer and Hernández, 2019) . Therefore, the narrative of adhering to social distancing and quarantine protocols may be fundamentally flawed when considering families under severe financial stress.

While many recent public debates and academic studies have focused on how health information and beliefs factor into the adoption of COVID-19 mitigation efforts, household and economic limitations play a significant role in the ability to adopt behaviors that involve J o u r n a l P r e -p r o o f spending more time at home, which sounds deceptively cost-free, but has potential financial and health costs, such as lost wages from voluntary adherence to stay-at-home orders by nonessential employees. More importantly, even if one remains employed, there are increased costs from used supplies and utilities due to increased hours spent at home (BBC, 2020; Meinrenken et al., 2020; U.S. Energy Information Administration, 2020b) .

Current energy poverty research is mainly carried out at the individual, small-scale level (e.g., building, person, household), but not at the population level (e.g., national building stock, cities, building typologies). The limited availability of detailed residential energy consumption data during the pandemic makes it challenging to understand the potential impacts of energy poverty and income disparities on health outcomes at the population level. Deeper insights into the presence and persistence of energy poverty and socioeconomic constraints during staying-at-home orders across populations are severely limited, thereby undermining effective health policy and interventions. Further, many policies have provided temporary protection against utility shut-offs or evictions due to non-payment, which has affected households of different income levels, so that counties or cities themselves may have differential success in outbreak control and subsequent population health outcomes.

This study moves beyond extant COVID-19 studies to demonstrate the multidimensional and interconnected factors of energy poverty, socioeconomics, healthcare resources, essential workers, health interventions (e.g., social distancing policies), and the spread of COVID-19 cases. Of course, while this study focuses on energy due to the expertise of the authors, to ensure we are making robust assumptions, these models apply to any household expenditure tradeoffs to ensure health and security during a pandemic. This study also presents the epidemiological models to explicitly consider how energy and household economic trade-offs might affect community-level success at mitigating COVID-19. This study is based on Giddens' theory of structuration (Giddens, 1984; Bryant and Jary, 1991) , which engages the relationship between social structure and human agency. The theoretical assumption here is that social structural context is something that both constrains and enables; social structure depends on the agency of individuals, but agency is also enabled and constrained by rules and resources (Whittington, 2015) .

This study highlights how energy poverty can drastically stratify the risks of COVID-19 infection, how contact rates within and across households are likely to be affected by public health intervention policies (i.e., social distancing and resource protecting policies), and how the synergy between those effects increases the poverty faced by the entire population. We estimated energy poverty of low-, middle-, and high-income households (LIHs, MIHs, and HIHs, respectively) in five U.S. counties during the first phase of stay-at-home orders from April to July 2020. This study also considers the age-based demography of the county to estimate the percent likely to be employed and thus affected by shut-downs, as well as the likely proportion of a county's workforce classified as essential workers, who would be exempted from social-distancing policies, and assumed over-representation of essential workers in LIHs, MIHs, and HIHs.

Energy poverty were considered as a proxy for inelastic costs that directly support household health and safety in this study. Income differences in energy poverty led to differences in COVID-19 exposure risks. They can critically influence the duration of illness and concomitant risks of death, assuming that limited economic resources compromise adequate access to healthcare before and during infection. We further consider households' skewed ability to perform their jobs while safely practicing social distancing. Specifically, we examine four cases of health intervention policies, including 1) "Do Nothing," 2) "Social Distance Only," 3) "Economic Support," and 4) "Try Everything." This study then parameterized the epidemiological model to reflect the conditions of the five counties, chosen for their differences among the relevant socioeconomic metrics. Specially, we used a 3-tiered SEIR model of COVID-19 in a population stratified into three socioeconomic levels: HIHs, in which there was no direct trade-of and families can be assumed to absorb the opportunity costs of social distancing without compromising any other health-supportive resources; MIHs, in which families can absorb the costs of social distancing while maintaining other healthsupportive resources for up to a year; and LIHs, in which families do not have a sufficient economic buffer to be able to maintain health-supportive resources while engaging in social distancing.

Our model demonstrated that the socioeconomic status (i.e., number of households unable to adopt social distancing due to the constraints of utility and household expenditure) of a county drastically influenced the expected course of an epidemic outbreak in the population. Under assumed uniform etiology and mixing patterns across cities (i.e., discounting the differences based on access to healthcare, public transportations, and underlying health conditions not correlated directly with income, instead of focusing only on demographic and economic differences), we observed that the model for each county produced drastically different baseline results in the expected outbreak size over the first 120 days after the introduction of a novel COVID-like infection (Fig. 1) .

In understanding the dynamics, our model also showed how, after the introduction of novel infection (for simplicity and consistency, we assumed this to be introduced via MIHs), we observed critically different patterns in which socioeconomic strata of households were likely to shoulder the poverty of disease and death overtime as the COVID outbreak progressed (Fig. 2) . For example, in Los Angeles, there was an early transition in both disease and death poverty from MIHs to LIHs (Fig. 2e ,f), while Philadelphia experienced a longer delay after introduction, before the poverty of cases shifted from the initially infected MIHs to LIHs (shifting from day 20 to day 30; Fig. 2e ,i); the lag in the shift of death poverty was even more substantial (shifting from day 37 to day 65; Fig. 2f ,j). Hidalgo and Los Angeles have very high ratios of LIHs and HIHs compared to the other counties studied, and the mixing rates of the populations yield a slower COVID exposure for the LIH community in these counties; however, this trend is the result of slower progress in spread rather than long-term expected dynamics (note the lower total case and death numbers). The relative dynamics of MIH and LIH cases in Allegheny and Oakland is due to the graphs presenting actual numbers of cases, rather than per capita cases. The mixing rates of different sizes between subpopulations mean that the epidemic growth is projected to slow in absolute case numbers in the LIHs in Allegheny, Oakland, and Los Angeles, relative to MIH cases. For HIH households, our evidence-based assumption is that more households can effectively adopt social distancing strategies while maintaining financial security, meaning that HIHs will be less impacted than either MIHs or LIHs. In other words, LIHs' inability to make overall health-supportive choices (e.g., seeking healthcare, affording medicine, social distancing) due to economic limitations mean they were less able to avoid the infection themselves. LIHs both caught and transmitted the disease more quickly than MIHs or HIHs and acted as a source of ongoing exposure to higherincome households. Therefore, they suffered the most significant poverty and functioned as the greatest barrier to effective, population-wide outbreak control.

Considering the outcomes of health interventions on these populations, we observed that a county's socioeconomic composition also drastically impacted the intervention's effectiveness (Fig. 3) . For the interventions meant to reduce overall transmission as a blanket policy, affecting socioeconomic sub-populations differently (e.g., "Social Distance Only" policy), we see a ~60% to 78% reduction in symptomatic cases from the baseline scenario ("Do Nothing"). In the absence of social distancing, however, health intervention policies aimed solely at supplementing the economic hardships faced by lower-income populations, who would then be less constrained in the individual choices available to them for selfprotection (e.g., by accessing appropriate medical care or paying utilities that support household health; the "Economic Support" policy), we were still able to achieve between 6% to 14% reductions in symptomatic COVID cases over the first 120 days of an outbreak. When both resource protective strategies and social distancing policies were imposed together ("Try Everything"), the reduction in total cases ranges from 66% to 81%; however, the improvement of the combined strategies was synergistic rather than additive, ranging between an additional 2% to 5% improvement, depending on the socioeconomic composition of the county.

We repeated the same comparison of outbreak intervention scenarios, using the models adapted to fit the observed case incidence data from the COVID-19-outbreaks in each county (i.e., baseline parameters for interaction and social distancing were tailored so that the resulting epidemic curve produced similar growth and cumulative symptomatic case counts over 120 days after the first identified case of COVID-19 to the reported outbreak for those counties post-COVID-19 introduction; Fig. 4 ). In this scenario, our model suggests that each county's social interaction rates and underlying healthcare resource accessibility drastically impacted the expected outcome of health interventions, both in magnitude within each county and relative impact across counties (compare Fig. 3 with Fig. 4) . Unsurprisingly, social distancing was more effective in more densely populated counties (now observable due to the tailored interaction rates); for example, the tailored results of Los Angeles showed a reduction in COVID cases of over 90% (Fig. 4) , where in the untailored COVID case, Los Angeles achieved 60% reduction. Given these tailored scenarios, however, economic support policies alone were capable of achieving up to 38% reduction in cases. This result means that LIHs acted as drivers of the ongoing outbreak for the entire community due to their economic limitations whether lockdowns were achievable or not, but there is a greater impact when social distancing cannot be imposed. Therefore, resource protection strategies tailored to alleviate financial constraints for LIHs can protect the whole population. While economic support strategies were seen to be more effective overall, their benefits over lockdown policies were reduced, meaning resource protection strategies may be an effective strategy in the absence of social-distancing mandates. Still, they may not be cost-effective to enact once lockdown policies are in place.

This study demonstrates how household economic constraints of utility and health expenditure may affect a county's ability to effectively enact pandemic mitigation policies. The findings suggest that the security of household utilities is necessary to support household health and safety. Additionally, a county's social interaction rates and health infrastructure accessibility significantly impacts the outcomes of health policy interventions, so that lowincome households bear most of the poverty of disease and death. Inequities in access to resources (e.g., utilities and healthcare) and health intervention policies (e.g., social distancing) hinder low-income households' ability to protect themselves from infection. The findings provide implications for management of future disasters beyond the COVID-19 pandemic. In particular, there is a critical need for policies to address energy and healthcare affordability and accessibility among vulnerable communities.

There are a few limitations to the present study. First, this paper focuses on the effects that household expenditure, including healthcare and utilities, have on counties' risks of infection during outbreak situations; future research can separate different household burdens such as expenses for rent, food, transportation, etc. Second, this study did not separate racial background from other demographic differences in economic stratification to highlight how financial resources themselves have the potential to drastically stratify the risks to households. Future research could investigate the interconnected constraints of racial/ethnic background, socioeconomics, and disease exposure. Third, this study did not include agebased probabilities of infection or death, to focus on household economics alone can impact economic-epidemiological dynamics. However, it should be noted that these differences are contributing factors in the case and death counts observed for each county due to the agespecific nature of original-strain COVID-19 outcomes; therefore, by fitting the model independently for each location, these factors are captured implicitly. Finally, we tailored the mixed rates of each county to approximate the magnitude of the outbreak in the "do nothing" case by August 2020, while also discussing these cases as "pure strategies", to highlight how emergent household necessities (e.g., the ability to socially distance or afford electricity) can complicate a county or city's ability to control COVID-19 and other illnesses. It is likely, however, that every country had a mixture of policies that shifted over time. Future research addressing these limitations may further clarify the ways in which socioeconomic, household, and utility constraints worsen a county's overall infection exposure risk. This project was conceived by C.-F. Chen and N. Fefferman. G. Bonilla compiled the majority of data used in the analysis and wrote the data sources; C.-P. Kuo estimated energy consumption poverty. N. Fefferman created the math models and analyzed the data. C.-F. Chen, and N. Fefferman lead the manuscript writing. H. Nelson wrote the literature review and edited the entire manuscript.

The authors have no financial or non-financial interests associated with the material in this manuscript.

One or more of the authors of this paper self-identifies as an underrepresented ethnic minority in science. One or more of the authors of this paper self-identifies as a member of the LGBTQ+ community. One or more of the authors of this paper self-identifies as living with a disability. One or more of the authors of this paper self-identifies received support from a program designed to Increase minority representation in science

Differences in un-fitted model outcomes across counties due solely to differences in demographic and socioeconomic make-up over the first 120 days after the first identified case.

The left column shows symptomatic infectious cases overtime in each socioeconomic category of households in each county. The right column shows cumulative deaths over time in each socioeconomic category of households and as an all-households total in each county.

Delta labels indicate the percentage improvement from the combined strategy above that achieved by social distancing alone.

Delta labels indicate the percentage improvement from the combined strategy above that achieved by social distancing alone.

Detailed methods are provided in the online version of this paper and include the following: 

Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Chien-fei Chen (cchen26@utk.edu).

This study did not generate new unique reagents.

Data and Code Availability  This paper analyzes existing, publicly available data. Dataset accession numbers are listed in the key sources table. 

Code for the epidemiological model was written in MATLAB and is available from the lead contact upon request.  All additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

J o u r n a l P r e -p r o o f

This study selected five counties for epidemiological comparison: Los Angeles, CA; Philadelphia, PA; Oakland, MI; Allegheny, PA; and Hidalgo, TX, based on patterns in vulnerability ranking across selected socioeconomics and health-related variables between each county. The selection of counties was based on population density and per capita income, as well as percentages of people below the national poverty level, people of racial minorities, people without insurance, households with energy poverty by county, and households that spend 30% or more of their income on primary care physicians. Counties were chosen to be maximally distinct from each other in their pattern of vulnerability across these measures (using pairwise Euclidean distance per metric). The following section describes each specific variable source. 

The CCVI was used to select our countries and measure socioeconomic and health vulnerabilities at the county-level to indicate the communities that may be less resilient to the impacts of the pandemic (Surgo Ventures, 2021). The CCVI builds on the U.S. Center for Disease Control and Prevention's (CDC) Social Vulnerability Index (SVI), a validated metric that uses census tract and county-level data (CDC Social Vulnerability Index, 2020). The CCVI's six themes include (1) socioeconomic status, (2) household composition and disability, (3) minority status and language, (4) housing type and transportation, (5) epidemiological factors, and (6) healthcare system factors. Each county is ranked from least vulnerable to most vulnerable in each of these categories.

For this study, data on COVID-19 cases and deaths was observed from January 21 to July 31, 2020, using data from John Hopkins' University (COVID-19 United States Cases by County, 2021) and USA FACTS (US Coronavirus Cases & Deaths by State, 2021). The data sources use three main methods to collect this data: first, by drawing aggregate county-level data from the Covid Tracking Project, John Hopkins' utilizes data from 56 U.S. states and territories under CDC guidelines for test positivity (Prevention, 2020; Project, 2021) . Second, USA FACTS indicates where presumed cases are included as positive cases and adjusted per capita to represent the cumulative total. Lastly, USA FACTS estimates the gaps in daily cumulative cases and deaths by direct referencing or scraping from state and local agencies (FACTS, 2021). Both sources were used in validation of model fit for cases and deaths independently for each county, directly capturing local factors (such as age distribution) that could be expected to influence the dynamics of the outbreak, to ensure accuracy.

Public intervention policy: Using Safegraph mobility data in 2020 (Social Distancing Metrics, 2020), this study analyzed the state-level stay-at-home and social distancing orders that limited movement from areas of residence to places of interest. Safegraph is a data consortium that provides accurate location data for human migration patterns and has been used in various COVID-19 studies (Charoenwong, Kwan and Pursiainen, 2020; Huang et al., 2020; Lamb, Kandula and Shaman, 2020; Chang et al., 2021) . The anonymized data in the present study were collected utilizing cellphone pings coupled with the average dwell time per day for each county. The number of devices and the length of time at home were then averaged to create a proxy representing the change in movement for each county population compared to non-pandemic conditions. It is important to note that public intervention policy was only used in the "COVID-19" fitted scenarios and not in the "random disease in the cities using this model shape, but not with COVID parameters" case. Further, county mandate information does not include the factors such as social gatherings, movement restrictions, and curfews; therefore, this paper assumed the counties were following state social distancing guidelines.

Additionally, instead of modeling a city's specific public intervention policies over time to avoid limiting the scope of such interventions to specific actions, we chose to examine the four following cases of policies: 1) "Do Nothing" -there were no social distancing policies and the expected differential economic impacts on household health-supportive spending remain in place, such as LIHs having less access to healthcare or health-supportive utilities due to economic constraints; 2) "Social Distance Only" -those who were not essential workers were allowed to social distance, but nothing was done to alleviate disparities in health-supportive spending for LIHs; 3) "Economic Support" -there was no social distancing attempted, but there were programs ensuring that pandemic-related loss of income would not compromise household health (e.g., policies that limited the ability of utility companies to disconnect services, public assistance covering medical care and/or medicine, etc.); and 4) "Try Everything" -social distancing was enacted and supplemented by policies that alleviate additional economic poverty on LIHs. In reality, every county had a mixture of these policies, but we tailored the mixed rates of each county to approximate the magnitude of the outbreak in the "do nothing" case by August 2020. Further, while each county has taken different policy actions over time, we discussed these cases separately as "pure strategies" to highlight how emergent household necessities (e.g., the ability to socially distance or afford electricity) can complicate a county or city's ability to control COVID-19 and other illnesses.

Essential workers: An essential worker provides public health and safety, essential products, or other infrastructure support during the COVID-19 pandemic; however, they are more likely to be exempted or prohibited from adopting social distancing policies. Workforce statistics for essential workers were retrieved by analyzing the number of workers per industry and calculating each industries' labor force percentage from the U.S. Bureau of Labor Statistics (BLS) (U.S. Bureau of Labor Statistics, 2021) and is based on methodology proposed in a recent study by the United Way (Area, 2021) . The state-level statistics were used as a proxy for each of the counties because county-level data were not available.

We used a combination of demographic factors and economic levels to estimate population levels of contact throughout different socio-demographic strata. The total population and percent of people over 65 years old and less than 17 years old were retrieved from the CDC COVID-19 index (Surgo Ventures, 2021). The estimates of household income level were based on nationally adjusted household sizes and cost of living relative to the area, as well as the percentage of households by county-level that represented low-, medium-, and high-income were estimated using data from the Pew Research Center (Bennett, Fry and Kochhar, 2020) . We defined LIHs as two-thirds of the national median, medium-income as two-thirds to double the median, and high-income as more than double the median.

We established the 2020 county-level energy consumption poverty (ECB) database for LIHs. Due to the lack of official energy consumption data at zip code or county-level in 2020, the historical energy expenditure (electricity, fuel, and natural gas) and poverty estimation among LIHs were collected from the Low-Income Energy Affordability Data (LEAD) Tool by the National Renewable Energy Laboratory (NREL) and the U.S. Department of Energy (DOE) (Ma et al., 2019) and U.S. Energy Information Administration (EIA) (U.S. Energy Information Administration, 2020a). Technically, the 2020 county-level LIHs' ECB data was estimated from county-level information on LIHs' annual income in 2014-2018 and 2020 state-level energy consumption data by using the following equation: where ECB 2020 ij was 2020 ECB for the th county in the th month for LIHs, Exp 2020 ijk was the sum of energy expenditure for the th source (electricity, fuel, and natural gas) of the th county in the th month in 2020, and Income 2014−2018 was the average income of th county in the th month during 2014-2018 for LIHs. Exp 2020 ijk was further calculated by using energy expenditure for the th source of the th county during 2014-2018 (Exp 2014−2018 ik ), energy expenditure fluctuation ratio in th month for kth source compared with the monthly average ( Exp 2014−2018 ijk Exp 2014−2018 ik ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ⁄ ), and 2020 state-level residential energy consumption ratio for kth source of ith county in jth month compared with 2014-2018 data ( Con 2020 ijk Con 2014−2018 ijk ⁄ ). The applied monthly energy consumption indices for electricity, fuel, and natural gas in each state were sales of electricity to residential sector, prime supplier sales volume (propane), and nature gas consumption by residential sector. ). The missing values of LIHs' ECB data were replaced with the estimations from the multiple imputation technique (Wang and Johnson, 2018; van Ginkel et al., 2020) , which includes county-level CCVI data as variables due to its completeness for all counties and socioeconomic relationship with ECB.

This study did not consider racial and demographic differences in economic stratification (i.e., HIHs, MIHs and LIHs) to highlight how financial resources themselves have the potential to drastically stratify the risks to households, even before further etiological differentiation (thereby also avoiding the potential circular logic that racial differences in health outcomes may be due to poorer socioeconomic conditions). Based on this breakdown, the authors estimated how much of the population's contact rates within and across households were likely to be affected by social distancing policies, including stay-at-home orders. We did consider the age-based demography of the city as part of estimating the percent likely to be employed (and thus affected by shut-downs), but did not include age-based probabilities of infection or death, to highlight again how household economics alone can impact economicepidemiological dynamics. We further included the likely proportion of a city's workforce classified as essential workers, who would be exempted or prohibited from adopting social distancing policies. We also assumed over-representation of essential workforce in LIHs and MIHs and used COVID-19-inspired rates for etiology of infection (assuming population-level mass averages without differentiating by age, race, or gender).

Based on household income level, we explicitly considered what proportion of household income would be expended on utilities as a proxy for inelastic costs that directly support household health and safety. These differences naturally lead to differences between households in exposure risks, but also critically influence the likely duration of illness (and concomitant risks of death) experienced by individuals who catch COVID-19, assuming that limited economic resources compromise adequate access to critical healthcare before and during active infection. We further consider the skewed ability of these households to perform their jobs while safely practicing social distancing (e.g., working from home, limiting contact with the public, etc.). To be most conservative, we assumed no direct job loss due to either public health policies (such as lockdowns) or from illness-related absenteeism; economic losses due to illness are felt only as temporary losses in income during protective protocols or illness. This means that all the differences in our study come only from the tradeoff in exposure and healthcare -relaxing this assumption would meaningfully increase the poverty borne by lower income families in negative economic and health outcomes.

1. Epidemiological Model We employed a Susceptible-Exposed-Asymptomatic-Infectious-Recovered-Dead (SEAIRD) model (Cohen, 1992; Liu et al., 2021) with socioeconomically dependent proportions of the population able to effectively shelter at home and/or afford other health-supportive resources. The former impacts the rates of COVID-19 exposure, where interactions with others decreases as income status increases, and the latter decreases the duration of infection while increasing the probability of recovery relative to death. We first defined the population of Susceptible individuals in each socioeconomic status class, . Similarly, we defined Exposed individuals in each status class, , as those who have been infected but were neither symptomatic nor capable of transmitting infection to others. To reflect the possibility that individuals may have never become infectious, we allowed individuals to progress directly from the Exposed class into the Recovered class, . Alternatively, individuals may have progressed from Exposed to the first phase of Infectiousness, , where individuals could transmit the infection but were not yet symptomatic. Individuals, then, could either recover or progress to the second phase of infectiousness, , in which they were both infectious and symptomatic. For simplicity, we assumed that both asymptomatic and symptomatic infectious individuals were equally likely to infect a susceptible individual. Individuals in could either recover or progress to diseaserelated death, .

To capture the dynamics of this system, we also defined the composite value, , , which captured the probability of successful infection transmission due to contact between Infectious individuals of status i and Susceptible individuals of status j. We separately defined , , which captured the probability of contact between an individual in status i and an individual of status j in the absence of social change in response to COVID-19. We defined , ̅̅̅̅ to denote the probability of contact when both individuals i and j were practicing social change in response to COVID-19. Note that status was assumed to affect the possibility of social distancing, such that LIHs were less able to effectively social distance. Further, we assumed that as socioeconomic status increases, the percent impact of social change in response to COVID-19 also increases (i.e., the , ̅̅̅̅ decrease), reflecting the proportion of "essential workers" required to report to work despite the desire to socially distance.

This study also defined the rate of becoming infectious, , and the rates of progression from to as 1 and from to as 2 , each of which is assumed to be status-independent. We defined the rates of recovery from , , and classes as 0, through 2, , respectively, which were dependent on status as a proxy for both underlying health and access to healthcare, as this access critically depends on economic resources, which may be depleted by expenditure on household access to utilities.

Lastly, we defined to capture the decreased rate of recovery from both and , respectively, due to compromised access to health-related resources (up to and including the luxury of convalescence when ill) in the absence of social change in response to COVID-19. To further incorporate the resource cost poverty incurred by social change in response to COVID-19, we defined ̅̅̅̅̅̅̅̅̅ to reflect alleviation of limitations in resources, such that ̅̅̅̅̅̅̅̅̅ > , reflecting the intervention of a policy to ensure ongoing access to critical health-supportive resources, such as energy. As socioeconomic status increases, the impact of social change in response to COVID-19 costs decreases (i.e., increases) to reflect the increased economic capacity to handle incurred costs (whether due to lost salary from furloughs, hiring help to perform disease-exposure risky tasks, or other poverty . Again, to highlight the processes we wish to consider most clearly, we assumed no births, deaths from any cause other than the disease, or movement into or out of the population.

Based on these definitions, we have defined the baseline dynamics of the model, in the absence of social change in response to COVID-19, in the following way: We then modify this baseline model by the use of the appropriate combinations of , , , ̅̅̅̅, , and ̅̅̅̅̅̅̅̅̅ to consider our four scenarios: the baseline scenario of "Do Nothing", where we used , and unaltered; the "Social Distance Only" scenario, where we used , ̅̅̅̅ and ; the "Economic Support" scenario, where we used , and ̅̅̅̅̅̅̅̅̅; and the "Try Everything" scenario, in which we use , ̅̅̅̅ and ̅̅̅̅̅̅̅̅̅. These values indicate assumed percentage-based corrections for cross-socioeconomic interaction rates under unaltered societal function (estimated curve fit to previous, non-COVID outbreaks). To calculate , ̅̅̅̅, we used assumed estimates of the percentage of the households in each socioeconomic category that had at least one worker employed in a job that would have been classified as essential, , such that = {0.7,0.3,0.1} . We then calculated , ̅̅̅̅ = , . To tailor each of these calculations to each specific county, we used the weighted average of , and , ̅̅̅̅ , reflecting the overall percentage of that counties' essential labor force reported, scaled by the percentage of the population reported to be between the ages of 18 and 65 (reflecting the assumed demographic description of most of the workforce itself). These values were initially based on estimates of transmission from Weitz et. al (Weitz et al., 2020) , and were assumed to increase the probability of transmissible infection as household income decreased, reflecting increased probability of interaction due to decreased access to indoor leisure spaces both within and outside of the home, and the decreased probability of having avoidable public interaction because of employment.

Further etiological parameters were tailored based on data for each county, including the local case fatality rate (reflecting local differences in healthcare capacity and baseline health of the population), and mixing rates governing all the , terms, reflecting differences in average contact rates between individuals in different counties (due to patterns in travel, urban planning, etc. Values for the reduction in average recovery rate, :

Low-income 0.6 Medium-income 0.9 High-income 1

From which ̅ was calculated as ̅ = ( − ) + (1 − ), and where c indicated a decreased rate of recovery due to compromised ability to rest and undertake healthcarerelated activities while continuing to work (assumed to be 0.2).

Values for duration of infection based on underlying health condition, :

Low-income 0.06 Medium-income 0.07 High-income 0.09

These values were estimated using medium-income households as the presumed average, representing a fourteen-day duration of infectivity until immune protection. While COVID-19 is now known to be transmissible mostly within a ten-day window, we could assume this for all future potential pandemics, and chose fourteen days since that was the initial window considered for public health response policy estimates. Alterations for duration of infection based on household income were estimated as scaling from medium-income households based on known all-cause health corrections. Values for 1, were then calculated as 1, =

US States with the Most Essential Workers

Coronavirus: Domestic electricity use up during day as nation works from home

Are you in the American middle class? Find out with our income calculator

Household energy insecurity: dimensions and consequences for women, infants and children in low-and middle-income countries

(eds) Giddens' theory of structuration: A critical appreciation

Energy access is needed to maintain health during pandemics

Agency for Toxic Substances and Disease Registry

Mobility network models of COVID-19 explain inequities and inform reopening'

Social connections with COVID-19-affected areas increase compliance with mobility restrictions

Coronavirus comes home? Energy use, home energy management, and the social-psychological factors of COVID-19

Infectious Diseases of Humans: Dynamics and Control

COVID-19 United States Cases by County (2021) Johns Hopkins University. Available at

Report: Low-Income Households, Communities of Color Face High 'Energy Burden' Entering Recession

The Constitution of Society: Outline of the Theory of Structuration

Rebutting Existing Misconceptions About Multiple Imputation as a Method for Handling Missing Data

COVID-19 assistance needs to target energy insecurity

Keeping warm and staying well: findings from the qualitative arm of the Warm Homes Project

Understanding "energy insecurity" and why it matters to health

Exploring the Housing and Household Energy Pathways to Stress: A Mixed Methods Study

Time-series clustering for home dwell time during COVID-19: What can we learn from it?', medRxiv

Energy, Poverty, and Health in Climate Change: A Comprehensive Review of an Emerging Literature', Frontiers in Public Health

Differential COVID-19 case positivity in New York City neighborhoods: Socioeconomic factors and mobility', Influenza and other Respiratory Viruses

Fuel poverty and human health: A review of recent evidence

A new SEAIRD pandemic prediction model with clinical and epidemiological data analysis on COVID-19 outbreak', Applied Intelligence

Low-Income Energy Affordability Data (LEAD) Tool. Golden, CO. Available at: energy.gov/eere/slsc/maps/lead-tool

Exploring the link between energy security and subjective wellbeing: A study of 22 nations

New Data Suggest COVID-19 Is Shifting the Burden of Energy Costs to Households, Columbia University

Sociodemographic disparities in energy insecurity among low-income households before and during the COVID-19 pandemic'

Calculating Percent Positivity

Justice, fuel poverty and disabled people in England

Energy justice: Conceptual insights and practical applications', Applied Energy

Energy Information Administration (2020b) Stay-at-home orders led to less commerical and industrial electricity use in April

US Coronavirus Cases & Deaths by State (2021) USAFACTS. Available at

An Examination of Discrepancies in Multiple Imputation Procedures Between SAS ® and SPSS ® An Examination of Discrepancies in Multiple Imputation Procedures Between SAS and SPSS

Modeling shield immunity to reduce COVID-19 epidemic spread

Giddens, structuration theory and strategy as practice', Cambridge Handbook of Strategy as Practice

Energy efficiency and energy justice for U.S. low-income households: An analysis of multifaceted challenges and potential

No supplemental information.