County-Level Social Determinants of Health and COVID-19 in Nursing Homes, United States, June 1, 2020–January 31, 2021

Abstract

Objectives:

Nursing homes are a primary setting of COVID-19 transmission and death, but research has primarily focused only on factors within nursing homes. We investigated the relationship between US nursing home–associated COVID-19 infection rates and county-level and nursing home attributes.

Methods:

We constructed panel data from the Centers for Medicare & Medicaid Services (CMS) minimum dataset, CMS nursing home data, 2010 US Census data, 5-year (2012-2016) American Community Survey estimates, and county COVID-19 infection rates. We analyzed COVID-19 data from June 1, 2020, through January 31, 2021, during 7 five-week periods. We used a maximum likelihood estimator, including an autoregressive term, to estimate effects and changes over time. We performed 3 model forms (basic, partial, and full) for analysis.

Results:

Nursing homes with nursing (0.005) and staff (0.002) shortages had high COVID-19 infection rates, and locally owned (−0.007) or state-owned (−0.025) and nonprofit (–0.011) agencies had lower COVID-19 infection rates than privately owned agencies. County-level COVID-19 infection rates corresponded with COVID-19 infection rates in nursing homes. Racial and ethnic minority groups had high nursing home–associated COVID-19 infection rates early in the study. High median annual personal income (−0.002) at the county level correlated with lower nursing home–associated COVID-19 infection rates.

Conclusions:

Communities with low rates of nursing home infections had access to more resources (eg, financial resources, staffing) and likely had better mitigation efforts in place earlier in the pandemic than nursing homes that had access to few resources and poor mitigation efforts. Future research should address the social and structural determinants of health that are leaving racial and ethnic minority populations and institutions such as nursing homes vulnerable during times of crises.

During the COVID-19 pandemic, the United States has had higher COVID-19 infection rates and deaths than the rest of the world. ¹ As of March 25, 2021, more than 125 million cases of COVID-19 had been confirmed worldwide. Of these, approximately 30 million confirmed cases and 540 000 deaths had occurred in the United States.^1,2 The COVID-19 pandemic may exacerbate existing health disparities rooted in structural systemic inequities and social determinants of health.^{3 -6} Institutional settings, such as prisons ⁷ and nursing homes, ⁸ and county-level demographic characteristics have been reported as substantial risks for COVID-19 cases and deaths. ⁹ One study also examined the influence of county-level socioeconomic factors on the potential for community-level spread of COVID-19 into neighboring counties. ¹⁰

Although socioeconomic factors, regardless of race and ethnicity, are a major risk factor for COVID-19 transmission, researchers have highlighted systemic racial and ethnic disparities as critical to understanding how COVID-19 is affecting the United States.^5,9 McLaren ¹¹ found that racial and ethnic minority populations accounted for the highest rates of COVID-19 cases and deaths in the United States, after controlling for socioeconomic status, health care access, and occupational characteristics. Other research has attributed high rates of COVID-19 cases and deaths among racial and ethnic minority populations to neighborhood living conditions and limited capacity to socially distance.^12,13 Large cities on the East and West Coasts were most vulnerable to higher COVID-19 infection rates early in the pandemic because of high-density cities and travel patterns; however, as time progressed, southern states, which have a history of strained race relations, had a rapid uptick in cases and deaths, as did rural areas.^9,14,15 The southern states have been less likely than many other states to follow social distancing recommendations, and racial and ethnic minority populations are facing the brunt of infections and deaths. Many southern states and communities have limited health care and resource capacities, low rates of health insurance coverage, and less supportive social policies to mitigate the effects.^9,14,15

During the early months of the pandemic, nursing homes were described as “ground zero” ¹⁶ and “death pits” ¹⁷ in relation to COVID-19 cases and deaths. From the beginning of July 2020 until the beginning of March 2021, the number of COVID-19 cases among residents in nursing homes was stable at about 10 000 cases per week until a peak period from November 2020 to January 2021, when the number of cases was >30 000 per week; in addition, this period included a peak of >6000 deaths per week from early December 2020 to early February 2021. ¹⁸ Nursing home residents are vulnerable because they typically have multiple comorbid conditions, live in small rooms with limited space between resident beds, and require assistance with activities of daily living. ¹⁹ Furthermore, staff members provide hands-on care and services for many residents, often not knowing whether a resident has COVID-19. Staff member movement among residents increases the probability of transmission resulting from the asymptomatic spread of COVID-19. ²⁰ A study early in the pandemic at a single nursing home in the United States found that 74% of residents and half of onsite staff members were asymptomatic. ²¹ Early studies of COVID-19 cases and deaths in nursing homes indicated that nursing homes that had more beds, were located in urban areas, ²² had few registered nurses on staff, ²³ and served a high percentage of Medicaid-insured and Black/African American residents ²⁴ were most likely to have COVID-19 cases or deaths. ²⁵

Staffing challenges such as low pay, difficult work environments, negative job perception, limited autonomy, and organizational climate have been problematic at nursing homes for decades. ²⁶ Because of low wages, staff members often work in multiple nursing homes, thus increasing transmission of COVID-19 between facilities. Nursing assistants, the direct care workers who provide most of the hands-on care in nursing homes, are predominantly single women of color, have low levels of education, and have a median annual income of approximately $23 300. ²⁷

Insufficient nurse staffing in nursing homes is related to adverse events, substandard care, inadequate monitoring, and poor quality of care and resident outcomes. ²⁸ Studies have found increases in the quality of care in nursing homes with more nurses on staff, while accounting for resident acuity. ²⁹ Personal protective equipment (PPE) was in short supply at nursing homes during April and May 2020 in the early stages of the pandemic; internationally and in the United States, supplies were limited because of disruptions in the supply chain and prioritization of hospitals to receive PPE, making it difficult for nursing homes to obtain PPE for staff members.^30,31

Long-standing systemic racial inequities likely contribute to nursing home–related COVID-19 disparities. Socioeconomic and racial disparities in nursing homes are well documented; specifically, nursing homes that serve a higher proportion of Black residents have fewer nurses and staff members, have more deficiencies, are located in poorer counties, and are reimbursed primarily by Medicaid compared with nursing homes serving chiefly White residents. ³² Two studies have highlighted racial and ethnic differences in COVID-19 deaths among nursing homes with higher proportions of non-White residents when compared with those having the highest proportions of White residents.^33,34 Black populations and the communities in which they reside are also likely to have higher rates of COVID-19 hospitalization and mortality than White populations and communities.^35,36

With the COVID-19 pandemic disproportionately affecting the United States and certain populations and settings, we investigated associations among nursing home COVID-19 infection rates, characteristics of nursing homes, and county-level demographic characteristics. We further explored these relationships during an extended period to analyze trends. Based on previous research on COVID-19 and nursing homes and county-level factors,^37,38 we hypothesized that infection rates would differ statistically based on nursing home characteristics and community characteristics and would vary over time.

Methods

Data Sources

We constructed panel data from the Centers for Medicare & Medicaid Services (CMS) minimum dataset, ³⁹ CMS nursing home data, ⁴⁰ 2010 US Census data, ⁴¹ 5-year estimates from the 2012-2016 American Community Survey, ⁴² and county-level COVID-19 infection rates. ⁴³ Facility-level data (quality rating, survey rating, staffing rating, overall rating, number of monetary fines, and total sum of fines) came from the 2019 CMS nursing home data to account for effects from the COVID-19 pandemic. Nursing homes must maintain substantial compliance with the Medicare and Medicaid conditions of participation. When a nursing home does not maintain compliance, CMS may use various enforcement mandates to include a civil monetary penalty or fine. ⁴⁴ CMS facility data use a 5-star rating system, developed in 2008, with 5 being the best and 1 being the worst. The overall quality rating comprises the health inspections rating, staffing rating, and quality measures rating, with the health inspections rating as the base. ⁴⁵ We also used information from the 2019 CMS minimum dataset on the type of ownership, change of ownership in previous 12 months, and total number of residents (as a measure of facility size).

In May 2020, CMS began collecting COVID-19 infection data among residents and staff members of nursing homes. Because of irregularities in the first 2 weeks of data, we used COVID-19 infection rate data from June 1, 2020, through January 31, 2021. We aggregated the data into 7 five-week segments, allowing for smoothing of weekly spikes and for analysis of trends among various community types, based on demographic profiles. Variables included the total resident infection rate and the number of weeks during which nursing staff shortages occurred, as well as clinical staff, nurse aids, and other staff shortages. County-level data from the US Census and the American Community Survey included population per square mile, median annual personal income, percentage of the population aged >65, and racial and ethnic composition. We used the number of county-level COVID-19 infections to calculate county infection rates.

A total of 15 454 nursing homes are in the United States and were originally considered for analysis. For our study, we excluded all nursing homes identified as owned by a hospital and any special-focus facilities (ie, those with a poor record of performance on recent health inspections) because they may have had different pressures and access to resources during the pandemic than government-owned, nonprofit, and private agencies not identified as special-focus facilities. In addition, previous research concluded that hospital-owned nursing homes have a substantially different operating environment and strategic behavior than freestanding public and private facilities. ⁴⁶ We excluded 5464 facilities from analysis. The final dataset included 69 891 observations from 7 five-week periods across 9990 (64.6%) nursing homes in the United States.

Statistical Analyses

Using COVID-19 infection rates internal to nursing homes (or facility infection rate) as the dependent variable, we found a nonnormal distribution (model residuals) exhibiting autocorrelation (Table 1). Before settling on the final analysis, we ran initial analyses to determine the need for autocorrelation correction. With previous experience using these variables, we felt that collinearity justified analysis of the base, partial, and full models. This approach allowed for examining whether any instability in estimation due to collinearity existed. A Durbin–Watson test suggested a second-order lag of autocorrelation. Therefore, we selected a maximum likelihood estimator and included an autoregressive term. We also included a lagged measure of facility infection rate in the analysis; including this lag omitted the initial period of the panel data. A lagged dependent variable (ie, COVID-19 infection rates) considers the correlation between previous observations of the dependent variable and the current observation; in addition, serially correlated error terms can be a source of heteroscedasticity, which can reduce the efficiency of the ordinary least squares estimates and bias estimates of SE. However, inclusion of the lagged measure did not substantially change the estimated effects or estimated significance of these effects in the rest of the model. With testing of hypotheses focused on the effects of race and ethnicity and county demographic characteristics over time, and because the initial period was judged to contain essential data, we retained the model with no lagged measure of the facility infection rate. We also assessed the sum of squared errors (SSE), the Akaike information criterion (AIC), and the R² statistics across the maximum likelihood estimation models to interpret the effects of the variables included (Table 1).

Table 1.

Fit comparisons for the base, partial, and full models in a study on county-level social determinants of health and COVID-19 infection in nursing homes during 7 periods, United States, June 1, 2020–January 31, 2021 ^a

Fit statistics	Base model	Partial model	Full model
Sum of squares	1917.696	1883.378	1880.022
Mean sum error	0.028	0.027	0.027
Log likelihood	26 313.21	26 942.46	27 000.64
Durbin–Watson	2.009	2.007	2.007
Akaike information criterion (AIC)	−52 556.4	−53 766.9	−53 855.3
Weighted AIC	−52 556.4	−53 766.8	−53 855.1
Transformed regression R2	0.080	0.097	0.099
Total R2	0.088	0.104	0.106
No. of observations	69 696	69 696	69 696

^aThe base model included nursing home attributes, staffing, COVID-19 infection rates, and county demographic characteristics. The partial model allowed for effects related to race and ethnicity to vary over time. The full model allowed for effects of average age, county density, and median annual personal income to vary over time.

We ran 3 forms of the model. The base model included nursing home attributes, staffing, COVID-19 infection rates, and county demographic characteristics. We modeled the effects of time on facility infection rate using binary variables to allow for nonlinear effects. The binary variables allowed us to model nonlinear effects of time. For example, in the constructed panel data, we used several data points for each of the 7 periods. Instead of assigning each period a number (0-6) and using that number as an explanatory variable, each period was given a variable (eg, period 1). That variable is then set to 1 if that observation is in that period and to 0 if it is not in that period. This approach allows the regression to provide an estimate of the average infection rate of each period without assuming a linear (or any other constrained geometrical) relationship between time and infection. The base model examined for constant effects through time in the panel.

The partial model allowed for effects related to race and ethnicity to vary over time. The full model allowed for effects of average age, county density, and median annual personal income to vary over time. The results of all 3 models allowed for assessing the differences among them. The estimates of change in infection rates during each period in relation to the percentage of the county-level population aged >65, county population per square mile, and county median annual personal income are not captured in the partial model. We performed statistical analyses, including t tests to compare the total US sample of nursing homes and the sample that we used and maximum likelihood estimation, using SAS version 9.4 (SAS Institute, Inc). For the t tests and maximum likelihood estimation tests, we considered P ≤ .05 to be significant. We used ArcGIS Pro 2.5 (Esri) to create the map. The Appalachian State University Institutional Review Board reviewed this study and considered it exempt because we used previously de-identified data.

Results

Using the SSE, we found greater improvement between the base model and the partial model (1917.696 vs 1883.378) than between the partial model and the full model (1883.378 vs 1880.022) (Table 1). Similar outcomes were found in the R² (0.088 vs 0.104 and 0.104 vs 0.106, respectively) statistics. Therefore, we favored the partial model in the interpretation of the maximum likelihood estimates of COVID-19 infection rates in nursing homes.

We found high rates of nursing home–associated COVID-19 cases from June 1, 2020, through January 31, 2021, across the southern and central region of the United States (Figure). The sample of 9990 facilities did not differ substantially from the total number of US nursing homes in staffing and quality rating, ownership, and deficiencies; however, we found significant differences in the number of residents and overall rating (Table 2).

Figure.

COVID-19 cases among residents in nursing homes, regardless of age, United States, June 1, 2020–January 31, 2021. Nursing home—level COVID-19 cases were aggregated at the county level for visualization normalized by the county-level population aged ≥85.

Table 2.

Characteristics of nursing homes and county-level demographic characteristics, United States, 2019 ^a

Variable	Sample	United States	P value b
Nursing home variables
No. of nursing homes	9990	15 454
No. of residents (average)	88.22 (51.17)	85.80 (52.80)	<.001
Type of ownership, % (SD)
For profit	70.84 (45.45)	69.90 (45.90)	<.001
Local government	5.46 (22.71)	5.30 (22.40)	<.001
State government	0.97 (9.81)	1.10 (10.40)	<.001
Federal government	0.09 (3.00)	0.10 (3.20)	.001
Nonprofit	22.64 (41.85)	23.50 (42.40)	<.001
Total no. of fines (average per facility) c	0.49 (0.83)	0.47 (0.83)	.14
Total fines, $ (average per facility) c	15 534.00 (52 956.08)	15 196.19 (52 280.89)	.62
Overall rating (average per facility) d	3.09 (1.42)	3.13 (1.42)	.02
Survey rating (average per facility) d	2.78 (1.27)	2.81 (1.28)	.07
Quality rating (average per facility) d	3.61 (1.26)	3.63 (1.25)	.14
Staffing rating (average per facility) d	2.85 (1.20)	2.91 (1.22)	<.001
Health deficiencies (average per facility)	6.67 (5.61)	6.68 (5.66)	.92
Fire deficiencies (average per facility)	3.71 (4.24)	3.72 (4.28)	.89
County variables
No. of counties with nursing homes	2411	2987
Median annual personal income, $ (SD)	23 547.18 (5370.13)	23 277.01 (5371.05)	.07
% County aged >65 (average)	0.16 (0.04)	0.16 (0.04)	.07
Race, average % (SD) of county
Non-Hispanic White	0.83 (0.16)	0.83 (0.16)	.73
Non-Hispanic Black	0.09 (0.14)	0.09 (0.15)	.53
Non-Hispanic Asian	0.01 (0.02)	0.01 (0.05)	.40
Hispanic	0.08 (0.13)	0.08 (0.13)	.80
Non-Hispanic Other e	0.06 (0.07)	0.07 (0.18)	.02

^aData sources: Nursing home data came from the Centers for Medicare & Medicaid Services. ³⁹ County data came from the 2010 US Census ⁴¹ and the 2012-2016 American Community Survey. ⁴²

^bDifferences, using t tests, between the sample and US nursing home totals were significant at α = .05.

^cCenters for Medicare & Medicaid Services Nursing Home Penalty data. ⁴⁰

^dNursing home 5-star quality rating from Centers for Medicare & Medicaid Services. Quality rating ranges from 1 to 5, with 5 being the best and 1 being the worst. ⁴⁵

^e“Other” includes American Indian/Alaska Native, Native Hawaiian/Other Pacific Islander, other race, or ≥2 races.

Of 9990 nursing homes, most (n = 7703, 70.8%) were for-profit agencies (Table 2). Nursing homes on average had <1 fine (0.5) in 2019; however, the average total monetary value of the fines was >$15 000. Of the star rating scales, the quality rating had the highest average (3.6) and the survey rating had the lowest average (2.8). Nursing homes averaged 6.7 health deficiencies and 3.7 fire deficiencies during 2019.

We found significant associations between nursing home–associated COVID-19 infection rates and nursing home resident population size, nursing home ownership, staffing rating, weeks with nursing and staffing shortage, and county-level COVID-19 infection rates. Nursing homes with more residents tended to have lower COVID-19 infection rates (−0.000) than nursing homes with fewer residents. Local-owned (–0.007) or state-owned (−0.025) nursing homes and nonprofit agencies (−0.011) had lower COVID-19 infection rates than for-profit nursing homes. Nursing homes receiving higher staff ratings had lower COVID-19 infection rates (−0.005) than nursing homes receiving lower staff ratings. Nursing homes that had more weeks with shortages of nursing (0.005), nurses’ aides (0.001), and other staff members (0.002) had higher COVID-19 infection rates than nursing homes with adequate staffing. An increase in county-level COVID-19 infection rates (0.003) corresponded with higher rates of nursing home–associated COVID-19 cases.

The highest nursing home–associated COVID-19 infection rates occurred during periods 4 and 5 (late October–December 31, 2020) (Table 3). COVID-19 infection rates at nursing homes declined steadily in counties with a higher percentage of Black, Asian, and Hispanic populations from period 2 to period 6 (beginning in mid–July and forward) compared with counties with a higher percentage of White populations; during the earliest period (June and earlier) of the COVID-19 pandemic, racial and ethnic minority populations were the hardest hit. However, compared with counties that had a higher percentage of White people, counties that had a higher percentage of people from other races (ie, multiple races, American Indian/Alaska Native, Native Hawaiian/Other Pacific Islander) had a gradual increase in nursing home–associated COVID-19 infection rates, peaking in period 5. Counties with higher median annual personal incomes and a higher percentage of people aged ≥65 had lower COVID-19 infection rates than counties that had lower median annual personal incomes and populations with lower percentages of people aged >65. Less dense, or more rural, counties were more affected by nursing home–associated COVID-19 cases as time progressed, when compared with more urban and populated counties. However, the estimate for period 4 was slightly positive (0.014), indicating an increased rate among counties with higher population density when compared with counties with lower population density.

Table 3.

Maximum likelihood estimates of COVID-19 infection rates in nursing homes in relation to facility characteristics, county-level COVID-19 infection rates, and nursing home COVID-19 infection rates during 7 five-week periods ^a in relation to racial and ethnic minority population at the county level (2010 US Census), county-level proportion aged >65, county-level population density, and county-level median annual personal income, United States ^b

Variable	Base model c		Partial model c		Full model c
	Estimate	P value d	Estimate	P value d	Estimate	P value d
Intercept	0.086	<.001	0.064	<.001	0.008	<.001
Total no. of residents	0	.01	0	.01	0	.01
Ownership e
Change of ownership	0	.99	0	.99	0	.98
Local government	−0.007	.01	−0.007	.01	−0.007	.01
State government	−0.025	<.001	−0.025	<.001	−0.025	<.001
Federal government	−0.004	.86	−0.005	.82	−0.005	.82
Nonprofit	−0.011	<.001	−0.011	<.001	−0.011	<.001
Fines
No. of fines	−0.002	.02	−0.002	.01	−0.002	.01
Total fines, $ f	2.51E-08	.06	2.49E-08	.06	2.48E-08	.06
Ratings
Overall	0.003	.02	0.003	.02	0.003	.02
Survey	−0.001	.59	−0.001	.53	−0.001	.51
Quality	−0.001	.08	−0.001	.06	−0.001	.06
Staffing	−0.005	<.001	−0.005	<.001	−0.005	<.001
Health deficiencies	0	.54	0	.60	0	.65
Fire deficiencies	0	.93	0	.83	0	.81
No. of weeks with staff shortage
Nurses	0.005	<.001	0.005	<.001	0.005	<.001
Other clinical	0.001	.63	0.001	.51	0.001	.50
Nurses’ aids	0.002	.01	0.001	.05	0.001	.05
Other staff members	0.003	<.001	0.002	<.001	0.002	<.001
County COVID-19 infection rate g	0.004	.01	0.003	.03	0.002	.12
Period 1	0.019	<.001	0.015	<.001	0.045	.02
Period 2	0.013	<.001	0.026	<.001	0.080	<.001
Period 3	0.016	<.001	0.053	<.001	0.107	<.001
Period 4	0.075	<.001	0.143	<.001	0.257	<.001
Period 5	0.130	<.001	0.179	<.001	0.294	<.001
Period 6	0.053	<.001	0.067	<.001	0.093	<.001
% County non-Hispanic Blackg,h	−0.040	<.001	0.052	<.001	0.062	<.001
Period 1			0.044	.02	0.052	.01
Period 2			0.003	.89	0	.99
Period 3			−0.119	<.001	−0.131	<.001
Period 4			−0.268	<.001	−0.303	<.001
Period 5			−0.248	<.001	−0.271	<.001
Period 6			−0.044	.02	−0.042	.04
% County non-Hispanic Asiang,h	0.054	.01	0.281	<.001	0.118	.02
Period 1			−0.248	<.001	−0.072	.32
Period 2			−0.252	<.001	−0.047	.54
Period 3			−0.323	<.001	−0.160	.03
Period 4			−0.571	<.001	−0.331	<.001
Period 5			−0.190	.002	0.064	.39
Period 6			−0.017	.77	0.068	.35
% County Hispanicg,h	−0.001	.77	0.059	<.001	0.064	<.001
Period 1			0.155	<.001	0.157	<.001
Period 2			−0.032	.18	−0.033	.17
Period 3			−0.111	<.001	−0.115	<.001
Period 4			−0.175	<.001	−0.189	<.001
Period 5			–0.216	<.001	−0.229	<.001
Period 6			−0.042	.07	−0.043	.07
% County non-Hispanic other racesg,h	0.005	.68	−0.006	.87	0.035	.34
Period 1			−0.146	.004	−0.171	<.001
Period 2			0.003	.95	−0.038	.47
Period 3			0.013	.79	−0.028	.59
Period 4			0.033	.51	−0.048	.35
Period 5			0.164	<.001	0.088	.09
Period 6			0.008	.87	−0.007	.90
% County aged >65	−0.044	.02	−0.047	.01	0.007	.90
Period 1					0.048	.51
Period 2					0.011	.88
Period 3					0.012	.87
Period 4					−0.125	.08
Period 5					−0.198	.01
Period 6					0.036	.62
County population density (10 000 residents  per square mile)	−0.015	<.001	–0.047	<.001	−0.009	.07
Period 1					−0.019	.01
Period 2					−0.010	.19
Period 3					0.006	.38
Period 4					0.014	.05
Period 5					−0.013	.08
Period 6					−0.016	.03
County median annual personal income, in $10 000s	−0.002	<.001	−0.002	<.001	0	.29
Period 1					−0.002	<.001
Period 2					−0.002	<.001
Period 3					−0.002	<.001
Period 4					−0.003	<.001
Period 5					−0.003	<.001
Period 6					−0.001	.12
Autoregressive model 1 i	−0.002	.57	0.009	.02	0.009	.01
Autoregressive model 2 i	0.075	<.001	0.075	<.001	0.074	<.001

^aPeriod 0 (June 1–July 5, 2020); period 1 (July 6–August 9, 2020); period 2 (August 10–September 13, 2020); period 3 (September 14–October 18, 2020); period 4 (October 19–November 22, 2020); period 5 (November 23–December 27, 2020); period 6 (December 28, 2020–January 31, 2021).

^bData sources: Centers for Medicare & Medicaid Services, ³⁹ USAFacts.org, ⁴³ Centers for Medicare & Medicaid Services nursing home data, ⁴⁰ 2010 US Census, ⁴¹ and 2012-2016 American Community Survey, 5-year estimates. ⁴²

^cThe base model included nursing home attributes, staffing, COVID-19 infection rates, and county demographic characteristics. The partial model allowed for effects related to race and ethnicity to vary over time. The full model allowed for effects of average age, county density, and median annual personal income to vary over time.

^dMaximum likelihood estimates considered significant at α = .05.

^eThe reference group is for-profit ownership.

^fA $1 increase in fines is a 0.00000000251 increase in the infection rate. Or, a $10 000 increase would be associated with an increase in the infection rate of 0.000251.

^gThe time period reference group is period 0 (June 1–July 5, 2020).

^hThe reference group is non-Hispanic White.

ⁱThe dependent variable (nursing home COVID-19 infection rate) was measured 7 times (5-week increments) for each facility. The average infection rate across all periods was 0.075 (SD = 0.183) infections per resident. However, this average includes multiple observations of individual facilities.

Discussion

Our study examined associations between COVID-19 infection rates at nursing homes and nursing home and county demographic characteristics from June 1, 2020, through January 31, 2021, when vaccinations were beginning to be administered (nursing home vaccinations began in December 2020). In addition, we examined 7 five-week periods across the larger period. Our study found an association between nursing home attributes—type of ownership, staff rating, and weeks with nursing and other staff shortages—and increased COVID-19 infection rates. This finding suggests that local- or state-owned facilities and nonprofit facilities were more protective against COVID-19 infection than for-profit facilities, and that having adequate and skilled staff members was a key to limiting the number of COVID-19 cases. The for-profit ownership type was a significant predictor of COVID-19 infections in a recently published review. ⁴⁷ In our study, overall star rating contributed to increased COVID-19 infection rates across all models. The overall score was based heavily on the survey score and may reflect a nursing home culture of providing high-quality care, which means staff members interact with the residents frequently and provide more opportunities for the virus to spread than in agencies with less interactions between staff members and residents.

High COVID-19 infection rates at the county level also corresponded with increased COVID-19 infection rates in nursing homes. Frontline nursing assistants, who provide direct care to patients, are often in racial and ethnic minority groups and work multiple jobs, which often lack health benefits, paid time off, and a supportive work environment. ⁴⁸ Nursing assistants and other staff members may contribute to the spread of COVID-19 in nursing homes because of their interactions with other networks in the community that have a high risk of contracting COVID-19. ⁴⁹ Further evidence supports that racial disparities persist in protections found in the occupational context, in that racial and ethnic minority populations disproportionately work in jobs (eg, meatpacking and agricultural work, construction work, nursing homes, prisons) with elevated exposure to COVID-19. ⁵⁰ These settings at high risk of COVID-19 transmission often lack proper PPE, are unable to support physical distancing guidelines, and lack sufficient sick-leave support. Shippee et al ⁵¹ argued early in the pandemic for more resources to be directed toward staffing needs in low-income communities with a high percentage of racial and ethnic minority populations to address existing disparities. Racial and ethnic disparities were found in nursing home–associated COVID-19 cases in the early months of the pandemic. ⁵²

Our results imply that after the initial periods, when large urban areas with a high percentage of racial and ethnic minority populations were hit the hardest, ⁵³ nursing homes located in more rural and less populated areas had an increase in COVID-19 infection rates. The highest COVID-19 infection rates shifted from states in the Northeast in the earliest months to states in the South and Midwest, along with Southern California. At the same time, however, communities with higher proportions of American Indian/Alaska Native and Native Hawaiian/Other Pacific Islander populations also had higher nursing home–associated COVID-19 infection rates than communities that were predominantly White. Counties with a high median annual personal income and a high proportion of people aged ≥65 continued to be more protected against nursing home–associated COVID-19 infection rates than counties with a low median annual personal income and a low proportion of people aged ≥65.

It is likely that communities with low rates of nursing home–associated COVID-19 infections have more resources and that their state is using increased mitigation strategies to protect the older adult population and nursing homes. Li et al ⁵⁴ reported an association between state-level restrictions and COVID-19 case rates in nursing homes. Abouk and Heydari ⁵⁵ also reported on the significant effects of state social distancing policies on reducing adverse COVID-19 outcomes. Another recent study revealed political inclination predicted the COVID-19 case fatality ratio in rural counties, particularly in the Midwest, and that American Indian/Alaska Native and Asian/Pacific Islander populations were most vulnerable to COVID-19 cases and deaths. ⁵⁶ The importance of political values is associated with individual behaviors of citizens, ⁵⁷ state- and county-level policy responses,^{58 -60} and, ultimately, intention and willingness to vaccinate.^{59 -61} Future studies should consider political ideology when examining COVID-19 outcomes and pandemic responses.

Strengths and Limitations

This study had several limitations. First, because the COVID-19 pandemic is evolving, new data are available daily and can create challenges, such as inaccurate data, underreporting, and nursing home residents dying in hospitals rather than in nursing homes, for researchers and analysts. Second, in merging the datasets, we had to use county-level data before COVID-19 began; however, by incorporating nursing home data, we overcame some scale limitations and potential ecological fallacy. Strengths of the study were the integration of multiple data sources and scales (county-level and nursing homes), the longitudinal nature, and the entire United States being represented.

Conclusion

Two complex public health issues are occurring simultaneously in the United States: COVID-19 and systemic racism and structural inequities.^{62 -66} In the United States and globally, nursing homes, which house many older adults in underserved areas (ie, areas with limited resources and capacity), ⁶⁷ were identified by scholars ¹⁶ and journalists ¹⁷ early in the pandemic as hotspots of COVID-19 infection. In our study, for-profit nursing homes, nursing homes with low staff ratings, and nursing homes with nursing and staff member shortages were associated with higher COVID-19 infection rates during all 7 periods analyzed than government-owned and nonprofit nursing homes and nursing homes with high staff ratings and adequate numbers of nurses and staff members. Similar findings of poor quality of care and poor resident outcomes have been reported in nursing home quality studies throughout the past decades, where quality ratings are low in for-profit nursing homes, facilities with low staff ratings, and facilities with high staff shortages.^{68 -72} An article published in 2021 argued that the ownership structures (ie, for-profit, government-owned, nonprofit) should be scrutinized and reporting guidelines should be updated because of a lack of financial transparency. ⁷³ Our study findings indicate a particular need to examine for-profit agencies. Our study highlights that surrounding county-level inequities, largely counties with a high percentage of racial/ethnic minority populations and low income levels, correlated with increased risks for COVID-19 cases found in nursing homes, when also accounting for nursing home characteristics. This finding was particularly apparent in the early period of the pandemic; as time progressed during the study period, COVID-19 infection rates expanded to more socially vulnerable rural areas, ⁷⁴ and other racial and ethnic minority populations were affected. It is evident that nursing homes with a shortage of resources (financial resources, PPE), capacity (nurses, nurses’ aides), and staffing (administration and other frontline workers) have been vulnerable to COVID-19 infection.^19,75,76 The inequities in vulnerability are most likely the result of racial and ethnic disparities, population density (rurality), low median annual personal income levels, and other social determinants of health. In the short term, public health research and intervention should focus on preventing and mitigating outbreaks of infectious disease (ie, COVID-19) in nursing homes located in predominantly racial and ethnic minority communities; however, attention needs to be directed toward addressing the social and structural determinants of health that leave racial and ethnic minority populations more socially vulnerable at all times, and particularly during times of crises, such as in a global pandemic.^77,78