Unemployment, Bankruptcies, and Deaths From Multiple Causes in the COVID-19 Recession Compared With the 2000‒2018 Great Recession Impact

Abstract

Objectives. To determine whether unemployment and bankruptcy rates are related to increased excess deaths during the COVID-19 recession and to examine whether the current recession-based mortality rate not only is dependent on COVID-19 but also continues the pattern of recessions, especially the Great Recession, in relation to chronic disease mortality rates and mental health disturbances (e.g., including suicide) from 2000 to 2018.

Methods. This study used pooled cross-sectional time series analysis to determine the impact of unemployment and bankruptcy rates on excess deaths from February to November 2020 for US states. The study used a second pooled cross-sectional time series analysis to determine whether the COVID-19‒ era recessional mortality continues the impact of prepandemic recessions (2000–2018) on multiple causes of mortality.

Results. Ten percent unemployment was associated with approximately 48 149 excess deaths, while, jointly with bankruptcies, their combined effect produced 35 700 and 144 483 excess deaths, for unemployment and bankruptcies, respectively. These health-damaging COVID-19‒recessional findings suggest a reiteration of the significantly increased major cause‒specific mortality during 2000 to 2018, mitigated by the size of the health care workforce.

Conclusions. Minimization of deaths attributable to the COVID-19 recession requires ample funding for the unemployed and underemployed, especially Black and Hispanic communities, along with significant investments in the health workforce. (Am J Public Health. 2021;111(11):1950–1959. https://doi.org/10.2105/AJPH.2021.306490)

The United States continues to experience an unpredictable COVID-19 pandemic, during which deaths have been accelerating since November 2020, and the national toll has reached 4000 persons per day. Overall, deaths have exceeded 595 000 as of June 9, 2021. The Centers for Disease Control and Prevention has estimated a reduction, in the first half of 2020, in years of life expectancy—with 2.7 years lost by African Americans, 1.9 years lost by Hispanic populations, and 0.8 years lost by Whites.¹ The current COVID-19 pandemic is exacerbated by the appearance of newer variants originating in the United Kingdom, South Africa, and Brazil, which bring further uncertainty to the death rate. The loss of life expectancy raises the question whether—or to what degree—the estimated deaths are partly the result of the abrupt national recession, which caused extraordinarily high unemployment rates and business closures in the onset of the COVID-19 pandemic. Is the accompanying recession a source of additionally increased excess deaths?² If the COVID-19‒based recession has, in itself, produced higher-than-expected mortality, is this a unique feature of the COVID-19 pandemic—increased mortality would thus be the result of a natural disaster—or is it a continuing effect of heightened unemployment and loss of income that, as evidenced during the Great Recession, may have increased national and state mortality during the first 2 decades of the 21st century?

The impact of COVID-19 on recession, especially unemployment, is clear from reports by economists.^3,4 But does COVID-19‒based unemployment by itself produce additional mortality beyond that initiated directly through the COVID-19 infection? Literature over 40 years, consistent with epidemiology at the individual level,^5‒9 shows, using a variety of methods, medium- to long-term (5- to 10-year) effects of recessions on elevated mortality (especially cardiovascular causes).^10‒15

But are there short-term effects within the same year of increased unemployment and bankruptcies on excess deaths during the COVID-19 period of February to November 2020? And, if so, are these effects predictable resumptions of the impact of prepandemic recessions on mortality during the first 2 decades of the 21st century, which includes the Great Recession and its aftermath? The evidence so far shows conflicting findings.¹⁶ The present single-year analysis of the health damage of unemployment supports the relationship between recessional factors and mortality rates; it approximates the types of calculations typical of business cycle analysis (although unemployment is understood to be a “lagging” business cycle indicator).¹⁷

To answer these questions, I explored the potential impact of unemployment and bankruptcies on total excess deaths over the COVID-19 pandemic period of February to November 2020. I furthermore examined the relationship between employment status and gross domestic product (GDP) declines during the prepandemic 2000–2018 period as potentially linked to mortality. Proximate causes of mortality include heart disease, cancer, stroke, diabetes, chronic lower respiratory disease, and suicide.

METHODS

In many observational studies, observations are available over a sequence of points in time (e.g., states and years as in our case). Examination of only 1 dimension (i.e., space or time) would limit us to perform classical cross-sectional or time-series regression analysis. Drawing on more advanced techniques^18,19 (i.e., pooled cross-sectional time series analysis [PCSTS]), allows us to model simultaneously both space and time components as discussed by Reibling.²⁰

Pooled Cross-Sectional Time Series Analysis

The PCSTS method combines 2 approaches. First, the more familiar one is cross-sectional analysis, where, in this case, US states were the units of analysis. I examined multiple cross-sectional analyses corresponding to the 19-year period of 2000 to 2018 for which all of the data representing the individual variables were available with respect to all 50 US states. Second, the same is true for the PCSTS analysis of the monthly period of February to November 2020, which I analyzed by state. I based all variables used in these PCSTS analyses on aggregated data (i.e., population rates rather than individual-level data).²¹ In addition to the cross-sectionality of this procedure in both periods (2000‒2018 and February‒November 2020), the technique simultaneously entails time-series analysis, involving variations over time in the individual predicted variables and the outcome variables—age-adjusted rates of overall mortality and those for major chronic diseases (apart from dementia) and suicide.²²

A more detailed discussion of the PCSTS approach is provided in Appendix A (available as a supplement to the online version of this article at http://www.ajph.org). References to the data sources for each of the variables are provided in Appendix B (available as a supplement to the online version of this article at http://www.ajph.org).

Major Variables in the Predictive Models

Three central variables, acting together, hypothetically promoted or harmed (in the case of the unemployment rate) overall population health during 2000 to 2018. These variables were the unemployment rate, GDP per capita with 5-year lag, and the size of the health care workforce. The lagged measure of GDP per capita was, hypothetically, the principal source of medium- and long-term national economic and population health gain.

Following GDP per capita with a 5-year lag, a key predictive variable was the very short-term impact of the unemployment rate, representing the immediacy of economic recession, during a single year (i.e., without lag). The third variable was the availability of health care represented by the size of the US health care workforce per 1000 of total employment. This variable is a potential central moderator of the effects of unemployment and GDP per capita with a 5-year lag. In this 3-variable overall hypothesis, all 3 variables act jointly to influence the overall mortality rate. The study ’s argument required the modeling of all 3 variables simultaneously. Additional control variables (including severe poverty, tobacco, opioid and extensive alcohol consumption, and environmental pollution) are listed in detail in the section on control variables, including rationales, operationalization, and references for each control variable.

Components of the National Economy

Journalistically, the unemployment rate has been the principally used indicator measuring the presence of recessions. Despite alternative measures for which there are plausible arguments, this rate is nevertheless the metric that is most commonly used in policy discussions of the implications of recession, both short- and long-term.

While past researchers have typically either concentrated on the medium-term or lengthy effects of unemployment on damaged health, recent literature at the macroeconomic level has tended to focus on the near-simultaneous relation between unemployment and mortality. Such refocusing assumed that both unemployment and mortality could potentially behave like typical business cycle indicators, the effects of which are often seen within a single year. However, in the case of unemployment, it is understood by the National Bureau of Economic Research to be a lagging indicator, the maximum effect of which can remain high at least a year or 2 following the end of economic recessions.²³ Thus, to capture the health implications of unemployment within the same year that unemployment comes to its peak, it was necessary to treat the unemployment without any lag (i.e., contemporaneously) from February to November 2020, thus capturing only very short-term effects. At the same time, I used estimates of GDP per capita with a 5-year lag to capture medium-term effects of recession and growth on mortality from 2000 to 2018.

Log Gross Domestic Product per Capita

The most comprehensive hypothesized variable benefiting population health is log GDP per capita with a 5-year lag for the period 2000 to 2018. This variable is as yet unavailable for February to November 2020. The GDP measure, with a 5-year lag, expresses the standard of living for the general society over the medium term within a full 5- to 6-year business cycle.²³ In the present study, I expressed GDP per capita in terms of a 5-year lag to take account of the cyclical implications of investments that require several years to result in health improvements (especially in health care technology and pharmaceuticals), in occupational‒ environmental health, and poverty minimization via social welfare expenditures affecting many categories of living conditions and elevated consumption of necessary goods and services (e.g., nutrition, clothing, shelter, electricity, transportation, rent).

Long-term GDP per capita has predominant importance for improved public health even though it includes elements of substantial economic inequality,²⁴ especially by US region, income group, educational status, and race/ethnicity. The economic inequality issue, in conjunction with long-term economic growth (for the general population and especially high-income groups) should not be underestimated. In the United States, rural, Midwestern, and Rust-Belt areas have fallen behind in economic growth and life span. It has been argued that these aspects of lagging development have been at least partly responsible for the emergent trends of the opioid crisis among younger populations and long-term loss in life expectancy (i.e., “deaths of despair”).²⁵

Health care workforce (2000–2018)

The size of the health care workforce over long periods in US history reflects a sustained upward trend. At the same time, the short-term and long-term effects of recessions have decreased the size of the health care workforce because of a loss of health insurance related to unemployment during recessions and a recessional loss of income to the population more generally, thus inhibiting investment in an expanded workforce. Thus, I separately investigated the trend in the size of the health care system, highly contingent on scientific and technological advances and long-term health policy, as to its long-term implications for societal health.

Total bankruptcies (February‒November 2020)

In the February to November 2020 analysis, in addition to unemployment, I also used bankruptcies (private and commercial) as a COVID-19 recessional predictor of excess deaths. In the 2020 monthly analyses of the impact of recession, it was important to take into account a somewhat broader set of measures—especially those that influence business as a whole. A prominent and traditional business cycle indicator is total bankruptcies,²⁶ which provides a more widespread sense of the extensiveness and depth of the COVID-19‒ initiated recession. The total bankruptcy rate provides an indication of losses to small and large businesses.

Control variables

I introduced behavioral risk factors as variables into the predictive models to distinguish them from the more direct effects of income change or unemployment.

The behavioral risk factors from 2000 to 2018 were as follows:

1. Major depressive episodes: The intention was to discriminate between depressive episodes that were clearly linked to recession and those that were not necessarily associated with recession, but rather emanated from major life events²⁷ and daily hassles,²⁸ including those that occurred at the workplace, in family life, and especially resulting from the loss and grief related to damaged health and mortality. Nevertheless, depression is a significant risk factor for poor health, low life satisfaction, and early mortality.

2. Smoking: Smoking, a basic behavioral risk factor, was measured at a 3-year lag because the long-term trend in industrialized societies (especially the United States) has involved major declines in smoking prevalence, greatly curtailing mortality from cardiovascular illnesses, malignancies, and chronic obstructive pulmonary disease. The 3-year lag was hypothesized because there is evidence that several years are often required after smoking cessation for the former smokers’ health to improve to a point that it returns to previous nonsmoker cardiopulmonary functioning.

3. Participation in physical activities: It is now widely accepted that participation in physical activities is a major source of health maintenance and improvement.

4. Alcohol-induced death rate: The literature is complicated regarding the health impact of the overall population consumption of alcohol. Especially for the cardiovascular illnesses, very high as well as virtually zero consumption are associated with elevated mortality, whereas moderate consumption appears to enhance longevity.²⁹ To specifically indicate chronically higher alcohol consumption, which elevates mortality for many chronic and mental health causes, I used age-adjusted alcohol-related deaths as a behavioral risk factor for mortality.

5. Tuberculosis: Tuberculosis incidence is especially high in low-income developing countries, but modest in industrialized countries. I used it as a predictive behavioral risk factor because of its intense association with chronic poverty, but not necessarily with trends in GDP or recession.

For references and further discussion of control variables, see Appendix C (available as a supplement to the online version of this article at http://www.ajph.org).

RESULTS

In the COVID-19 period of February to November 2020, unemployment was a significant predictor of excess deaths controlling for the number of COVID-19 cases, age, and Black and Hispanic racial/ethnic groups (Table 1). Using the same model, with identical controls, the combination of both recessional factors of unemployment and bankruptcies yielded an impact of further increased excess deaths (Table 2). In the analysis presented in Table 2, which combines unemployment and bankruptcies, the numerical effect on excess deaths related to unemployment slightly decreased because of the relationship between unemployment and bankruptcies that typically would occur in a recession. In the model that presents 10% unemployment (as currently estimated by the Federal Reserve and Department of the Treasury³⁰) as the only recessional variable, an additional 48 149 deaths were estimated (Table 1). In the recessional model that included both unemployment and bankruptcies, the estimates of 10% additional unemployment led to 35 700 excess deaths, and a 120-unit increase per 100 000 in bankruptcies led to approximately 144 483 deaths.

TABLE 1—

Cumulated Excess Deaths Rate per 100 000, All Causes, Pooled Cross-Sectional Time Series Analysis: 50 US States, February‒November 2020

Variables	Cumulated Excess Deaths, All Causesa	Coefficient (P)
Cumulated incident cases rate/100 000		0.02 (< .001)
2000-unit increase	150 537
4000-unit increase	301 074
6000-unit increase	451 612
8000-unit increase	602 149
10 000-unit increase	752 686
Unemployment rate as proportion (%) of total labor force aged ≥ 16 years, seasonally adjusted		1.47 (< .001)
5-unit increase	24 074
10-unit increase	48 149
15-unit increase	72 223
20-unit increase	96 297
25-unit increase	120 372
Percentage of population aged ≥ 75 years, July 1, 2019, estimate		7.62 (.018)
1-unit increase	24 997
2-unit increase	49 994
3-unit increase	74 991
4-unit increase	99 988
5-unit increase	124 985
Percentage of Black or African not Hispanic or Latino, July 1, 2019, estimate		1.03 (.001)
5-unit increase	16 927
10-unit increase	33 855
20-unit increase	67 709
30-unit increase	101 564
40-unit increase	135 418
Percentage of White Hispanic or Latino, July 1, 2019, estimate		0.28 (.35)
5-unit increase	4517
10-unit increase	9034
20-unit increase	18 067
30-unit increase	27 101
40-unit increase	36 135
Constant		−59.78 (.014)

Note. Data show expected additional excess deaths, all causes, when the independent variable increases by the specified units. The estimated coefficients from the excess deaths model were used to calculate the associated change of the cumulated excess deaths rate, all causes. Based on the total US population from the year 2019, the change of the cumulated excess deaths rate was converted to expected additional excess deaths numbers (e.g., a change of the unemployment rate by absolute 10% resulted in 48 149 additional excess deaths). The potential changes of the independent variables were derived by inspection of the descriptive statistics for each variable. Data and their sources of control variables are listed in Appendix B (available as a supplement to the online version of this article at http://www.ajph.org). For all key variables, a summary of descriptive statistics is provided in Appendix E (available as a supplement to the online version of this article at http://www.ajph.org).

^aEstimates for excess deaths were based on a 2019 US population of 328 239 523. The number of observations for the analysis was n = 500; R² = .65; Wald χ²(5) (Prob > χ²) = 1434.06 (P <.001).

TABLE 2—

Cumulated Excess Deaths Rate per 100 000, All Causes, Pooled Cross-Sectional Time Series Analysis: 50 US States, February‒November 2020

Variables	Cumulated Excess Deaths, All Causesa	Coefficient (P)
Cumulated incident cases rate/100 000		0.01 (< .001)
2000-unit increase	98 141
4000-unit increase	196 282
6000-unit increase	294 423
8000-unit increase	392 564
10 000-unit increase	490 705
Unemployment rate as proportion (%) of total labor force aged ≥ 16 years, seasonally adjusted		1.09 (< .001)
5-unit increase	17 850
10-unit increase	35 700
15-unit increase	53 549
20-unit increase	71 399
25-unit increase	89 249
Cumulated total bankruptcies rate/100 000		0.37 (< .001)
60-unit increase	72 242
120-unit increase	144 483
180-unit increase	216 725
240-unit increase	288 967
300-unit increase	361 208
Percentage of population aged ≥ 75 years, July 1, 2019, estimate		7.99 (.015)
1-unit increase	26 243
2-unit increase	52 485
3-unit increase	78 728
4-unit increase	104 971
5-unit increase	131 213
Percentage of Black or African not Hispanic or Latino, July 1, 2019, estimate		0.48 (.14)
5-unit increase	7 911
10-unit increase	15 822
20-unit increase	31 643
30-unit increase	47 465
40-unit increase	63 286
Percentage of White Hispanic or Latino, July 1, 2019, estimate		0.42 (.16)
5-unit increase	6 846
10-unit increase	13 692
20-unit increase	27 385
30-unit increase	41 077
40-unit increase	54 770
Constant		−73.19 (.003)

Note. Data show expected additional excess deaths, all causes, when the independent variable increases by the specified units. The estimated coefficients from the excess deaths model were used to calculate the associated change of the cumulated excess deaths rate, all causes. Based on the total US population from the year 2019, the change of the cumulated excess death rate was converted to expected additional excess deaths numbers (e.g., a change of the unemployment rate by absolute 10% results in 35 700 additional excess deaths). Analogously, when the cumulated bankruptcies rate per 100 000 changes by 120 units, then this will result in 144 483 additional excess deaths. The potential changes of the independent variables were derived by inspection of the descriptive statistics for each variable. Data and their sources of control variables are listed in Appendix B (available as a supplement to the online version of this article at http://www.ajph.org). For all key variables, a summary of descriptive statistics is provided in Appendix E (available as a supplement to the online version of this article at http://www.ajph.org).

^aEstimates for excess deaths were based on a 2019 US population of 328 239 523. The number of observations for the analysis was n = 500; R² = .62; Wald χ²(5) (Prob > χ²) = 2053.66 (P < .001).

Is this finding a unique, natural consequence of a recession related to a world pandemic or a partial reinstantiation of the effects of recessions during the 21st century, specifically 2000 to 2018? Like total mortality, all major chronic disease causes of death showed significant beneficial effects of log GDP per capita with a 5-year lag and mortality increases related to unemployment without lag (Table 3).

TABLE 3—

Six Causes of Death, Rates per 100 000, Pooled Cross-Sectional Time-Series Models: 50 US States, 2000–2018

Variables	Total (ICD-10: A00‒Y89), Coefficient (P)	Cancer (ICD-10: C00‒D48), Coefficient (P)	Diabetes (ICD-10: E10‒E14), Coefficient (P)	Heart Diseases (ICD-10: I05‒I52), Coefficient (P)	Stroke (ICD-10: I60‒I69), Coefficient (P)	CLRD (ICD-10: J40‒J47), Coefficient (P)
Log normal of 5-y lag of real GDP in thousands of chained 2012 $ per capita	−173.66 (< .001)	−43.02 (< .001)	−10.65 (< .001)	−116.02 (< .001)	−38.19 (< .001)	−5.65 (< .001)
Unemployment rate as proportion (%) of total labor force aged ≥ 16 y	1.72 (< .001)	1.70 (< .001)	0.14 (< .006)	1.06 (< .001)	0.18 (.052)	0.26 (< .001)
Employment in health care support per 1000 of total employment	−35.69 (< .001)	−8.07 (< .001)	−2.00 (< .001)	−20.26 (< .001)	−5.24 (< .001)	−0.77 (.12)
Tuberculosis incidence rate per 100 000 of total population	12.69 (< .001)	3.21 (< .001)	0.26 (.005)	8.62 (< .001)	2.06 (< .001)	−0.71 (< .001)
Major depressive episode in the past year (%) in total population aged ≥ 18 y	7.25 (< .001)	1.11 (< .001)	0.62 (< .001)	4.91 (< .001)	1.31 (< .001)	−0.40 (.013)
5-y lag of daily smoker prevalence in total population aged ≥ 18 y	5.85 (< .001)	2.50 (< .001)	0.15 (.002)	2.90 (< .001)	0.77 (< .001)	0.54 (< .001)
Age-adjusted alcohol-induced death rate per 100 000 of total population	1.03 (.01)	−0.97 (< .001)	0.19 (< .001)	−0.35 (.12)	−0.39 (< .001)	0.11 (.08)
Prevalence of not participated in any physical activities in population aged ≥ 18 y	2.46 (< .001)	0.09 (.29)	−0.01 (.78)	0.94 (< .001)	0.07 (.27)	0.4 (< .001)
Constant	1313.05 (< .001)	300.19 (< .001)	60.11 (< .001)	554.51 (< .001)	177.65 (< .001	54.29 (< .001)
R2 overall	0.86	0.89	0.63	0.85	0.78	0.75
Wald χ2 (23) (Prob > χ2)	5504.14 (< .001)	6102.29 (< .001)	676.04 (< .001)	6488.50 (< .001)	4133.83 (< .001)	452.07 (< .001)

Note. CLRD = chronic lower respiratory diseases; GDP = gross domestic product; ICD-10 = International Classification of Diseases, 10th Revision (Geneva, Switzerland: World Health Organization; 1992). The number of observations was 950. All models have been adjusted with 15 regional dummies for New England, Mideast, Great Lakes, Southwest, Rocky Mountains, Far West, Alaska, Arizona, California, Florida, Hawaii, Louisiana, Nevada, Oklahoma, and Wyoming. Data and their sources of control variables are listed in Appendix B (available as a supplement to the online version of this article at http://www.ajph.org). For a detailed discussion of each of the control variables, see Appendix C (available as a supplement to the online version of this article at http://www.ajph.org). For all key variables, a summary of descriptive statistics is provided in Appendix E (available as a supplement to the online version of this article at http://www.ajph.org).

The principal recessional variable, unemployment, showed evidence of elevated mortality during 2000 to 2018 (total mortality) and major chronic causes of death and suicide and continued to do so in the period February to November 2020 for total excess death. This indicates a continuous relationship of recession (especially indicated by unemployment rates) through the 2 research sample periods elevating mortality. The principal control variables showed strong positive relations to total mortality and mortality by major cause. However, the health care workforce with inverse relations to mortality featured as the strongest coefficient among predictive variables except GDP per capita with a 5-year lag (Table 3).

The impact of recessions during 2000 to 2018 on suicide was clear, while the predictive model and, therefore, the findings, were somewhat different from those of total mortality and that for major chronic diseases. For suicide, the unemployment rate proved to be a robust and significant related predictive factor. GDP per capita, the major indicator of national economic change, was not predictive for suicide, while, when instead median household income for GDP with 5-year lag (which de-emphasizes income inequality) was included, the model showed the hypothesized relationship. However, the control variables, expressing stress relationships and coping mechanisms, were positively related to and constituted significant risk factors in the occurrence of suicide (Table 4).

TABLE 4—

Suicide Death Rate per 100 000, Pooled Cross-Sectional Time Series Model: 50 US States, 2000–2018

Variables	Suicide (ICD-10: U03, X60‒X84, Y87.0), Coefficient (P)
Log normal of 5-y lag of median household income in chained 2018 $	−1.28 (.043)
Unemployment rate as % of total labor force aged ≥ 16 y	0.09 (< .001)
Illicit drug use in the past month (%) in population aged ≥ 26 y	0.14 (<.001)
Gallons of alcohol per capita aged 14 y and older	1.87 (< .001)
Age-adjusted alcohol-induced death rate per 100 000 of total population	0.38 (< .001)
Age-adjusted nontransport accidents (W00‒X59, Y86) death rate per 100 000 of total population	0.08 (< .001)
Nonmedical use of pain relievers in past year (%) in population aged ≥ 26 y	0.11 (.1)
Major depressive episode in the past year (%) in population aged ≥ 18 y	0.16 (.007)
Constant	15.37 (.024)

Note. ICD-10 = International Classification of Diseases, 10th Revision (Geneva, Switzerland: World Health Organization; 1992). The model has been adjusted with 6 regional dummies for New England, Mideast, Great Lakes, Rocky Mountains, Alaska, and California. The number of observations was 950; overall R² = .79; Wald χ² (23) (Prob > χ²) = 1936.32 (P < .001). Data and their sources of control variables are listed in Appendix B (available as a supplement to the online version of this article at http://www.ajph.org). For a detailed discussion of each of the control variables, see Appendix C (available as a supplement to the online version of this article at http://www.ajph.org). For all key variables, a summary of descriptive statistics is provided in Appendix E (available as a supplement to the online version of this article at http://www.ajph.org).

DISCUSSION

In the COVID-19 recession, both unemployment and bankruptcies exerted a substantial damaging impact on excess deaths. COVID-19 incidence, age, and race/ethnicity were controlled. This means that the ethnic/racial factors, which have been widely reported as being especially important,³¹ were adjusted for in these 2 models covering the COVID-19 recession era. In these models, the potential importance of race/ethnicity in the COVID-19 recession was further highlighted by the fact that unemployment rates in 2020 were considerably higher for Hispanic and Black populations than for the White population³² and, for the same period, bankruptcies were especially high for the African American population, especially Black women.³³

However, the Federal Reserve and Department of the Treasury announced on February 22, 2021, that the official unemployment rate of 3.6% should really be taken as close to 10.0% because of many unemployed persons completely leaving the labor force.³⁰ If that is correct, then our estimates of the impact of unemployment could be underestimated.

Comorbidities of COVID-19 Mortality

In addition, the recessional effects of COVID-19 could also extend to the chronic disease comorbidities of COVID-19 mortality, as was the case for elevated chronic disease mortality in the 2000–2018 period. Main contributing factors in the COVID-19 pandemic could include the influence of psychological stress, loss of access to health care because of pressure on the health care system, and loss of financial resources. In such cases, because the Black population has been more susceptible to many chronic diseases, their relatively low economic and occupational status may well put them at greater risk for recessional losses. The important implication is that far more extensive work is required to understand how key elements of economic recession intersect with race/ethnicity to produce much higher-than-expected mortality rates.

Support for the long-term relationship between unemployment and increased mortality has been found at the national level^9,12–15 and extensively tested for at the individual level in epidemiological studies over 40 years.^5–8,10,11 A principal contribution of this article is the demonstration that short-term, intense, and abrupt increases in unemployment have led to elevated excess deaths during the COVID-19 recession as well as expanded total and chronic disease mortality during the 2000–2018 period. Despite the fact that the unemployment-to-mortality relationship has been established for medium- and long-term relationships,^5,6,14 it is only recently that economists have begun to study this relationship at the macro level within a single year (i.e., the same year) in the attempt to use temporally coincident (i.e., contemporaneous) indicators in the technical business cycle terminology.

At first sight, it would seem that examining only the very-short-term mortality impact of unemployment would be counterintuitive with respect to the epidemiological tradition of research, which stipulates that the development of chronic diseases occurs over much of the life span. Previous research using this short-term approach has produced contradictory findings.¹⁶ Nevertheless, in keeping with the style of research that is specific to short-term business cycle analysis, I analyzed the measures of the effects of recession, unemployment over 2020 and 2000 to 2018, and bankruptcy rates during 2020 (over a single year) in this article. At the same time, longer-term analysis by economists have emphasized the effects, after a first year of peak unemployment, of hysteresis, or scarring.³⁴ Hysteresis refers to the multiple effects lagging high recessional unemployment during which employment and income losses persist.

Nevertheless, the mortality-inducing effects of long-term and severe economic loss are supported by the most prominent and ubiquitous findings in epidemiology, in industrialized countries and globally. This finding has become known as the “social gradient” or “health gradient.”³⁵ It specifies that the higher the level of socioeconomic status of a person or population, the lower will be the mortality rate. The direct inference is that elevation of socioeconomic status decreases population and individual mortality, while declines in socioeconomic status produce increases in mortality. This study is apparently one of the first that deals with the importance of GDP per capita, on both an annual and long-term basis, in the reduction of mortality for major causes during 2000 to 2018. In this article, the value added is that statistical tests were used to examine this hypothesis on a US state basis. The analysis over 2000 to 2018 provides the extensive statistical degrees of freedom necessary to examine the impact of the size of the health care work force on multiple causes of mortality, holding constant overall GDP per capita. The resulting significant impact of the health care workforce is one of the major sources of mortality reduction during 2000 to 2018. It is clear that investment in the health workforce, including public health, has been a major factor leading to improvement in the public’s health. Indeed, it is possible that, absent the mortality-reducing effects of the health care workforce during 2000 to 2018, the impacts of recession on increased mortality may have been greatly enlarged.

Public Health Implications

The COVID-19‒induced unemployment and bankruptcy rates are robustly related to increased excess mortality from February to November 2020. Economists and labor market specialists should therefore join public health researchers to formulate policies that reduce mortality. Rapid policy intervention is especially required for populations of low socioeconomic status and communities of color who have suffered inordinately from COVID-19 in terms of morbidity and mortality. The vulnerability of these groups is attributable not only to occupations that more frequently involve interpersonal contact but also to low socioeconomic position, with considerably higher unemployment, income loss, and bankruptcy occurring to these populations in recession. We need to reconsider the epidemiology‒public health disciplines involving natural disasters—epidemics, heatwaves, floods, hurricanes, etc. The typical approach in public health disaster relief has been to concentrate on mitigating the immediate disaster. But the corollary economic and social implications of natural disasters could have medium- and long-term implications for substantially expanded illness and mortality if these corollary effects are not attended to.

Although the COVID-19 recession is unique in being caused by a natural disaster, it follows a pattern of recessions during the 21st century, including the Great Recession, of increasing the total mortality rate and mortality specifically attributable to major chronic diseases, such as heart disease, cancer, stroke, diabetes, and chronic lower respiratory disease, as well as mental health consequences, such as suicide.³⁶ It is clearly important for Congress to provide sufficient funds to aid in vaccine distribution and COVID-19 treatment of the full population. At the same time, the minimization of deaths attributable to the COVID-19 recession requires ample funding for the unemployed and underemployed, and to individuals and businesses who have experienced and are experiencing major economic losses. Also, it is especially important for funding to mitigate the maximum effect of recession to communities of color, rural populations, and those “left behind” in previous governmental support efforts during the Great Recession era. Given the vulnerabilities of communities of color to the pandemic and its economic consequences, it would be important to investigate more precisely how different racial/ethnic groups’ health vulnerabilities interact with economic losses imposed by recessions of the 21st century. Furthermore, the prepandemic 2000–2018 analysis makes clear that the absolute size of the health workforce is a highly significant factor in mitigating the results of economic recession, low socioeconomic status, and poverty. Investments into enlarging the size of the health workforce, and more generally into public health, is an important priority for reducing overall health inequalities in American society, in both the short and long term.