Stability and Volatility in the Contextual Predictors of Working-Age Mortality in the United States

Abstract

The contextual predictors of mortality in the United States are well documented, but the COVID-19 pandemic may have upended those associations. Informed by the social history of disease framework (SHDF), this study examined how the importance of county contexts on adult deaths from all causes, drug poisonings, and COVID-19-related causes fluctuated during the pandemic. Using 2018 to 2021 vital statistics data, for each quarter, we estimated associations between county-level deaths among adults ages 25 to 64 and prepandemic county-level contexts (economic conditions, racial-ethnic composition, population health profile, and physician supply). The pandemic significantly elevated the importance of county contexts—particularly median household income and counties’ preexisting health profile—on all-cause and drug poisoning deaths. The elevated importance of household income may be long-lasting. Contextual inequalities in COVID-19-related deaths rose and then fell, as the SHDF predicts, but rose again along with socio-political disruptions. The findings support and extend the SHDF.

There are wide and long-standing inequalities in mortality between counties in the United States (Dwyer-Lindgren et al. 2016, 2017). Counties characterized by higher levels of education and income, lower prevalence of harmful behaviors such as smoking, and lower shares of marginalized populations tend to have lower mortality rates (Dobis et al. 2020; Graetz and Elo 2022; Shaw et al. 2016). Such inequalities were also evident during the COVID-19 pandemic, with economically disadvantaged counties and those with large shares of marginalized and health-vulnerable populations experiencing high COVID-19 death rates (Itzhak et al. 2022; Jones et al. 2023). Although geographic inequalities in deaths from COVID-19 and other causes during the pandemic have been documented, several key questions have gone unexplored. Specifically, it is unclear whether and how these inequalities fluctuated during the pandemic, whether any such fluctuation was typical prior to the pandemic, and what the fluctuation suggests about the predictors of county-level mortality now and in the near future. Informed by the social history of disease framework (SHDF; Clouston et al. 2016), this study provides insights into these questions for all-cause mortality and two causes of death of upmost importance during the pandemic: drug poisoning and COVID-19.

Background

Trends in U.S. life expectancy have been worrisome for decades, but declines in life expectancy during the COVID-19 pandemic were drastic. Between 2019 and 2021, life expectancy declined from 78.8 years to 76.1 years, a drop of 2.7 years (Arias et al. 2022; Arias and Xu 2022). Most of the decline was due to COVID-19 and unintentional injuries, predominantly drug poisonings. Among the two causes of death contributing most to the decline in life expectancy between 2019 and 2020, COVID-19 contributed 61.2%, and unintentional injuries contributed 11.7% (Arias and Xu 2022). Between 2020 and 2021, those percentages were 50.0% and 15.9%, respectively (Arias et al. 2022). Drug poisoning mortality rates were high before the pandemic, particularly among working-age adults, but surged during 2020 and 2021, claiming nearly 200,000 lives (National Institute on Drug Abuse 2023).

Prior to the pandemic, all-cause mortality rates varied considerably among U.S. counties. Counties with higher mortality rates had higher prevalence of behavioral and metabolic risk factors (e.g., smoking, obesity), poor physical and mental health, lower educational attainment and income, higher unemployment rates, and larger shares of some minoritized racial-ethnic groups (Dobis et al. 2020; Dwyer-Lindgren et al. 2017; Graetz and Elo 2022; McLaughlin and Stokes 2002). Likewise, prior to the pandemic, drug poisoning mortality rates were higher in counties with smaller shares of college-educated residents and higher poverty and unemployment rates (Cano et al. 2023; Monnat 2018, 2019; Monnat et al. 2019) and in communities where large shares of residents reported poor mental health (Kedia et al. 2020). Unlike with many other causes of death, prepandemic drug poisoning mortality was lower, on average, in counties with smaller shares of Black and Hispanic residents (Zhu et al. 2022), although the racial distribution of drug poisoning deaths may have shifted during the pandemic (Kariisa et al. 2022; Monnat 2023).

Going into the pandemic, these place-level “preexisting conditions” meant that certain counties were more vulnerable than others to experiencing high COVID-19 mortality rates as well as high rates of drug poisoning and all-cause mortality. In terms of COVID-19 mortality, significantly higher mortality rates were observed in counties with smaller shares of college-educated adults, lower median household income, higher poverty and unemployment rates, higher prevalence of preexisting health conditions, and larger shares of Black and Hispanic populations in the early waves of the pandemic but larger White population shares in later waves (Albrecht 2022; Cheng, Sun, and Monnat 2020; Jones et al. 2023; Madlock-Brown et al. 2022). Using a machine-learning approach to evaluate 53 county-level contexts, one analysis found that counties’ racial-ethnic composition, education levels, median household income, and poverty rates were the strongest predictors of COVID-19 mortality rates (Itzhak et al. 2022). Regarding all-cause mortality, Stokes et al. (2021) found that rates of excess deaths from January 2020 to March 2021 were higher in counties with lower median income and education and with worse health profiles.

In general, such ecological studies examined a snapshot of time and provided few insights into whether the associations between county contexts and mortality evolved during the pandemic. A few studies examined how county predictors of COVID-19 deaths fluctuated during the early months of the pandemic. One tracked COVID-19 deaths for six months following each county’s first experience with COVID-19, finding that the county contexts that mattered varied based on how long COVID-19 was present in a county (Madlock-Brown et al. 2022). A study of January through May 2020 found that the association between counties’ socioeconomic status (SES) and COVID-19 mortality flipped from positive to negative (Clouston, Natale, and Link 2021). A study of April through November 2020 found that the importance of county education, income, and racial-ethnic composition for predicting COVID-19 mortality fluctuated over time (Itzhak et al. 2022). During the early months of the pandemic, counties with larger shares of Black residents had higher COVID-19 mortality than did counties with larger shares of White residents, but this disparity disappeared in early 2021 and reversed in late 2021 (Bergmann, Ahlgren, and Stone 2022). Between the initial and Omicron waves, disparities in COVID-19 mortality between non-Hispanic White and Black adults shrank due to rising mortality among the former, declining mortality among the latter, and a growing mortality disadvantage in nonmetropolitan areas (Lundberg et al. 2023).

These temporal analyses raise intriguing questions, which this study aims to answer. For instance, how did the importance of county contexts for county mortality fluctuate for not only COVID-19 mortality but also for drug poisoning mortality—a primary contributor to the decline in life expectancy during 2020 and 2021—and for all-cause mortality? The temporal analyses also raise questions about how the fluctuation in county-level predictors of drug poisoning and all-cause mortality during the pandemic compared to that before the pandemic. In other words, does the importance of county contexts for mortality typically fluctuate over time, or was the fluctuation during the pandemic atypical? Were certain county characteristics, such as socioeconomic and racial-ethnic composition and counties’ population health profiles, more important for predicting mortality during the pandemic than before the pandemic, as is sometimes implied in scholarly work and media coverage? There is reason to question this assumption. For instance, using European data, Nielsen et al. (2021) concluded that the hyped sex differences in excess deaths during the pandemic were not unusual once put in temporal context.

Also unclear is how the importance of county contexts compares to that of states’ pandemic policies for predicting county deaths. Many states enacted containment policies, such as stay-at-home orders, to mitigate the spread of the virus, and economic support policies, such as rent moratoria, to counter the economic fallout. These policies affected health and mortality in complex ways (e.g., Boen et al. n.d.; Kaslow et al. 2023; Monnat et al. 2023; Wolf et al. 2024). For instance, containment policies like mask mandates reduced COVID-19 mortality (Hansen and Mano 2023), whereas containment policies like business closures increased mental health problems (Monnat et al. 2023) and drug poisoning deaths (Wolf et al. 2024). States’ bans on foreclosures and evictions reduced drug poisoning deaths (Wolf et al. 2024) and buffered the impact of economic precarity on mental health (Boen et al. n.d.) and COVID-19 death rates (Sun and Bisesti 2023).

Social History of Disease Framework

Our study is informed by the SHDF, which puts cause-specific mortality in temporal context to better understand how and why social inequalities in it change over time (Clouston et al. 2016). It hypothesizes four stages through which cause-specific mortality progresses in a population. Each stage has implications for social inequalities in diseases and their mortality risks. The first stage, “natural mortality,” captures the period in which there is little knowledge in the population about how to prevent or treat a disease and its mortality risk. Social inequalities in mortality may or may not exist in this stage. As knowledge emerges, it is unequally diffused across the population, ushering in the second stage, “producing inequalities.” Inequalities emerge because some individuals and communities have social, economic, and other structural advantages that facilitate access to lifesaving knowledge and technology. The third stage, “reducing inequalities,” occurs when knowledge about and technology for preventing and treating a disease and its mortality risks diffuse throughout the population. Success in the third stage can lead to the final stage, “reduced mortality and/or disease elimination.” Importantly, the framework also asserts that human interventions—for example, technology, medicine, or policies—can affect the duration of each stage and the magnitude and direction of the inequalities within each stage.

The SHDF is closely related to another framework for understanding health inequalities—fundamental cause theory (FCT). In its original conceptualization, FCT helps explain why the association between SES and health emerges and persists over time despite changes in epidemiologic environments and the mechanisms that connect SES to health (Link and Phelan 1995; Phelan, Link, and Tehranifar 2010). It asserts that individuals and communities with higher SES deploy their SES-related resources (e.g., money, power, knowledge, social connections) to obtain protective factors and avoid harmful factors regardless of the environment or mechanisms at play (Clouston et al. 2016). FCT has been extended to understand other how other social disadvantages, including racism (Phelan and Link 2015) and stigma (Hatzenbuehler, Phelan, and Link 2013), affect health. The SHDF further asserts that the importance of social (dis)advantages for cause-specific mortality must be placed in temporal and spatial context. For example, Clouston et al. (2016) used county-level data from 1969 to 2009 to show how the importance of county SES for predicting mortality from four causes of death differed across stages of the framework.

We propose that the SHDF has utility for conceptualizing how social inequalities in mortality evolved during the pandemic not just for COVID-19 deaths but also for drug poisoning and all-cause mortality. Based on the original application of the framework that analyzed county SES and cause-specific mortality (Clouston et al. 2016), we posit that the importance of county SES for cause-specific mortality may have changed during the exogenous shock of the pandemic. Studies of COVID-19 deaths early in the pandemic support this speculation (Clouston et al. 2021; Kamis et al. 2021). However, the framework’s potential for understanding social inequalities in mortality during the pandemic extends beyond place-level SES to other place-level markers of (dis)advantage, and it extends beyond COVID-19 mortality to drug poisoning and all-cause mortality. In the following, we describe these extensions for the current study.

Extending the Framework to the Current Study

To develop a fuller picture of how social inequalities in mortality fluctuated during the pandemic, we extend the framework in two ways. First, we examine all-cause mortality in addition to two causes of death of upmost importance during the pandemic, drug poisoning and COVID-19. All-cause mortality provides a complete picture, which may be distinct from that of any specific cause of death. For instance, place-level social inequalities in mortality from some causes may have widened but narrowed, persisted, or reversed for others (e.g., Miech et al. 2011), potentially leaving social inequalities in all-cause mortality unchanged. It is imperative to understand the net effect of these disparate trends and their implications for current and future mortality inequalities.

The second way that we propose extending the framework is to expand the scope of county-level (dis)advantages. In addition to SES and SES-related resources, we consider counties’ racial-ethnic composition and population health profiles. We posit that in the face of a shock like the pandemic, the importance of these factors for all-cause mortality, COVID-19 mortality, and drug poisoning may also change in systematic ways. Next, we describe some of these potential changes.

The association between county SES and mortality may have fluctuated for several reasons. Higher educated counties and those with higher median income may fare better on COVID-19 mortality if their residents can isolate by working from home and are able to avoid crowded stores by ordering food delivery (Reeves and Rothwell 2020). Residents of counties with high-quality internet may have more opportunities for telehealth to diagnose and treat physical and mental health conditions (Pandit et al. 2023). The comfort, resources, and control that higher income and education provide may also mean lower stress and anxiety—psychological hardships that increase risks of multiple causes of death. As an SES-related resource, health care infrastructure is also an important county-level resource. It may be more important in the face of a pandemic as demand for services increases.

In terms of county racial-ethnic composition, structural racism constrains access to flexible resources, like medical knowledge, power, and money, that are critically important during a pandemic (Garcia et al. 2021; Laster Pirtle 2020; Phelan and Link 2015). In line with FCT’s argument that structural racism is a fundamental cause of health disparities, the importance of county racial-ethnic composition for mortality may have increased during the pandemic in part because Black and Hispanic individuals were more likely than White individuals to be working in essential industries and live in crowded housing (Almagro et al. 2021; DeLuca, Papageorge, and Kalish 2020), conditions that increase disease transmission.

A county’s population health profile is also a marker of (dis)advantage that affects access to resources and the diffusion of information and support. Adverse health behaviors, metabolic vulnerabilities, and poor physical and mental health, which are tied to SES, elevate mortality risk (DeSalvo et al. 2006; Masters et al. 2013; Thun et al. 2013; Yang et al. 2020). Counties with large shares of health-vulnerable populations had a high risk of COVID-19 deaths and consequently, of an overburdened health care system, reducing resources to treat other health conditions. By contrast, counties with favorable health profiles may have fared better if their health, as a community resource, lessened the severity of cases, keeping hospital beds open and physicians available for all health care needs of all residents. The importance of such resources for mortality may fluctuate during a pandemic. For example, the importance of a county’s population health profile may increase when there is no vaccine or effective treatment. The importance may increase when mobility restrictions and an overburdened health care system reduce access to routine health care, which may be most detrimental for counties with high shares of residents with poor mental and physical health. The importance of counties’ mental health may increase because people with poor mental health are less likely to have beneficial social connections (Umberson and Montez 2010) through which helpful knowledge and resources related to the pandemic could be diffused. Indeed, in a study of another type of exogenous shock—Hurricane Katrina—Raker, Zacher, and Lowe (2020) found that the stressors it imposed had the largest adverse effects among people with preexisting psychological distress and fair/poor health. Already high rates of social isolation among individuals with poor mental health likely increased during the pandemic due to social distancing protocols, elevating the risk for multiple causes of death. In sum, understanding how prepandemic county (dis)advantages predicted mortality during the unique conditions of the pandemic requires expanding the focus beyond SES resources to other place-level markers of (dis)advantage.

Figure 1, Panel A illustrates how the importance of county contexts for county COVID-19 mortality rates may fluctuate over time according to the SHDF. For heuristic purposes, we use county SES as the measure of (dis)advantage. The figure illustrates that prior to the pandemic, the risk of dying from COVID-19 was effectively zero (or undefined) in both advantaged and disadvantaged counties. At the start of the pandemic, COVID-19 mortality rises for both types of counties because there is little knowledge about disease spread and containment strategies have not yet been enacted (Stage 1). As knowledge about the disease and how to reduce its spread emerges (e.g., physical distancing), advantaged counties are better equipped to respond. The residents of these counties are more likely to be able to work remotely, live in uncrowded housing, have food and other supplies delivered to their homes, and more. As a result, county contexts become important predictors of COVID-19 mortality, and inequalities between advantaged and disadvantaged counties increase (Stage 2). As knowledge diffuses, vaccines are developed and widely disseminated, and pandemic mitigation policies are enacted, COVID-19 mortality declines for disadvantaged counties, and inequalities diminish (Stage 3). Eventually, COVID-19 inequalities shrink, and mortality approaches zero for both types of counties (Stage 4). Whether social inequalities in COVID-19-related deaths across counties followed these hypothesized stages is unclear, although evidence from the early months of the pandemic provides some support (Clouston et al. 2021; Kamis et al. 2021). Also unclear is whether other markers of county-level (dis)advantages discussed previously followed similar patterns.

Figure 1.

Hypothesized Changes in the Importance of Counties’ Socioeconomic Contexts on County Mortality Rates across the Four Stages of the Social History of Disease Framework.

Source: Figure adapted from Clouston et al. (2016).

Figure 1, Panel B shows a hypothetical example of how the importance of county contexts for drug poisoning deaths may have fluctuated. In this example, disadvantaged counties had consistently higher drug poisoning mortality than did advantaged counties before the pandemic (Monnat 2018; Monnat et al. 2019). At the start of the pandemic, the figure shows drug poisoning mortality rising in both types of counties. During this time, many states imposed stay-at-home orders and closed schools and nonessential businesses, and many health care providers (including mental health and substance use treatment providers) cancelled appointments or moved to telehealth (Johnson et al. 2022). These institutional responses to the pandemic created novel and unrecognized threats across both advantaged and disadvantaged counties. States’ pandemic containment measures such as stay-at-home orders increased drug poisoning mortality (Wolf et al. 2024) likely through social isolation (Tull et al. 2020), disruptions to mental health and substance use treatment (McNeely et al. 2021; Mellis, Potenza, and Hulsey 2021), job loss (Gupta et al. 2023), increases in solitary drug use (Schneider et al. 2023), and more. Eventually, in this hypothetical example, drug poisoning mortality in advantaged counties starts declining toward its prepandemic level, thereby widening county inequalities. In this stage, compared to those in disadvantaged counties, people in advantaged counties may have more access to and resources to spend on social and recreational activities that boost mental health and social connections (Astell-Burt and Feng 2021; Larson et al. 2021; Taff et al. 2021), draw on their savings to protect against the mental strain of job loss (Despard et al. 2018), and leverage local social connections that may be more available in counties with good mental health (Yang et al. 2019). In this stage, the health care infrastructure may be less burdened in counties with advantaged population health profiles (Levin et al. 2023), potentially meaning there are more resources available for residents struggling with mental health and substance use problems and more hospital and emergency medical provider capacity to revive individuals who overdose. Inequalities start to wane in the third stage of the framework, as drug poisoning mortality starts to decline in disadvantaged counties toward the level of advantaged counties. The decline may reflect diffusion of information about accessing telehealth for substance use treatment (Jones et al. 2022), the widespread distribution of government economic supports (Wolf et al. 2024), the repeal of stay-at-home orders, bans on evictions and foreclosures to protect economically distressed households, and more. In the final stage, drug poisoning mortality rates approach similarly low levels in both types of counties.

Aims

This study aims to better understand the contextual predictors of working-age mortality during the first two years of the COVID-19 pandemic in the United States and determine how they compare to the prepandemic period. The study uses county-level data to assess (1) how the importance of prepandemic county contexts for predicting deaths from all causes, drug poisoning, and COVID-19-related causes among working-age adults fluctuated during the pandemic; (2) how the importance of county contexts prior to and during the pandemic differed; and (3) their importance compared to states’ pandemic policies. We examine four domains of prepandemic county contexts: socioeconomic conditions, racial-ethnic composition, population health profile, and physician supply. The selection of these contexts was informed by the studies highlighted previously and the county health rankings and roadmap conceptual model for understanding all-cause mortality (Remington, Catlin, and Gennuso 2015). We focus on counties because they are the smallest geographic unit for which mortality rates are available for the entire United States.

We focus on working-age adults (ages 25–64), whose mortality has been rising in recent decades (National Academies of Sciences, Engineering, and Medicine [NASEM] 2021). We focus on them because (1) higher U.S. death rates vis-à-vis other high-income countries in 2020 and 2021 were most pronounced for adults under 65 years (Bor et al. 2023), (2) U.S. life expectancy declines in 2020 and 2021 disproportionately involved COVID-19 deaths in midlife (Masters, Aron, and Woolf 2024), and (3) although older adults had the highest COVID-19 mortality rates, working-age adults experienced the largest percentage increase in all-cause mortality over our study period. ¹ Moreover, apart from COVID-19, unintentional injuries, driven by drug poisonings, was the cause of death that increased most during the pandemic (Arias and Xu 2022), and drug poisoning mortality is highest among working-age adults (Monnat 2023).

We extend prior work on how social inequalities in death fluctuated during the pandemic in several ways. First, rather than analyzing only COVID-19 deaths, we also consider drug poisoning and all-cause mortality because the pandemic’s impact was much broader than any single cause of death. Second, we include the prepandemic period, starting in 2018, to provide perspective on how much fluctuation was typical in the years preceding the pandemic. In addition, we account for states’ pandemic policies. The findings provide novel insights into how the exogenous shock of the pandemic disrupted the importance of contextual predictors of mortality.

Data and Methods

County Death Data

County death counts were from the restricted Multiple Cause of Death files, accessed through a data use agreement with the National Center for Health Statistics (NCHS; 2023). We calculated death counts for each county in each quarter, starting with the first quarter (Q1) of 2018 and ending with fourth quarter (Q4) 2021, the last quarter of data in the restricted files. For each county-quarter, we calculated sex-specific deaths among adults ages 25 to 64. We examined deaths from all causes and from two specific causes. The two causes and their ICD-10 codes included drug poisoning (X40–X44, X60–X64, X85, Y10–Y14) and deaths with COVID-19 as an underlying or contributing cause (U07.1). Including deaths that involved COVID as either underlying or contributing helped capture most COVID-19-related deaths and reduced the impact of misclassification of COVID-19 sequalae.

County Contextual Variables

We obtained 10 variables from the County Health Rankings and Roadmap (CHRR; 2022) data sets. We included three socioeconomic variables. In the CHRR, education was the percentage of adults ages 25 to 44 with at least some postsecondary education. Unemployment was the percentage of individuals ages 16 and older who were unemployed. Income was median household income. To capture the racial-ethnic composition of a county, we included the percentages of the population that were non-Hispanic Black (NHB) and Hispanic. We used four markers of population health. Smoking was the percentage of adults who were current smokers. Obesity was the percentage of adults with a body mass index ≥ 30. Self-rated health was the percentage of adults reporting fair or poor health. Mental health was the average number of self-reported mentally unhealthy days in past 30. Lastly, physicians was the ratio of population size to the number of primary care physicians.

All variables were measured circa 2018, the start of the study. We used 2018 because some variables in the CHRR are not harmonized across the 2018 to 2021 data sets, and we wanted to ensure that our findings were not contaminated by changes in survey sampling or measurement. The 2020 CHRR provides unemployment, income, and race-ethnicity for 2018 and education for a five-year period spanning 2014 to 2018. The 2021 CHRR provided smoking, self-rated health, mental health, and physicians for 2018 and obesity for 2017 (2018 data for obesity were not available in any CHHR). The CHRR variables and original sources are listed in Appendix Table A1 in the online version of the article. The CHRR had complete data for 2,991 of the 3,141 counties in the 50 states. Our analyses included the contiguous 48 states, excluding Connecticut, for a final count of 2,953 counties, representing 98% of the U.S. population. Connecticut switched from county-level reporting to “planning area” reporting in 2022, which complicates interpolating 2021 county population counts.

County Control Variables

We accounted for rural–urban context using the nine category USDA Economic Research Service’s 2013 rural–urban continuum codes (RUCCs) based on the 2010 census population. In the initial wave of the pandemic, COVID-19 mortality rates were highest in large metropolitan areas, but in later waves, they were highest in nonmetropolitan areas (Lundberg et al. 2023; Paglino et al. 2023). We adjusted for county age distribution in each quarter with three variables measuring the percentage of adults ages 25 to 64 who were 25 to 34, 35 to 44, and 45 to 54 (Wolf et al. 2024; Wolf, Montez, and Monnat 2022). The regression models, described in the following, also included the size of the population ages 25 to 64 in each county as an “offset” to account for the fact that counties with more population tend to have more deaths. County population counts came from the U.S. Census Bureau (2021, 2023), which provided counts for the midpoint of each calendar year. We used linear interpolation to obtain quarterly population estimates for each county by sex.

States’ COVID-19 Mitigation Policies

The Oxford COVID-19 Government Response Tracker contained indices summarizing the daily intensity of policy responses to the pandemic (Hale et al. 2021). We used the economic support index (ESI), which included policies, such as direct cash payments and eviction moratoria, that were intended to offset the economic impacts of the pandemic. We also used the containment and health index (CHI), which included measures intended to reduce the spread of the virus, such as school closings and masking mandates. ² Each index was additive, giving equal weight to each policy. The indices were highly correlated with indices derived from a latent variable approach (Hale et al. 2021). Our analysis used the quarterly average of each index for each state.

Analytic Strategy

We merged all data into one county-quarter data set. Some variables were time invariant (county contextual variables and RUCC), and some varied by quarter (deaths, population, age distribution, and state policy indices). We converted all predictor variables to Z scores (M = 0, SD = 1) to facilitate coefficient comparisons across time.

We then estimated the number of deaths, d, in county c in quarter q using the following negative binomial regression specification with dispersion parameter r and an offset (θ_c,q) to account for the size of each county’s working-age population in each quarter.

$$d_{c,q}~NegativeBinomial({\mu }_{c,q}\times {\theta }_{c,q},r)$$

$$\begin{array}{l}\ln ({\mu }_{c,q})=b_0+b_1{E}_c+b_2{R}_c+b_3{H}_c\\ +b_4{P}_c+b_5{S}_q+b_6{C}_{c,q}.\end{array}$$ (1)

It modeled the natural log of the number of deaths in a county in a quarter (µ_c,q) from the four types of county contexts (socioeconomic, E; race-ethnic composition, R; health, H; and physicians, P), states’ pandemic policy indices, S, starting in second quarter (Q2) 2020, and county controls, C (age distribution, RUCC). Standard errors were clustered at the state level to account for similarities among counties within states. Models were estimated with Stata 18.0.

In preliminary analyses, we assessed collinearity among the 10 county variables of interest (results are in Appendix Tables A2–A5 in the online version of the article). We (1) examined all pairwise correlations, (2) assessed variance inflation factors, and (3) compared the size and direction of the estimated coefficient for each variable from a model including only that variable to a model with all 10 variables. Analyses indicated that including both self-rated health and smoking was potentially problematic such that smoking should be excluded from the all-cause and COVID-19 models and self-rated health should be excluded from the drug poisoning model. We estimated the models accordingly.

To address our second aim, we assessed whether county context was more (or less) important for predicting deaths during the pandemic than before the pandemic. We describe our approach here using county income as an example. To test whether county income became more (or less) important for predicting deaths during a pandemic quarter, say Q2 2020, than it was just prior to the pandemic, we compared the income–mortality association in that quarter to the association in the year 2019. In this example, we first restricted the data set to all quarters of 2019 and Q2 2020. We created a time variable, where time = 0 for 2019 and time = 1 for Q2 2020. To Equation 1, we added interaction terms between time and each county variable. We used the model estimates to compute the average marginal effects (AME) of income in the two time periods and test whether they were different (see Mize 2019). We repeated this approach for each quarter of the pandemic. These analyses used the Spost13 mtable and mlincom commands in Stata 18.0.

Results

Descriptive statistics of county contextual variables are in Table 1. They are based on the original units of measurement. As expected, counties differed on these variables. For instance, the average unemployment rate across counties was 4.1% but ranged from 1.5% to 18.1%.

Table 1.

Descriptive Statistics for County Variables.

	Average	SD	Minimum	Maximum
Some college education or higher (%)	58.11	11.68	20.38	100.00
Unemployment (%)	4.10	1.40	1.54	18.09
Household income ($1,000s)	52.88	13.85	25.39	140.38
Non-Hispanic Black (%)	9.11	14.32	.00	85.41
Hispanic (%)	9.56	13.60	.61	96.36
Current smokers (%)	21.26	4.07	7.08	40.94
Obesity (%)	33.47	5.90	11.00	53.70
Fair or poor self-rated health (%)	20.07	5.03	8.59	41.00
Number of mentally unhealthy days in past 30	4.68	.66	2.69	7.29
Ratio of population to primary care physicians	2,688.30	2,281.14	152.00	2,8523.00

Note: Estimates are based on data from the County Health Rankings and Roadmap (2022). All data are for 2018 except for obesity (from 2017) and education (a five-year period estimate spanning 2014–2018). N = 2,953 counties.

All-Cause Deaths

Models estimating county all-cause deaths among working-age men are in Table 2. Each column contains incident rate ratios (IRRs) by quarter; RUCC and age distribution are not shown. The IRRs are exponentiated coefficients from the regression models. As an example, Column 1 contains IRRs for men’s deaths in Q1 2018. Each dimension of county contexts (economic, racial-ethnic, population health, physicians) was associated with men’s deaths net of each other. For instance, a standard deviation increase in counties’ median household income was associated with 10% (IRR = .90, p < .001) fewer deaths in Q1 2018 net of other variables in the model. Each standard deviation increase in the percentage of adults reporting fair or poor health was associated with 12% more deaths in Q1 2018 (IRR = 1.12, p < .001). County education level and unemployment rates were not significantly associated with men’s all-cause deaths net of other variables.

Table 2.

Incident Rate Ratios for County All-Cause Deaths among Working-Age Men by Calendar Year and Quarter.

	2018				2019				2020				2021
	Q1	Q2	Q3	Q4	Q1	Q2	Q3	Q4	Q1	Q2	Q3	Q4	Q1	Q2	Q3	Q4
Economic
Some college plus	1.00	.99	1.00	1.00	1.00	1.00	1.00	1.01	1.01	.99	1.01	1.00	1.01	1.02	.96	.99
Unemployment	.99	1.00	1.00	1.00	1.00	.99	.98+	1.00	.99	1.02	1.02	1.00	1.01	1.02+	.99	1.03*
Income	.90***	.91***	.91***	0.91***	.90***	.90***	.89***	.90***	.91***	.93**	.90***	.88***	.91***	.89***	.88***	.87***
Race-ethnicity
Black	1.03***	1.03**	1.03**	1.03**	1.03*	1.03**	1.04***	1.03**	1.03**	1.07***	1.06***	1.01	1.05***	1.03**	1.04**	.97*
Hispanic	.97+	.97+	.97*	.96**	.97*	.97+	.98	.97+	.96**	.98	1.00	1.00	1.02	.97	1.00	.99
Population health
Smoking	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—
Obesity	1.03**	1.03***	1.03**	1.03***	1.03**	1.03***	1.03**	1.03**	1.02**	1.00	1.01	1.04***	1.03*	1.02+	1.02	1.05***
Fair/poor SRH	1.12***	1.10**	1.10**	1.12***	1.11***	1.09**	1.10***	1.11**	1.16***	1.15**	1.21***	1.16***	1.16***	1.12***	1.15**	1.04
Mental health	1.08***	1.09***	1.09***	1.08***	1.08***	1.09***	1.08**	1.08**	1.06**	1.06+	1.04	1.04	1.11***	1.10***	1.10*	1.09**
Medical care
Population to physician	.98*	.98*	.98*	.98**	.98***	.99+	.99	.98*	.97***	.98*	.98**	.98*	.97**	.97***	.99	.98*
State policies
Economic support										1.00	.99	.96	1.02	1.01	.89*	.98
Containment										1.05	.94*	.99	.92+	.93*	.96	.97
Pseudo R2	.11	.10	.10	.11	.11	.11	.10	.10	.10	.09	.12	.11	.12	.12	.11	.10

Note: Models exclude Alaska, Hawaii, and Connecticut. N = 2,953 counties in each quarter (Q). County death data from the National Vital Statistics System for 2018 to 2021; county contextual data from County Health Rankings and Roadmap (2022) for 2018. All variables were converted to Z scores. Models adjusted for county age distribution and rural-urban designation. Standard errors were clustered by state. Population to physician = ratio of county population to primary care physicians; SRH = self-rated health; mental health = number of mentally unhealthy days in the past 30 days.

⁺p < .1. *p < .05. **p < .01. ***p < .001.

Other columns in Table 2 contain model results for subsequent quarters. To better discern continuity and change across quarters, we converted the results into AMEs and plotted them in Figure 2. The AMEs can be interpreted as the estimated difference in the number of deaths (for an average sized county) among working-age men for an approximate 1 SD increase in a variable. For instance, net of other variables in the model, in Q1 2018, among working-age men, an approximate standard deviation increase in counties’ median household income was associated with 3.9 fewer deaths, and an approximate standard deviation increase in the percentage of adults reporting fair or poor health was associated with 4.2 more deaths.

Figure 2.

Average Marginal Effects of County Contexts on County All-Cause Deaths among Working-Age Men by Calendar Year and Quarter.

Note: The average marginal effects are calculated from the model estimates reported in Table 2. N = 2,953 counties in each quarter. County death data from the National Vital Statistics System for 2018 to 2021; county contextual data from County Health Rankings and Roadmap (2022) for 2018.

Several patterns are noteworthy. First, during 2018 and 2019, the importance of each county variable for predicting county deaths was remarkably stable. Higher income was consistently associated with lower death rates, whereas obesity, unfavorable self-rated health, more mentally unhealthy days, and higher percentage NHB were consistently associated with higher death rates, with little fluctuation in the AMEs. That stability ended in Q2 2020. From then on, there was striking volatility in the importance of several county variables for predicting mortality. There were visibly large increases in the importance of county income (Figure 2, red line) and self-rated health (Figure 2, green line) during much of 2020 and 2021 and mental health (Figure 2, gray line) in 2021. During most quarters of 2020 and 2021, a 1 SD difference in county self-rated health or income was associated with sizable differences in all-cause mortality. In 2021, a 1 SD difference in county mental health was also associated with sizable differences in all-cause mortality compared to prior years. In just three quarters did states’ pandemic policies predict men’s all-cause deaths net of county variables (in Table 2, the ESI policy index was associated with men’s mortality in third quarter [Q3] 2021, and the CHI policy index was associated with men’s mortality in Q3 2020 and Q2 2021, at α < .05). These findings for all-cause deaths and drug poisoning and COVID-19-related deaths are robust to including or excluding the two state policy indices (figures showing the robustness are in the Appendix in the online version of the article).

Using 2019 as the baseline, we assessed whether the visible increases in the importance of some county variables for predicting men’s deaths during a pandemic quarter versus 2019 were statistically significant. Results of the interaction models described previously show that several increases in the AMEs were indeed significant. The results are summarized in Table 3. Consider county income as an example. During Q1 and Q2 2020, the association between county income and men’s deaths was not significantly different than it was in 2019 (e.g., the AME for income in Q2 2020 minus the AME in 2019 was 1.34, p = .15). Starting in Q3 2020, county income was a significantly stronger predictor of men’s deaths than it was in 2019. Higher median income in Q3 2020 and most subsequent quarters was more protective against all-cause mortality than it was in 2019. Overall, the pattern depicts sizable fluctuations in the importance of county variables for predicting men’s deaths, particularly income, self-rated health, and mental health. Some variables, such as self-rated health, appear to have returned to their prepandemic level of importance by Q4 2021, whereas other variables, such as income, showed little sign of returning to their prepandemic level of importance. No variable became less important than it was before the pandemic. Consistent with other studies (e.g., Bergmann et al. 2022), the association between county percentage NHB and deaths flipped at the end of 2021 such that counties with smaller shares of NHB residents had higher mortality than those with larger shares.

Table 3.

Difference ^a in the Importance of County Variables for Predicting County All-Cause and Drug Poisoning Deaths in Each Quarter of 2020 and 2021 versus the Year 2019.

	Men								Women
	2020				2021				2020				2021
	Q1	Q2	Q3	Q4	Q1	Q2	Q3	Q4	Q1	Q2	Q3	Q4	Q1	Q2	Q3	Q4
All causes
Some college plus	.14	−.97*	.45	.21	.01	.66	−1.95+	−.75	−.09	.20	.52	.45	−.49	.29	−1.47*	−.60
Unemployment	−.13	1.50***	.49	−.08	.59	.75**	−1.52**	1.38*	.00	.59*	.11	.12	.29	.30	−1.01*	.61
Income	.50+	1.34	−.97*	−1.69***	−.47	−1.15***	−3.41***	−3.29***	−.08	.28	−.84**	−1.39***	−.81*	−1.15***	−2.57***	−2.86***
Obesity	−.14	−1.11+	−.32	.78*	.19	−.27	.07	1.41***	.15	−.48	.23	.32	.15	.17	.01	1.04**
Black	.12	1.96***	1.32**	−.53	1.31*	.24	.97	−2.17***	.20	.97**	.51+	−.35	.38	−.14	.84*	−1.44***
Hispanic	−.62*	−.08	.52	.74	1.72**	−.44+	.19	.23	−.09	−.30	.53	.04	.44	−.41	−.77	−.49
Fair/poor SRH	2.00**	2.26	5.42***	3.53***	3.69**	2.04**	4.76*	−1.32	.48	1.78*	3.15***	2.10**	2.18**	1.43**	3.76***	−.98
Mental health	−.45	.14	−1.59*	−.53	1.28+	.81	1.61	1.08	−.26	−.31	−1.16**	−.49	.07	.01	.40	−.00
Population to physician	−.49+	−.35	−.37	−.20	−.61+	−.54*	.10	−.46	−.10	−.17	.32+	.32	.00	−.19	.34	.73***
Drug poisoning
Some college plus	.01	.05	−.01	.02	.26	−.16	−.02	−.18	−.10	.04	−.07	−.06	.07	.04	−.01	.04
Unemployment	.03	.33***	.19+	.26**	.21	.23+	.32+	.17	−.07	−.02	.04	.11+	−.02	−.04	.05	.03
Income	.01	.03	−.14	−.10	.03	−.33+	−.15	−.37*	−.10	−.11+	−.02	−.08	−.05	−.29**	−.24**	−.19**
Smoking	.18	.31	.15	.13	.33	−.09	.01	−.13	.01	.11	−.03	−.09	.09	.03	−.11	.04
Obesity	−.03	−.11	−.18	−.17+	−.09	−.24*	−.07	−.36**	−.03	−.08	−.01	.01	.05	−.09+	−.07	−.10
Black	.02	.05	.13	.01	.26+	.07	−.02	.05	.03	.01	.04	−.03	−.04	.06	.01	−.02
Hispanic	.04	−.21	.09	−.06	.19	−.31	.04	−.31	−.10	−.09+	−.02	−.11	−.00	−.03	−.17	−.16
Mental health	.03	.63**	.42+	.44*	.82**	.94***	.95***	.69**	−.08	.25**	.16+	.13	.29*	.32***	.21+	.17+
Population to physician	−.22	−.22	−.06	−.22+	−.42**	−.34*	−.39*	−.39*	.06	−.03	.20**	.01	−.04	.05	−.01	.06

Note: County death data from the National Vital Statistics System for 2018 to 2021; county contextual data from County Health Rankings and Roadmap (2022) for 2018. Population to physician = ratio of county population to primary care physicians;, SRH = self-rated health; mental health = number of mentally unhealthy days in the past 30 days. N = 2,953 counties in each quarter.

^aDifference = average marginal effect [AME] of the county variable during a pandemic quarter minus the AME of the county variable in 2019. The AME units of measurement are counts of deaths per county.

⁺p < .1. *p < .05. **p < .01. ***p < .001.

We replicated the analyses for all-cause deaths among working-age women. The findings are similar to those for men, so we include the model results in the Appendix (Table A6; Figure A1) in the online version of the article. In Table 3, we show the results of tests for whether the importance of county variables for predicting women’s all-cause deaths during the pandemic were statistically different from that in 2019. The two most striking findings for women are similar to those for men. That is, the significant and sizable increase in the importance of self-rated health for all-cause deaths during the pandemic dissipated by Q4 2021, but the significant and sizable increase in the importance of income continued with little sign of slowing. The observed pattern, for both men and women, aligns most closely with Stage 2, increasing inequalities, of the SHDF.

Drug Poisoning Deaths

The regression models for men (see Appendix Table A7 in the online version of the article) show that unemployment rate and poor mental health were consistently associated with higher drug poisoning mortality and that the physician supply was consistently associated with lower drug poisoning mortality before and during the pandemic. The association between county mental health and drug poisoning deaths was strong both before and during the pandemic. For example, in Q1 2021, a standard deviation increase in the average number of mentally unhealthy days was associated with a 41% increase in men’s drug poisoning deaths net of other variables in the model. A standard deviation difference in states’ CHI policy index during 2020 and Q1 2021 was also associated with a sizable increase in men’s drug poisoning deaths. For instance, in Q1 2021, a standard deviation increase in the CHI was associated with a 62% increase in drug poisoning deaths. Figure 3 shows AMEs for each variable across the study period. The stability of the predictors of drug poisoning mortality during the 2018 to 2019 period ends in early 2020, when the importance of mentally unhealthy days for predicting drug poisoning deaths began to notably increase over time.

Figure 3.

Average Marginal Effects of County Contexts on County Drug Poisoning Deaths among Working-Age Men by Quarter.

Note: The average marginal effects are calculated from the model estimates reported in Appendix Table A7 in the online version of the article. N = 2,953 counties in each quarter. County death data from the National Vital Statistics System for 2018 to 2021; county contextual data from County Health Rankings and Roadmap (2022) for 2018.

We tested whether the importance of the county variables for predicting men’s drug poisoning deaths during the pandemic was significantly different from that in 2019. As shown in Table 3, Q2 2020 marks a shift, when the importance of unemployment and mentally unhealthy days significantly increased compared to 2019. The elevated importance of mentally unhealthy days persisted through Q4 2021, the end of our study period. This pattern aligns most closely with Stage 2 of the SHDF.

The results for women are not as striking as those for men. As shown in Table 3, the importance of mentally unhealthy days significantly increased by Q2 2020 compared to 2019. By the end of the study period, the importance of mentally unhealthy days had started to wane, but the importance of household income became increasingly stronger. The regression results for women and plots of AMEs are in Appendix Table A8 and Figure A2, respectively, in the online version of the article.

COVID-19-Related Deaths

During Q1 2020, some county variables, such as unemployment, income, and percentage NHB, were associated with COVID-19-related mortality (model results in Appendix Table A9 in the online version of the article). For example, a standard deviation increase in county income was associated with a 47% increase in men’s COVID-19-related deaths, reflecting the fact that COVID-19 first hit large affluent cities. Although 47% is a sizable relative increase, the absolute increase was negligible given the small number of these deaths in that quarter. This is evident in Figure 4. In Q1 2020, an approximate 1 SD increase in county income predicted just .1 more COVID-19-related deaths.

Figure 4.

Average Marginal Effects of County Contexts on County COVID-19-Related Deaths among Working-Age Men and Women by Quarter.

Note: The average marginal effects are calculated from the model estimates reported in Appendix Table A9 in the online version of the article. N = 2,953 counties in each quarter. County death data from the National Vital Statistics System for 2018 to 2021; county contextual data from County Health Rankings and Roadmap (2022) for 2018.

Overall, the AMEs in Q1 2020 for women and men in Figure 4 show little importance of county variables for predicting COVID-19-related deaths, consistent with Stage 1 of the SHDF. Their importance grew in Q2 2020, consistent with Stage 2. For the next year, a 1 SD difference in county self-rated health was associated with a large difference in COVID-19-related deaths for men and women net of other variables. Then, between Q1 and Q2 2021, following the widespread dissemination of COVID-19 vaccines (Mathieu et al. 2023), the AMEs of the county variables diminished, consistent with Stage 3 of the framework. By Q2 2021, the AMEs were negligible, similar to Stage 1. However, subsequently, large inequalities resurfaced in Q3 2021, and many remained elevated in Q4 2021 or were still expanding. One of the most notable was county income. After flipping from a positive association in Q1 and Q2 2020 (higher income counties had more COVID-19-related deaths), not only did county income became strongly associated with COVID mortality, but higher income counties now had much lower COVID-19 mortality rates. Also notable are the findings for the state policy variables. States’ CHI policies were statistically significant predictors in Q1 and Q3 of 2021, when the country experienced spikes in COVID-19 deaths. A 1 SD increase in states’ ESI policies was associated with a large decrease in COVID-19-related mortality during Q3 of 2021.

Discussion

This study documented how the COVID-19 pandemic disrupted the importance of county-level predictors of mortality among working-age adults. It expanded knowledge about these predictors of mortality during the pandemic in several ways. Specifically, it simultaneously assessed multiple predictors, compared the importance of each predictor during the pandemic to its importance before the pandemic, and examined drug poisoning and all-cause deaths in addition to COVID-19-related deaths. The findings provide new insights into how the pandemic altered the importance of county predictors of adult mortality. We discuss several noteworthy findings next.

First, the importance of county contexts for predicting county deaths fluctuated during the first two years of the pandemic. The fluctuation was striking for COVID-19-related deaths and all-cause deaths. The importance of county variables for predicting these two outcomes varied markedly from quarter to quarter during the pandemic, more so than did the importance of county variables for predicting drug poisoning deaths. The fluctuation during the pandemic was atypical compared to the 2018 to 2019 period. That is, the importance of county variables for predicting all-cause and drug poisoning deaths was stable during 2018 and 2019, a sharp contrast to the volatility during 2020 and 2021. The volatility showed some signs of settling down by the end of 2021.

Second, by simultaneously examining 10 county variables, we found that some were better predictors of working-age deaths during the pandemic than they were before it. In general, prior to the pandemic, a 1 SD difference in county income or self-rated health predicted relatively large differences in all-cause deaths, but the magnitude of the associations with county income and self-rated health became even stronger during most of 2020 and 2021. Also notable, the association between county mental health and all-cause deaths became stronger during 2021 compared to 2019. Regarding drug poisoning deaths, county mental health stood out during the pandemic as an increasingly strong predictor over time, especially for men. For COVID-19-related deaths, county self-rated health was a strong predictor until the large-scale rollout of COVID-19 vaccinations in early 2021, after which the associations between county income and percentage NHB and COVID-19 deaths grew stronger over time. Collectively, these findings highlight the associations between counties’ SES and population health profiles and multiple types of mortality during the study period. They imply that counties went into the pandemic with a set of preexisting conditions that put some at greater risk of experiencing high rates of not only COVID-19 mortality but also all-cause and drug poisoning mortality.

Third, the associations between county contexts and deaths were robust to states’ pandemic policies: Models with and without policy indices provided similar findings. Also informative, the two policy indices were irregularly associated with all-cause mortality across the study period. Net of county contexts, the indices were significantly (α < .05) associated with men’s and women’s mortality in three quarters each, with fairly small AME magnitudes. This does not imply that such policies were ineffective. It does, however, raise broader questions for further study. For instance, was the effectiveness of states’ pandemic policies constrained by individuals’ and communities’ social, economic, and health conditions? How many lives might be saved in the next public health crisis by proactively investing in population health and economic well-being? Our findings for COVID-19-related deaths are similar to those for all-cause mortality in several ways, at least during 2020. In several quarters of 2021, states’ pandemic policies were strong predictors of COVID-19-related deaths, but the associations dissipated by Q4. Our findings for drug poisoning tell a different story, however. County mental health was a key predictor of drug poisoning deaths during the pandemic, but in many quarters of 2020 and 2021, more stringent containment policies were also key predictors of drug poisoning mortality. Our associational findings are consistent with other studies showing that state policies that restricted in-person interaction increased risk factors for drug use, such as poor mental health, anxiety, and depression (Brooks et al. 2020; Monnat et al. 2023), and that drug poisoning mortality rose more in 2020 in counties located in states with more stringent restrictions on in-person interactions (Wolf et al. 2024).

Our fourth finding relates to the SHDF, intended to explain how social inequalities in cause-specific mortality rise and fall over time. Although our study is not a formal test of the original framework, we used and expanded it to illustrate how social inequalities in mortality fluctuated during the pandemic. We found that social inequalities in COVID-19-related deaths followed the first three stages just as the framework hypothesized. Inequalities were small at the start of 2020 (Stage 1). They subsequently expanded (Stage 2) given that some counties had numerous advantages. For example, some counties had healthier populations who were not only at lower risk of dying from COVID-19 should they contract the disease but who also placed less burden on the health care system. Such counties also had greater capacity to learn about and follow emerging public health guidance, such as physical distancing, thanks to jobs that allowed remote work, uncrowded residences, and the financial means to have necessities delivered. Once an effective vaccine was developed and made widely available in early 2021 (Lopes and Stokes 2021), social inequalities began to shrink (Stage 3) and effectively disappeared in Q2 2021.

But this is where progress stopped. The disease and its mortality risk were not negligible, as predicted in Stage 4, and the process reverted to Stage 2 as inequalities reemerged. They did so despite highly publicized evidence about preventive behaviors and widespread availability of vaccinations. The inequalities may have reappeared simply because the COVID-19 variant during that period was particularly deadly (Tabatabai et al. 2023). Another possibility is that they reappeared because of social, cultural, and political factors, such as escalating politicization of the pandemic and backlash against public health guidelines and vaccine mandates (Gadarian, Goodman, and Pepinsky 2022). Indeed, research shows that COVID-19 mortality rates in later waves of the pandemic were higher in counties with lower vaccination rates and larger shares of votes for Trump in the 2020 presidential election (Jones et al. 2023), and our results demonstrate that state containment policies were no longer important for predicting COVID-19 mortality by Q4 of 2021.

The SHDF makes room for such interruptions. Clouston et al. (2016) speculated that human interventions—for example, knowledge, technology, medicine—and inefficiencies in the diffusion of information can disrupt the progression, with long-term consequences for social inequalities in and overall levels of mortality. The COVID-19 pandemic provides a case study of how politics, collectivities, and individual behaviors and attitudes can alter the hypothesized stages of the SHDF. Our findings also demonstrate the utility of using the SHDF to examine social inequalities in drug poisoning mortality and all-cause mortality and expanding the set of county contexts to be examined beyond strict SES measures. The unique conditions of the pandemic called for an expansion of the SHDF beyond its original formulation.

Our analysis showed that the pandemic disrupted a “steady state” of social inequalities in all-cause and drug poisoning deaths. The importance of many county variables for predicting these deaths significantly increased during 2020 and 2021. Our interpretation is that the overall temporal patterns most closely align with stage 2 of the framework, growing inequalities. The rising importance of county income for all-cause mortality is one of the most striking features. Higher income counties were not only better able to weather the peak of the storm, but their advantages seem to have accumulated during the study period, pulling them even further ahead. The increasing association between county prevalence of adverse mental health and drug poisoning mortality is also striking, suggesting that during an exogenous shock, such as a pandemic, a county’s preexisting mental health profile can be a highly influential predictor of mortality.

Limitations and Future Directions

When interpreting the findings from this study, one should keep in mind that our unit of analysis was counties, not individuals. Therefore, the findings should not be interpreted as reflecting individual-level associations. Another caveat is that our study estimated associations, not causal effects, between county variables, state pandemic policy indices, and county deaths. Nevertheless, claims about causality are unnecessary for addressing our research questions. We also note that death certificates can misclassify drug poisoning and COVID-19 deaths, and misclassification varies geographically due to factors such as resources and qualifications of the certifier and the availability of COVID-19 testing (NASEM 2021; Stokes et al. 2021). Partly as a result, COVID-19 deaths may have been underreported in 2020 and 2021, especially in disadvantaged counties, such as those with low incomes (Stokes et al. 2021). One implication is that our estimated associations between county characteristics and COVID-19-related mortality may be somewhat conservative. We also note that our measures of county contexts represent the prepandemic period. This decision was based on the fact that most of these annual measures change slowly over time, not all measures are harmonized across 2018 to 2021, several measures are unavailable for 2021, and any changes in these annual measures are unlikely to produce an immediate effect on county mortality. Our decision to use 2018 measures is also supported by our overarching aim to understand the importance of prepandemic county contexts on working-age mortality.

We did not adjust for counties’ COVID-19 vaccination rates because doing so in our modeling framework would have obscured the temporal and spatial variation that is central to our study and the SHDF. Recall, the framework posits that the widespread dissemination of new knowledge and technology for avoiding a new disease is the core process through which social inequalities in the disease are reduced and eliminated. Controlling for county vaccination prevalence, which modified individuals’ susceptibility and behaviors, would have obscured the county-level variation that we aimed to observe using our modeling approach (as previously mentioned, our supplementary analyses show that accounting for states’ pandemic policies did not affect the model results). Including this information could be useful in future work aiming to test how county-level vaccination rates, pandemic policies, and more may have acted as pathways creating the patterns that we documented. Lastly, our study ends in Q4 2021, the last year of the NCHS restricted mortality files available to us but not the end of the COVID-19 pandemic. It will be informative to monitor ongoing inequalities as new data become available.

Conclusion

This study sheds light on how the exogenous shock of the pandemic exacerbated the importance of certain county contexts for predicting county deaths during 2020 and 2021 compared to 2019. In general, the self-rated physical and mental health of residents and median household income going into the pandemic were key predictors of county deaths for much of 2020 and 2021. The association between county self-rated health and all-cause mortality grew stronger during the pandemic but was returning to its prepandemic size by Q4 2021. However, the association between county income and mortality continued to grow larger for all-cause, drug poisoning, and COVID-19-related deaths. Our findings also provided an unusual case study for the SHDF. They showed how social inequalities in deaths can rise and fall in a population, as predicted by the framework, but then unexpectedly rise again, potentially due to social, cultural, political, and other such structural forces.

Author Biographies

Jennifer Karas Montez is a professor of sociology, Gerald B. Cramer Faculty Scholar in Aging Studies, and director of the Center for Aging and Policy Studies at Syracuse University. Her research focuses on trends and inequalities in U.S. population health since the 1980s and the growing influence of U.S. state policies on those outcomes. Her work has been supported by the U.S. National Institute on Aging, Robert Wood Johnson Foundation, an Andrew Carnegie Fellowship, and more.

Shannon M. Monnat is professor of sociology, director of the Center for Policy Research, and Lerner Chair in Public Health Promotion and Population Health at Syracuse University. Her research examines trends and geographic differences in health and mortality. She has authored over 70 peer-reviewed journal articles as well as numerous book chapters and policy briefs. Her work has been funded by the National Institutes of Health, Robert Wood Johnson Foundation, U.S. Department of Agriculture, and several other agencies and foundations.

Emily E. Wiemers is an associate professor in the Department of Public Administration and International Affairs and an O’Hanley Faculty Scholar at Syracuse University. Her work examines intergenerational ties and economic well-being across the life course and has been published in leading journals in economics, demography, and gerontology. She is currently the principal investigator on a National Institute on Aging funded project examining the consequences of family support up and down the generational ladder for well-being during the pandemic.

Douglas A. Wolf is an emeritus professor of public administration and international affairs at Syracuse University. His research addresses topics in demography, gerontology, public policy, and population health. His current work focuses on late-life disability and the role of state and local policies on public health outcomes.

Xue Zhang is a research associate in the Department of City and Regional Planning at Cornell University. Her research focuses on population health, regional economics, equity, and government policy. Her current research employs a human ecological framework to examine geographic differences in demographic structure, social determinants of health, age-friendly planning, social policy, and population health outcomes with a specific focus on rural-urban differences.