Excess all-cause mortality across counties in the United States, March 2020 to December 2021.

Official Covid-19 death counts have underestimated the mortality impact of the Covid-19 pandemic in the United States. Excess mortality, which compares observed deaths to deaths expected in the absence of the pandemic, is a useful measure for assessing the total effect of the pandemic on mortality levels. In the present study, we produce county-level estimates of excess mortality for 3,127 counties between March 2020 and December 2021. We fit two hierarchical linear models to county-level death rates from January 2015 to December 2019 and predict expected deaths for each month during the pandemic. We compare observed deaths to these estimates to obtain excess deaths for each county-month. An estimated 936,911 excess deaths occurred during 2020 and 2021, of which 171,168 (18.3%) were not assigned to Covid-19 on death certificates as an underlying cause of death. Urban counties in the Far West, Great Lakes, Mideast, and New England experienced a substantial mortality disadvantage in 2020, whereas rural counties in these regions had higher mortality in 2021. In the Southeast, Southwest, Rocky Mountain, and Plains regions, there was a rural mortality disadvantage in 2020, which was exacerbated in 2021. The proportion of excess deaths assigned to Covid-19 was lower in 2020 (76.3%) than in 2021 (87.0%), suggesting that a larger fraction of excess deaths was assigned to Covid-19 later in the pandemic. However, in rural areas and in the Southeast and Southwest a large share of excess deaths was still not assigned to Covid-19 during 2021. SIGNIFICANCE: Deaths during the Covid-19 pandemic have been primarily monitored through death certificates containing reference to Covid-19. This approach has missed more than 170,000 deaths related to the pandemic between 2020 and 2021. While the ascertainment of Covid-19 deaths improved during 2021, the full effects of the pandemic still remained obscured in some regions. County-level estimates of excess mortality are useful for studying geographic inequities in the mortality burden associated with the pandemic and identifying specific regions where the full mortality burden was significantly underreported (i.e. Southeast). They can also be used to inform resource allocation decisions at the federal and state levels and encourage uptake of preventive measures in communities with low vaccine uptake.

Introduction

The Covid-19 pandemic has had a substantial impact on mortality in the United States, leading to declines in life expectancy not previously observed since the end of World War II (1, 2). Estimates of excess mortality, which compare observed deaths to those expected in the absence of the pandemic, suggest that the true death toll of the pandemic is much larger than indicated by the official Covid-19 death tallies (3–7). For example, one study estimated that 550,000 excess deaths occurred between March 2020 and February 2021 and that approximately one quarter of these excess deaths were assigned to causes other than Covid-19 (8).

Excess deaths may have been assigned to causes other than Covid-19 for several reasons. Lack of access to testing in the community, combined with the inconsistent use of post-mortem testing for suspected cases, likely resulted in a large share of undiagnosed Covid-19 infections and deaths, especially early in the pandemic and in rural areas (9–12). Additionally, persons with comorbid conditions, such as cardiovascular disease, diabetes, hypertension, and pulmonary disease, may have had their cause of death assigned to the comorbid condition rather than to Covid-19 (13). Unattended Covid-19 deaths occurring in the community may have been especially likely to be assigned to another cause of death (14). Individuals who developed cardiovascular disease or diabetes as a result of the post-acute sequelae of Covid-19 and subsequently died also may not have had their deaths assigned to Covid-19 (15, 16). Finally, excess deaths not assigned to Covid-19 may also reflect deaths indirectly related to the pandemic, including deaths associated with reductions in access to health care, hospital avoidance due to fear of Covid-19 infection, increases in drug overdoses, and economic hardship leading to housing and food insecurity (17–23).

There are multiple benefits to using excess mortality rather than assigned Covid-19 deaths to assess the mortality impact of the Covid-19 pandemic. First, estimates of excess mortality are more comparable spatially than Covid-19-assigned mortality, because states use different definitions to assign Covid-19 deaths and local death investigation systems may have different policies and resources that affect assignment of Covid-19 deaths.[1] Second, as many Covid-19 deaths go uncounted, excess mortality is likely to provide a more accurate measure of the pandemic impact for purposes of resource allocation. Thus, continued tracking of excess mortality across time and space is vital to clarifying the total impact of the pandemic and where its impact is greatest, and to identifying the most appropriate policy responses and interventions.

Prior studies of excess mortality have primarily focused on national and state-level estimates (5, 6), but revealing the true mortality impact of the Covid-19 pandemic at the county-level is especially valuable for several reasons. First, because counties are the administrative unit for death investigation, excess mortality estimates have the potential to help identify counties with substantial Covid-19 under-counts that would benefit from additional training in cause-of-death certification (25). Such estimates may also be valuable for informing local public health workers, community leaders, and residents of the true impact of the pandemic, thus potentially increasing vaccination and uptake of other protective measures (26). These estimates may also be used to target federal and state emergency resources, such as funeral assistance support from the Federal Emergency Management Agency (FEMA). Finally, estimating excess mortality at the county-level also enables analyses of the factors affecting mortality associated with the pandemic, including geographic dimensions like the urban/rural continuum.

One prior study generated predictions of excess mortality at the county level but was limited to data from 2020 and pooled small counties together to increase precision (4). Expanding these estimates to 2021 is critical because the geography of the pandemic has likely changed markedly since 2020 as a result of policy shifts, the availability of vaccines, and the emergence of new variants. In the present study, we developed two hierarchical models to estimate excess mortality for 3,127 harmonized counties[2] for the period from March 2020 to December 2021. We also evaluated the extent to which excess deaths are accounted for in official Covid-19 death tallies as an indicator of potential under-reporting of Covid-19 mortality. In addition to generating county-level estimates of excess mortality, we produced several aggregations of the county estimates to examine differences in excess mortality by U.S. Bureau of Economic Analysis (BEA) region and across metropolitan (metro) vs. non-metropolitan (nonmetro) areas.

Results

Across the United States, 459,764 excess deaths occurred between March and December 2020, and 477,147 excess deaths occurred in 2021. This equals 936,911 excess deaths during calendar years 2020 and 2021, of which 765,743 (81.7%) were assigned to Covid-19 as an underlying cause of death and 171,168 (18.3%) were not assigned to Covid-19. In 2020, excess death rates were highest in nonmetro areas (207 deaths per 100,000 residents) followed by large central metro areas (173 deaths per 100,000 residents), whereas in 2021, excess death rates were highest in nonmetro areas (227 deaths per 100,000 residents) followed by small or medium metros (163 deaths per 100,000 residents). Regionally, the excess death rate in 2020 was highest in the Mideast (206 deaths per 100,000 residents) followed by the Southwest (188 deaths per 100,000 residents). In 2021, however, the excess death rate was highest in the Southeast (205 deaths per 100,000 residents) and the Southwest (198 deaths per 100,000 residents). The areas with the highest excess death rates in 2020 were large central metros in the Mideast and nonmetro areas in the Southwest and Southeast. In 2021, the areas with the highest excess death rates were nonmetro areas in the Southwest, nonmetro areas in the Southeast, and small or medium metros in the Southwest and Southeast (Table 1). Supplemental Table S1 provides estimates of Covid-19 and excess mortality rates for each state in 2020 and 2021. Excess death rates were highest in Mississippi (301 deaths per 100,000 residents) followed by Arizona (246 deaths per 100,000 residents) in 2020 and in West Virginia (298 deaths per 100,000 residents) followed by Mississippi (271 deaths per 100,000 residents) in 2021.

Table 1.

Excess mortality, Covid-19 mortality, and the ratio of Covid-19 to excess mortality by metropolitan-nonmetropolitan status and BEA region

		2020					2021
		Number of Deaths			Rates per 100,000 PY		Number of Deaths			Rates per 100,000 PY
BEA Region	Metro Status	Excess	COVID	Ratio	Excess	COVID	Excess	COVID	Ratio	Excess	COVID
Total	Total	459,764	350,695	0.763	167	127	477,147	415,048	0.870	144	125
Total	Large Central	146,578	111,101	0.758	173	131	123,302	109,824	0.891	122	108
Total	Large Fringe	98,568	82,432	0.836	140	117	87,332	87,734	1.005	103	103
Total	Medium/Small	135,180	97,689	0.723	164	119	161,891	133,750	0.826	163	134
Total	Non Metro	79,438	59,473	0.749	207	155	104,622	83,740	0.800	227	182
Far West	Region Total	52,939	39,938	0.754	112	85	72,562	60,076	0.828	128	127
Far West	Large Central	32,389	24,486	0.756	128	97	37,827	32,883	0.869	126	109
Far West	Large Fringe	6,852	5,022	0.733	95	69	9,258	7,456	0.805	106	86
Far West	Medium/Small	11,853	8,935	0.754	97	73	19,858	15,831	0.797	136	108
Far West	Non Metro	1,845	1,495	0.810	71	58	5,619	3,906	0.695	179	124
Great Lakes	Region Total	73,063	54,660	0.748	186	139	61,766	56,302	0.912	131	143
Great Lakes	Large Central	22,117	16,654	0.753	204	154	13,113	12,153	0.927	101	94
Great Lakes	Large Fringe	17,662	13,111	0.742	165	123	15,597	13,644	0.875	121	106
Great Lakes	Medium/Small	19,546	13,919	0.712	182	130	17,686	16,211	0.917	138	126
Great Lakes	Non Metro	13,738	10,976	0.799	192	154	15,370	14,294	0.930	179	167
Mideast	Region Total	85,924	76,661	0.892	206	184	40,782	55,015	1.349	82	132
Mideast	Large Central	39,938	33,275	0.833	288	240	11,885	16,404	1.380	72	100
Mideast	Large Fringe	30,257	29,937	0.989	174	172	10,283	19,796	1.925	49	94
Mideast	Medium/Small	11,875	10,504	0.885	149	132	12,222	13,237	1.083	127	138
Mideast	Non Metro	3,854	2,945	0.764	157	120	6,392	5,578	0.873	217	189
New England	Region Total	15,752	18,016	1.144	125	144	5,839	11,269	1.930	39	90
New England	Large Central	3,739	4,012	1.073	191	205	709	1,750	2.468	30	75
New England	Large Fringe	5,295	6,731	1.271	116	148	1,208	3,852	3.189	22	71
New England	Medium/Small	5,846	6,579	1.125	128	144	2,187	4,254	1.945	40	77
New England	Non Metro	872	694	0.796	59	47	1,735	1,413	0.814	98	80
Plains	Region Total	29,544	24,705	0.836	164	137	21,150	21,814	1.031	98	121
Plains	Large Central	3,561	2,764	0.776	150	117	2,433	2,167	0.891	86	77
Plains	Large Fringe	5,968	4,578	0.767	142	109	4,712	4,645	0.986	93	92
Plains	Medium/Small	9,518	7,464	0.784	157	123	6,005	6,305	1.050	82	86
Plains	Non Metro	10,497	9,899	0.943	196	184	8,000	8,697	1.087	124	135
Rocky Mountain	Region Total	11,815	8,649	0.732	113	83	14,193	12,516	0.882	112	118
Rocky Mountain	Large Central	1,717	1,223	0.712	108	77	907	1,031	1.137	48	54
Rocky Mountain	Large Fringe	2,889	2,114	0.732	116	85	2,749	2,078	0.756	91	69
Rocky Mountain	Medium/Small	4,174	3,068	0.735	102	75	5,776	5,282	0.914	115	105
Rocky Mountain	Non Metro	3,035	2,244	0.739	132	98	4,761	4,125	0.866	171	148
Southeast	Region Total	124,271	81,065	0.652	175	114	175,879	130,234	0.740	205	182
Southeast	Large Central	19,341	12,753	0.659	148	98	24,800	18,120	0.731	159	116
Southeast	Large Fringe	23,284	16,312	0.701	129	91	33,283	26,869	0.807	153	123
Southeast	Medium/Small	48,893	30,717	0.628	177	111	72,189	52,697	0.730	216	158
Southeast	Non Metro	32,753	21,283	0.650	262	170	45,607	32,548	0.714	304	217
Southwest	Region Total	66,456	47,001	0.707	188	133	84,976	67,822	0.798	198	190
Southwest	Large Central	23,776	15,934	0.670	150	100	31,628	25,316	0.800	165	132
Southwest	Large Fringe	6,361	4,627	0.727	110	80	10,242	9,394	0.917	143	131
Southwest	Medium/Small	23,475	16,503	0.703	254	178	25,968	19,933	0.768	233	179
Southwest	Non Metro	12,844	9,937	0.774	288	223	17,138	13,179	0.769	320	246

Notes: Death rates were calculated by aggregating deaths and population over counties within each BEA region and metropolitan-non-metropolitan area. Estimates for 2020 correspond to the period March - December 2020. Estimates for 2021 refer to the period January to December, 2021.

Figure 1 shows excess death rates across all counties in the United States. Between 2020 and 2021, excess deaths shifted from the Northeast and Midwest to the South and to the West and from metro to nonmetro areas. In 2020, excess mortality was higher in metro counties in the Far West, Great Lakes, Mideast, and New England (Table 1). In 2021, however, excess mortality was higher in nonmetro areas than metro areas in these regions. In the Southeast, Southwest, Rocky Mountain, and Plains regions, excess mortality was higher in nonmetro areas in both 2020 and 2021. Nationally, excess mortality was higher in nonmetro areas than metro areas, and the disparity between nonmetro and metro areas was greater in 2021.

Figure 1.

County excess mortality rates per 100,000, 2020–2021

Notes: Panels A and B show the geographic distribution of excess death rates in 2020 (A) and 2021 (B) as estimated by comparing the expected number of deaths from our model to the actual number of deaths. Panels C and D report excess and COVID deaths rates for the counties with the highest excess deaths rates 2020 and 2021 respectively. Counties with less than 30,000 residents and less than 60 COVID deaths across the two years were excluded from the rankings in the barplots.

Supplemental Figure S1 shows actual and expected death rates for the U.S. by month during 2020 and 2021. Three peaks in mortality are apparent: (1) early 2020, (2) end of 2020 / start of 2021, and (3) end of 2021. Figure 2 breaks down trends in excess death rates within each BEA region by month throughout the period. Excess death rates peaked in the Mideast in early 2020, primarily in large metro areas. Excess death rates also increased markedly in New England and the Great Lakes during this time. Around the end of 2020, a second peak resulted in high excess death rates in the Far West, Great Lakes, Southwest, and Southeast regions. A third peak was observed in most of the regions near the end of 2021. Supplemental Figure S2 shows actual and expected death rates for the largest county in each BEA region - metro combination by month during 2020 and 2021. Some counties only experienced one peak in mortality (e.g. Kings County, New York), whereas others experienced three distinct peaks (e.g. Navajo County, Arizona).

Figure 2.

Monthly Excess Deaths Rates by BEA Region and Metro Status, 2019–2021

Notes: This graph shows aggregated trends in excess mortality at the monthly level between 2019–2021 stratified by BEA region and metro status.

In absolute terms, the Southeast was the region with the most excess deaths in both 2020 and 2021 followed by the Mideast and Great Lakes in 2020 and by the Southwest in 2021 (Figure 3). While New York was the state with the most excess deaths in 2020, Texas, California, and Florida had the most excess deaths in 2021. Excess mortality was also less concentrated in large metro areas and large fringe areas in 2021 than it was in 2020. Supplemental Figure S3 provides a comparison of states and regions in terms of excess death rates, rather than absolute counts.

Figure 3.

Excess Deaths by BEA Region and Metro Status, and by State, 2020–2021

Notes: The top panel presents absolute excess deaths by BEA region disaggregated by metro status (left) and state (right) for 2020; the bottom panel shows the same for 2021.

Figure 4 plots the proportion of excess deaths assigned to Covid-19 across counties in the United States. In both 2020 and 2021, counties across the country reported substantial numbers of excess deaths not assigned to Covid-19 as an underlying cause of death. In total, 76.3% of excess deaths were assigned to Covid-19 in 2020, whereas in 2021, 87.0% of excess deaths were assigned to Covid-19. This equals 109,069 excess deaths that were not assigned to Covid-19 in 2020 and 62,099 excess deaths that were not assigned to Covid-19 in 2021, for a total of 171,168 deaths. Despite the increase in assignment of Covid-19 deaths from 2020 to 2021, many regions still had areas with a low proportion of excess deaths assigned to Covid-19 during 2021, such as nonmetro areas in the Far West (69.5%) and Southeast (71.4%). The Southeast and Southwest were the regions with the lowest overall assignment of excess deaths to Covid-19 in 2021.

Figure 4.

Percentage of Excess Deaths not Assigned to Covid-19, 2020–2021

Notes: Panels A (2020) and B (2021) show the geographic distribution of proportion of excess deaths not assigned to COVID in 2020 and 2021. Panels C (2020) and D (2021) report excess and COVID deaths rates for the counties with the lowest COVID to excess ratios in 2020 and 2021. Counties with less than 30,000 residents and less than 60 COVID deaths across the two years were excluded from the rankings in the barplots.

In 2020, assignment of excess deaths to Covid-19 was similar in large central metro areas (76.3%) compared to nonmetro areas (74.9%), whereas in 2021, assignment was lower in nonmetro areas (80.0%) than large central metro areas (89.1%). Despite this overall trend, several regions (Mideast and New England) had lower assignment in nonmetro areas in 2020 than in metro areas. This suggests that assignment of excess deaths to Covid-19 in nonmetro areas was low in many regions across the study period. In 2021, assignment of excess deaths to Covid-19 was particularly low in nonmetro areas in the Southeast. Supplemental Figure S4 identifies the states with the lowest assignment of excess deaths to Covid-19. Many of these states were in the Southeast and included South Carolina, Louisiana, Mississippi, and North Carolina. In contrast, several states in New England including Massachusetts, New Jersey, Connecticut, and Rhode Island reported more Covid-19 deaths than excess deaths. Overall, few states reported more deaths assigned to Covid-19 than our estimates of excess deaths (5 states in 2020 and 14 states in 2021).

Supplemental Figure S5 plots excess death rates for each county against their Covid-19 death rates. Counties above the 45-degree reference line represent areas where the excess death rate was higher than the Covid-19 death rate, indicating there were excess deaths not assigned to Covid-19 in these counties. In 2020, the majority of counties in all regions except New England were above the 45-degree line, demonstrating that there were excess deaths not assigned to Covid-19 in these areas. Excess deaths not assigned to Covid-19 occurred in counties across the metro-nonmetro continuum. In 2021, the majority of counties in the Southeast, Southwest, Far West, and Rocky Mountains regions remained above the 45-degree line, demonstrating that they still had excess deaths not assigned to Covid-19. In particular, the counties with the highest proportion of excess deaths not assigned to Covid-19 tended to be nonmetro counties. Most counties in the Mideast, Plains, and Great Lakes, however, no longer had excess deaths not assigned to Covid-19 in 2021.

Discussion

In the present study, we produced estimates of excess deaths associated with the Covid-19 pandemic from March 2020 to December 2021 across 3,127 harmonized counties in the United States. Our study found that nearly 940,000 excess deaths occurred in the U.S. between 2020 and 2021, of which more than 170,000 were not assigned to Covid-19 on death certificates. This indicates that excess deaths were 22% higher than deaths assigned to Covid-19 during this period.

Prior studies of excess mortality have largely produced estimates for the year 2020 (3–6, 27), leaving patterns of excess mortality during 2021 under-studied. The Center for Disease Control and Prevention (CDC) however has reported a provisional estimate of approximately 944,000 excess deaths in the U.S. from March 2020 to December 2021, which is very close to our estimate (28). Woolf et al. identified 522,368 excess deaths from March 1, 2020 to January 2, 2021, which is higher than our estimate of approximately 470,000 deaths for the year 2020 (6). Islam et al. reported 458,000 excess deaths during 2020, which is close to our estimate (29). A prior estimate by Stokes et al. found 438,386 excess deaths in 2020, which is lower than our estimate due to differences in methods and time horizons for predicting expected deaths (4).

There are multiple potential advantages to using county-level data to generate estimates of excess mortality at the state and national levels. Predicting expected mortality at the state or national levels assumes that all areas within the state or nation have the same background trends of mortality. However, in actuality, different regions and metro and nonmetro counties of the U.S. have experienced varied long-term mortality trends (8, 30, 31). Our approach may produce more reliable estimates of expected mortality in the absence of the pandemic, because the projections build on historical trends at the county-level.

Another key advantage of producing excess mortality estimates at the county-level is the opportunity to examine differences across the urban-rural continuum, which is not possible with state-level data. By exploring how excess mortality varies by region and time across the urban-rural continuum, we gained several novel insights about patterns of mortality and the assignment of Covid-19 as an underlying cause on death certificates during the pandemic.

One major finding of this study is that there were similar numbers of excess deaths in 2020 and 2021, which is noteworthy as vaccinations were available for much of 2021. Despite the strong efficacy of vaccines, gaps in uptake likely contributed to high excess mortality in 2021, which may persist into the future if these vaccination gaps are not closed. This finding may also reflect federal and state governments’ decision to prioritize individual-level interventions over population-based strategies designed to protect the communities at greatest risk for Covid-19 death, such as financial support for family and medical leave, improved ventilation of schools and workplaces, and vaccine delivery programs organized in coordination with community partners (32).

We found substantial variation in excess mortality by US region and across the urban-rural continuum, which differed between 2020 and 2021. In the Far West, Great Lakes, Mideast, and New England, there was a substantial urban mortality disadvantage in 2020, which was reversed in 2021 to yield a rural mortality disadvantage. In the Southeast, Southwest, Rocky Mountain, and Plains regions, there was a rural mortality disadvantage in 2020, which was exacerbated in 2021. This suggests that the pandemic has impacted rural areas heavily, especially in 2021, suggesting a need for increased preventative measures in these areas where vaccination remains low (33).

Another finding of this study is that excess death rates exceeded Covid-19 death rates across most counties in all BEA regions during 2020 except New England. This indicates that there were excess deaths not assigned to Covid-19 reported in these regions during 2020. Between 2020 and 2021, Covid-19 and excess mortality then converged in some regions of the country (e.g. Mideast and Great Lakes). In other regions (e.g. Southeast and Southwest), excess death rates continued to exceed Covid-19 death rates in most counties, indicating that excess deaths not assigned to Covid-19 persisted. This finding may have several explanations. In the Mideast, the gap between Covid-19 and excess mortality in 2020 likely reflected the fact that the pandemic affected this region early in the pandemic when access to testing was extremely limited and the clinical manifestations of Covid-19 were unclear (27). The elimination of this gap by 2021 suggests that as the pandemic progressed, the Mideast was able to more effectively capture Covid-19 deaths. It is also possible that some indirect effects of the pandemic, such as disruptions in health care access, also decreased during this time. In other regions such as the Southeast and Southwest, the gaps between Covid-19 and excess mortality persisted into 2021, which may relate to the continued lack of Covid-19 testing in many Southeastern and Southwestern counties even as the pandemic progressed (34–36).

Excess deaths exceeded deaths assigned to Covid-19 in rural areas across many regions, especially in 2021. Many counties in the Southeast and Southwest also had low levels of assignment of excess deaths to Covid-19 throughout the pandemic. Rural counties may have under-counted Covid-19 deaths throughout the pandemic as a result of the exceptionally high death rates many rural areas faced during the Winter surge of 2020–2021 and the subsequent Delta surge in Summer/Fall 2021 (37–41). Other contributing factors may have included under-resourced health care systems that were unable to care for patients with Covid-19 and/or other non-Covid-19 conditions, under-resourced death investigation systems in which medical examiners or coroners did not pursue post-mortem Covid-19 testing, and partisan beliefs regarding the Covid-19 pandemic that may have influenced cause of death assignment and the likelihood of testing (19, 25).

Although our study does not distinguish between uncounted Covid-19 deaths and deaths indirectly related to the pandemic, emerging literature suggests that a large share of the excess deaths not assigned to Covid-19 likely represent uncounted Covid-19 deaths (10, 11, 42). For example, one recent study by Lee et al. found that approximately 90% of excess mortality between March 2020 and April 2021 could be attributed to the direct effects of SARS-CoV-2 infection (42). This possibility is also supported by investigative reporting during the pandemic which has documented widespread irregularities in cause of death assignment resulting from over-burdened and under-resourced death investigation systems (43). Discrepancies between Covid-19 death rates and excess death rates are problematic because they have the potential to mislead scientists and policymakers about which areas were most heavily affected during the pandemic. Failure to accurately capture Covid-19 deaths also points to an urgent need to modernize the death investigation system in the United States, including expanding budgets for medical examiner officers and eliminating the archaic coroner system (25).

In New England, we observed a different pattern around assignment of Covid-19 deaths than in the other BEA regions. In this region, a large share of counties had higher Covid-19 than excess death rates, a pattern that became more pronounced between 2020 and 2021. Several explanations may exist for this pattern including that other causes of death (i.e. influenza) declined in these areas or that the economically privileged status of many of these counties shielded their residents from the negative indirect effect of the pandemic by allowing them to work-from-home and avoid household crowding. Finally, it is possible that deaths were over-assigned to Covid-19 in these areas due to different cause of death assignment protocols for Covid-19. For example, until March 2022, Covid-19 deaths in the state of Massachusetts included any death that occurred within 60 days of a Covid-19 diagnosis, which differed from other states and guidelines from the Council of State and Territorial Epidemiologists that recommended states use a 30 day window (44).

The study had several limitations. First, the study relied on publicly available data, which were subject to suppression of death counts fewer than 10 in a given county-month. We addressed this limitation by pooling information across different geographical levels through the use of hierarchical models and by taking advantage of the additional information provided by yearly death counts, however, our estimates remain uncertain in areas with small populations and few deaths. Second, some counties and states have experienced prolonged reporting delays of Covid-19 deaths, which could affect our estimates of the proportion of excess deaths assigned to Covid-19, particularly in more recent months. Third, we were unable to distinguish between excess deaths that represented uncounted Covid-19 deaths and excess deaths indirectly related to the pandemic. Future research should examine this distinction to clarify the extent to which excess deaths not assigned to Covid-19 reported in this study represent under-reporting of Covid-19 deaths. Fourth, our study examined all-cause mortality and did not explore differences in trends using cause-specific death rates. Assessing geographic and temporal differences in excess death rates by cause-of-death would allow for a deeper understanding of the mechanisms driving trends in excess mortality overall and is an important direction for future work. Fifth, we used underlying cause of death data to identify deaths assigned to Covid-19 and thus did not identify deaths where Covid-19 was listed as a contributing cause. Finally, due to data limitations, our model does not account for differences in age structure between counties. Since the pandemic has affected older populations more significantly, some differences in mortality observed between counties may simply reflect differences in their age distribution.

In conclusion, the present study generated novel estimates of Covid-19 and excess mortality for 3,127 harmonized county units over the period from March 2020 to December 2021. In contrast to official Covid-19 death tallies, which are subject to differential underreporting and fail to capture indirect pandemic effects, the present estimates more fully account for the true toll of the pandemic across local areas and provide a more comparable measure of the Covid-19 mortality burden. As such, these estimates may be useful for additional work to investigate the determinants of excess mortality throughout the pandemic and may also be useful for communicating Covid-19 risks with local communities where the direct tallies have hidden the full extent of the pandemic’s consequences.

Materials & Methods

Yearly and monthly death counts at the county level were extracted from CDC WONDER online tool. See Methods Supplement for further details about data extraction procedures. We extracted death counts by all causes of death and from Covid-19. Causes of death were selected from the Multiple Cause of Death database using the provisional counts for 2020 and 2021 and the final counts for 2015–2019. A death was assigned to Covid-19 when Covid-19 was listed as the underlying cause of death using the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) code U07.1.[3] For all years except 2020, yearly deaths include all deaths between January to December. For 2020, yearly deaths include only deaths that occurred between March and December to better reflect the pandemic period. All death rates computed for 2020 were adjusted to account for the shorter exposure period.

To convert the number of deaths into rates, we used publicly available yearly county-level population estimates from the Census Bureau.[4] To compute monthly rates, we assumed linear growth between each two time points. For the August-December period in 2021, for which no population estimates are available, we assumed the population remained constant at its value in July.

We grouped counties into 4 metropolitan-nonmetropolitan categories (large central metro, large fringe metro, medium or small metro, nonmetro) based on the 2013 NCHS Rural-Urban Classification Scheme for Counties (45) and into 8 Bureau of Economic Analysis (BEA) regions (Far West, Great Lakes, Mideast, New England, Plains, Rocky Mountains, Southeast, Southwest) (30). Next, we stratified each region by each metropolitan- nonmetropolitan category, leading to 32 geographic units. We also grouped small counties into county-sets according to the United States Census Bureau’s County Sets classification. County-sets were used, in addition to counties, within our model to improve the precision of estimates for small counties in the study. See Methods Supplement for further details about the geographic classifications used in this study.

To model the monthly county-level number of deaths for 100,000 residents (DR), we estimated two different hierarchical linear models. The first model predicts monthly DR for a county directly while the second one predicts yearly DR for a county and then distributes the corresponding number of deaths over the year by month according to weights computed over the All-Causes of Death data at the national level. Both models were estimated using the lme4 package for the R language (46) and fitted using monthly mortality data for the period January 2015 - December 2019. The need for these two different models arises as a result of the suppression procedure applied in the CDC WONDER tool to all data points (county-months in our case) with fewer than 10 deaths. As a result of suppression, small counties that rarely exceed 10 deaths in a given month have very few data points, and these data points are not representative of normal mortality conditions (that is why they were not suppressed). In other words, data points are not missing at random and non-missing data points for small counties reflect higher-than-normal mortality. The yearly-level model, by making use of the additional information on the yearly number of deaths, leads to more accurate predictions in counties with a high proportion of missing data points. Combining these two models led to a better overall performance compared to using either one in isolation.

The monthly-level model expresses the monthly number of deaths for every 100,000 residents (DR) as a function of time (in years), month (with dummy variables), and an intercept allowed to vary across counties, county-sets, and states. To make the model more flexible, the time slope is also allowed to vary across county sets. Formally[5]:

Where:

In the equations above, the subscripts c, cs, s, m, and y indicate county, county-set, state, month, and year respectively. The capital letters C, CS, and S indicate that a term is specific to the county, county-set, and state level equations respectively. Using county-sets as an intermediate level between counties and states helped us overcome estimation difficulties with counties with few data points due to suppression. Even when the county level intercept cannot be estimated with precision, the estimate will be pulled toward the mean of counties in the same state and county-set, which we were able to estimate more precisely. The yearly-level model follows a structure similar to the monthly model but, due to the smaller sample size, it is less complex. Yearly death rates are modeled as a function of time (in years) and an intercept, both allowed to vary across counties.

Where:

In both models, the intercepts and the slopes are allowed to be correlated. To obtain the number of deaths, we multiplied the estimated death rate by the corresponding population. To obtain the monthly number of deaths from the annual number of deaths for the yearly model, we first computed the average proportion of deaths occurring in each month over the 2015–2019 period and then distributed annual deaths accordingly.

To decide whether to use the yearly model or the monthly model, we computed the average percentage difference between the predicted yearly deaths for the period 2015–2019 and the actual yearly deaths. We then used the estimate from the yearly model for all counties in which the difference was larger than 10%. Applying this decision rule, we used the yearly model for 790 counties and the monthly model for the remaining 2322 counties. Further details about the model section are provided in the Methods Supplement.

We obtained confidence intervals for the death rates by sampling from the distribution of the models’ fixed effects 1000 times and using these samples to compute 1000 different predicted rates. We then computed the 2.5 and 97.5 percentiles of the resulting distribution. The intervals thus obtained do not reflect all of the models’ uncertainty but only the portion due to fixed effects. However, they are consistent with the historical variability in the death rates and can be used to get a sense of the estimates’ variability.

This study used de-identified publicly available data and was exempted from review by the Boston University Medical Center Institutional Review Board. Programming code was developed using R, version 4.1.0 (R Project for Statistical Computing) and Python, version 3.7.13 (Python Software Foundation).