Abstract
Introduction Excess mortality does not depend on labeling the cause of death and is an accurate representation of the pandemic population-level effects. A comprehensive evaluation of all-cause excess mortality in the United States during the first two years of the Covid-19 pandemic, stratified by age, sex, region, and race/ethnicity can provide insight into the extent and variation in harm.
Methods With Centers for Disease Control and Prevention (CDC)/National Center for Health Statistics (NCHS) data from 2014-2022, we use seasonal autoregressive integrated moving averages (sARIMA) to estimate excess mortality during the pandemic, defined as the difference between the number of observed and expected deaths. We continuously correct the number of monthly expected deaths to reflect the decreased population owing to cumulative pandemic-associated excess deaths recorded. We calculate excess mortality for the total US population, and by age, sex, US census division, and race/ethnicity.
Results From March 1, 2020, through February 28, 2022, there were 1.17 million excess deaths in the United States. Overall, mortality was 20% higher than expected during the study period. Of the excess deaths, 799,477 (68%) were among residents aged 65 and older. The largest relative increase in all-cause mortality was 27% among adults ages 18-49 years. Males comprised most of the excess mortality (57%), though this predominance declined with increasing age. A higher relative increase in mortality occurred among Hispanic and non-Hispanic Black persons. Excess mortality differed by region; the highest rates were in the South, including in the population ages ≥65 years. Excess mortality rose and fell contemporaneously with Covid-19 waves.
Conclusion In the first two years of the pandemic, the US experienced 1.17 million excess deaths, with greater relative increases in all-cause mortality among men, in Black and Hispanic people, and in the South.
Introduction
Disease-specific mortality determinations are subjective, which can lead to both overcounts and undercounts. For a new cause of death, such as infection with SARS-CoV-2, the subjectivity may be amplified because of variations in testing and the co-existence of the pathogen with other causes of death. All-cause excess mortality provides a better estimate of the overall burden of the pandemic as it does not depend on cause of death determinations. Also, excess mortality measurement by subgroups enables comparisons without the need for adjustment of health differences among populations, nor historical differences.
All-cause excess mortality has been observed during the Covid-19 outbreak in the United States.1,2 Marked disparities in either all-cause excess mortality or Covid-19-attributed mortality by age, gender, and race/ethnicity have been noted.3–6 Here, we provide an excess mortality update two years into the Covid-19 pandemic in the United States. These estimates, unlike others, reflect the differences in the denominator by region, age, and gender, race/ethnicity owing to variation in pandemic-associated excess mortality. Thus, the present analysis, which accounts for these changes, provides a more accurate (and higher) excess mortality estimate for the first two years of the US Covid-19 outbreak than official estimates. We also provide period and overall sub-analyses by age group, sex, geographic location, and race/ethnicity.
Methods
Data Sources
We use publicly available Centers for Disease Control and Prevention (CDC)/National Center for Health Statistics (NCHS) population data from 2014-2019, and monthly mortality data from January 2015-February 2020. We use autoregressive integrated moving averages (ARIMA) to project the expected populations for 2020-2022, and then smoothed the yearly increases by dividing the yearly change by 12 and adding 1/12 of the growth to each successive month. We further adjusted (i.e., lowered) the population during the pandemic period to in proportion to the cumulative excess mortality recorded during the pandemic, so as to capture the smaller than expected population, owing to pandemic-associated all-cause excess mortality, as described previously.7–9
Excess Mortality
Excess mortality is defined as the difference between the number of observed and expected deaths.
Statistical Analysis
We divided the two-year study period (March 1, 2020, through February 28, 2022) into 6 periods: Wave 1 (March 1, 2020-May 31, 2020), Wave 2 (June 1, 2020-September 30, 2020), Wave 3 (October 1, 2020-February 28, 2021), Spring 2021 (March 1, 2020-June 30, 2021), Delta (July 1, 2021-December 31, 2021), and Omicron (January 1, 2022-February 28, 2022).
Then, using CDC Wonder data, we apply seasonal ARIMA (sARIMA) to monthly mortality data from January 1, 2015-February 29, 2020 to project the monthly number of expected deaths for the period of March 1, 2020-February 28, 2022, normalizing for difference in days in each month. Expected deaths for each of the four US Census Bureaus is determined by summing estimated from the nine US Census Divisions, which in turn are determined by age and gender subgroup projections (ages 0-17, 18-49, 50-64, and ≥65 by sex). These 72 components are then summed to create US-level estimates. For the race and ethnicity analysis, adequate data for monthly projects are available for non-Hispanic Black, non-Hispanic White, and Hispanic (all races) populations only. US level race/ethnicity estimates are determined by summing 24 component parts (8 sARIMAs for each of the three race/ethnicity groups included ages 0-17, 18-49, 50-64, ≥65 by sex). State-level estimates reflect the summation age-specific component models for 37 states; 14 states with inadequate age-specific data are modeled without age-stratification. As we have described previously, we continuously correct the number of monthly expected deaths to reflect the decreased population owing to cumulative pandemic-associated excess deaths recorded. The raw number of excess deaths, observed-to-expected ratios, and incident rates per 100,000 person-periods are reported for excess mortality.7,8 Raw and incident rates for Covid-19-specific mortality (ICD U07.1) is also reported. Analyses were conducted with R version 4.1.2. Statistical significance was defined as a 95% CI that did not include 0.
This study used publicly available data and was not subject to institutional review board approval according to the National Center for Health Statistics policy.
Results
During the first two years of the Covid-19 pandemic in the US (March 1, 2020 – February 28, 2022), there were 1,173,156 all-cause excess deaths (5,817,974 expected, 6,991,130 observed, 337.1 per 100,000 pandemic period population), and observed-to-expected ratio of 1.20 (Figure 1A, Table 1; Supplemental appendix Figure 1). All-cause excess mortality varied by region (Figure 1A, Table 1), and rose and fell contemporaneously with Covid-19-specific mortality in all states and regional jurisdictions (Supplemental Appendix Figure 1A).
Regional analysis
During the overall study period, there were notable differences in excess mortality by Census Bureau region, (Figure 1A), by Census Division, and by state (Supplemental Appendix Figure 1). There was excess mortality in all four Census Bureau regions during each of the 6 periods (Table 2). Census bureau contribution (Figure 1B-C) to and share of excess mortality (Supplemental Appendix Figure 1D) changed over time. During the study period, the highest cumulative excess mortality was observed in the South (rate and number), and the lowest overall excess mortality was observed in the West (rate) or Northeast (number), (Figure 1, Table 2, Supplemental Appendix Figure 1). Overall, 43.3% of the excess mortality was in the South despite comprising 38.5% of the US population; this accounted for 507,454 excess deaths (398.9 per 100,000), corresponding to an observed-to-expected ratio of 1.22. By comparison, the least excess mortality was observed in the Midwest where 230,098 excess deaths occurred (337.1 per 100,000), corresponding to an observed-to-expected ratio of 1.17, representing 20.7% of US excess mortality despite comprising 19.6% of the population. After the initial wave, the lowest rates of excess mortality were in the Northeast (220.7 per 100,000) while rates were higher in the other regions (range: 291.0-380.0 per 100,000, Table 2).
Age analysis
Excess mortality was observed among all adult age groups during the overall study period and varied by age (Figure 2A-D, Supplemental Appendix Figure 2). In adults, there was excess mortality in each of the 6 periods, except for US residents ages 65 and older during the spring 2021 period (Table 1-2, Figure 2A)
Numerically, the highest number of excess deaths occurred in residents ages 65 and older (Table 1). However, the largest relative increase in mortality occurred among residents ages 18-49 (Table 1).
Statistically significant excess mortality was not observed among children ages 0-17 during the pandemic months of 2020 but was observed during the Delta wave (1,532 excess deaths, IR 2.1 per 100,000 population, observed-to-expected ratio of 1.09) and 2021 overall (2,228, IR. 3.1 per 100,000 population, observed-to-expected ratio of 1.07) (Table 1, Table 3) for both males and females. Census Divisions and states large enough to provide adequate data indicated similar regional and subregional trends. (Supplemental Appendix Figure 2).
Among residents ages 65 or older, the highest overall rates of excess mortality during the pandemic period occurred in the South, even though during Wave 1, the highest rate of excess mortality recorded in any period among any age group occurred in Northeast residents ages 65 and over (Supplemental Appendix Table 1).
Sex analysis
Males accounted for 57% of excess mortality during the study period (Table 1). There were 663,228 excess deaths in men and 509,928 in women during the study period. Males comprised most of the excess deaths in all age groups, although the male share of excess mortality decreased with age (Table 3, Supplemental Figure S4A-B). The rate of excess death was 419.5 per 100,000 in men and 305.4 per 100,000 in women.
Race/Ethnicity analysis
There were marked differences in excess mortality by race, and ethnicity (Table 4, Figure 3A-F, Supplemental Appendix Figure 3) at the US and regional level. The highest increase in mortality was observed among Hispanic people, followed by non-Hispanic Black people, followed by Non-Hispanic White people (Table 4). There were marked differences by age group and wave. The rate of excess death was 346.0 per 100,000 in White people and 458.3 per 100,000 in Black people and 298.0 per 100,000 in Hispanic people. There were 200,190 excess deaths in non-Hispanic Black people and 184,277 excess deaths in Hispanic people. The observed-to-expected ratio was highest in Hispanic people (1.40), followed by non-Hispanic Black people (1.28), and then non-Hispanic White people (1.16), reflecting differences in baseline mortality (Table 4).
Discussion
Our study quantifies the mortality burden of the first two years of the pandemic across the United States and establishes the number of excess deaths and how they are distributed. This is the first study to show the distinct regional and age-specific changes in all-cause excess mortality that occurred during the initial two years of the pandemic. Using methodology that accounts for population changes across the pandemic and employs seasonal autoregressive integrated moving averages, we find that the United States had 1,173,156 all-cause excess deaths, which represented a 20% higher number of deaths than would have been expected based on the pattern in the 5 years immediately before the pandemic. The number of deaths was 40% higher than expected in Hispanic people and 28% higher for Black people, compared with 16% for White people. The South suffered the greatest burden, with 507,454 excess deaths, representing 22% higher than expected, in comparison with the Midwest, which was 17% higher than expected. The Northeast experienced a 20% increase in mortality. Although experiencing the brunt of the initial wave, the smaller deviations in mortality occurred after the first wave (14%) as compared to elsewhere, where after the initial wave, deviations from expected deaths were 18% (Midwest), 22% (West), and 24% (South).
The regional differences show that even accounting for baseline differences in expected mortality and the variation in population, the South suffered the most mortality burden from the pandemic, with the share increasing later in the pandemic period. While the South generally has higher age-adjusted mortality than the rest of the nation at baseline, it is important to note that excess mortality captures increases beyond the usual elevated mortality relative to other regions. Moreover, the pattern reveals that while the Northeast bore the brunt of the early pandemic period, the South exceeded the Northeast especially as vaccines became available, consistent with national data about vaccine adoption. However, other Covid-19 mitigation efforts may have contributed to these findings. The tracking of the excess mortality with the infection waves suggests that these excess deaths are related to the virus, and not a by-product of other societal changes. The study puts a number on the excess mortality toll of the pandemic and provides an assessment that is not biased by labels on cause of death.
The study also illuminates how the toll of excess mortality fell disproportionately on Black and Hispanic people. (Note: in the next version of this manuscript, data on American Indian/Alaska Native, Native Hawaiian/Other Pacific Islander, and Asian populations will be included). These findings are consistent with other data describing how infection, hospitalization and death varied by social factors, including race and ethnicity. Race/ethnicity reflects social identity and race/ethnicity-based injustices derived from the organization of society and the racism that has spanned centuries and persists today. The excess deaths are beyond the higher historic death rates of these groups. Structural racism’s impact on health was augmented during the pandemic. Considering this, the US will need to reckon with the magnitude of the loss of life and the inherent unfairness of how the excess deaths are distributed among the population and seek solutions for the future.
The age-related findings also warrant attention. First, the vaccine rollout period (Winter and Spring 2021) was associated with a precipitous decrease in the share (and, in fact, the temporary elimination) of excess mortality contributed by the ≥65-year-old demographic. In fact, during 2021, the only age group to have less excess mortality than 2020, was the ≥65-year-old group, likely reflecting vaccine uptake. Remarkably, during April and May of 2021, a numerical majority of excess mortality were derived from adults ages 50-64 years, the only time a non-geriatric demographic represented most of the excess mortality during the pandemic. However, the share of deaths from the ≥65-year-old demographic soon became the majority again, as younger groups began to receive vaccinations, and as the Delta variant became dominant, which decreased vaccine effectiveness against severe disease in older populations.1011 Second, we believe we are the first to describe the appearance of all-cause excess mortality in US residents ages 0-17 during the Covid-19 period, which occurred during the Delta period, which was sufficient to cause statistically significant all-cause excess mortality for 2021 overall. Reasons for this are unclear, but 69% (1,532 of 2,228) excess deaths in this demographic recorded in 2021 occurred during the Delta wave.
The study also revealed that men had higher excess mortality, accounting for 57% of those deaths even though on a population basis, there are many more women in older age groups. Sex differences in hospitalizations and deaths have been noted, but the reason why men are more vulnerable is as yet unexplained. The method used in this study addresses the higher historic mortality rates for men compared with women, and these excess deaths are in addition to a baseline higher risk.
This study has notable features to strengthen the validity of our findings. By summing component age and sex models to create regional and ultimately US-level figures, we generated the highest fidelity excess mortality estimates of which we are aware, avoiding biases that are otherwise introduced due to secular demographic heterogeneity that exists across jurisdictions. Additionally, we corrected for the ongoing population changes (i.e., smaller increases than expected), owing to cumulative excess mortality observed during the pandemic. Together, our models indicate that the CDC’s point estimate for all-cause excess mortality during the first two years of the US Covid-19 outbreak is approximately 11% lower than the actual number, which corresponds to >115,000 more excess deaths during the 24-month study period.
The study has several limitations. First, while all-cause and Covid-19-specific mortality trends were highly aligned, we cannot definitively establish causality. We note that the strength of the excess mortality metric is that it encompasses all deaths, thereby more comprehensively capturing the population-level effects of the pandemic. However, analysis of disease-specific causes of death, particularly in younger demographics, may be fruitful. Second, we used mortality categorized by decedent residence (not occurrence) due to the data availability. Therefore, a small fraction of deaths may have occurred in jurisdictions other than the decedents’ home states. Third, the race/ethnicity data depends on what was reported on death certificates and the Census Division.
In conclusion, the Covid-19 pandemic was associated with an increase in all-cause deaths in the United States by more than 20% beyond what would have been expected based on recent historic norms, resulting in nearly 1.2 million more deaths than expected in the first two years. Several groups suffered considerably a higher deviation in all-cause deaths compared with the historic norms, including young people, non-Hispanic Black and Hispanic people, men, and those living in the South. Future pandemic planning must consider strategies to better mitigate the overall harm, and specifically address populations at highest risk. The challenge going forward is to develop pandemic response that mitigates harm, particularly where the identified risk is greatest.
Data Availability
All data produced in the present study are available upon reasonable request to the authors
Data Statement
Dr, Faust and Mr. Renton had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Conflict of Interest Disclosures
Dr. Krumholz reported receiving consulting fees from UnitedHealth, Element Science, Aetna, Reality Labs, F-Prime, and Tesseract/4Catalyst; serving as an expert witness for Martin Baughman law firm, Arnold and Porter law firm, and Siegfried and Jensen law firm; being a cofounder of Hugo Health, a personal health information platform; being a cofounder of Refactor Health, an enterprise health-care, artificial intelligence-augmented data management company; receiving contracts from the Centers for Medicare & Medicaid Services through Yale New Haven Hospital to develop and maintain performance measures that are publicly reported; and receiving grants from Johnson & Johnson outside the submitted work. The other authors declare no competing interests.
Acknowledgements
The authors wish to thank Katie Dickerson Mayes, MD, PhD, Andrew Marshall, MD, Jacob Aguilar, PhD, Kristen Panthagani, MD, PhD and Lauren Rossen, PhD for their assistance and expertise.