Vaccination, human mobility, and COVID-19 health outcomes: Empirical comparison before and during the outbreak of SARS‐Cov-2 B.1.1.529 (Omicron) variant

Abstract

The B.1.1.529 (Omicron) variant surge has raised concerns about the effectiveness of vaccines and the impact of imprudent reopening. Leveraging over two years of county-level COVID-19 data in the US, this study aims to investigate relationships among vaccination, human mobility, and COVID-19 health outcomes (assessed via case rate and case-fatality rate), controlling for socioeconomic, demographic, racial/ethnic, and partisan factors. A set of cross-sectional models was first fitted to empirically compare disparities in COVID-19 health outcomes before and during the Omicron surge. Then, time-varying mediation analyses were employed to delineate how the effects of vaccine and mobility on COVID-19 health outcomes vary over time. Results showed that vaccine effectiveness against case rate lost significance during the Omicron surge, while its effectiveness against case-fatality rate remained significant throughout the pandemic. We also documented salient structural inequalities in COVID-19-related outcomes, with disadvantaged populations consistently bearing a larger brunt of case and death tolls, regardless of high vaccination rates. Last, findings revealed that mobility presented a significantly positive relationship with case rates during each wave of variant outbreak. Mobility substantially mediated the direct effect from vaccination to case rate, leading to a 10.276 % (95 % CI: 6.257, 14.294) decrease in vaccine effectiveness on average. Altogether, our study implies that sole reliance on vaccination to halt COVID-19 needs to be re-examined. Well-resourced and coordinated efforts to enhance vaccine effectiveness, mitigate health disparity and selectively loosen non-pharmaceutical interventions are essential to bringing the pandemic to an end.

1. Introduction

Recently, the B.1.1.529 (Omicron) variant was identified on November 14, 2021 in South Africa and rapidly became the dominant strain of the Coronavirus disease 2019 (COVID-19) pandemic circulating in the world. Preliminary evidence suggests more transmissibility and an increased risk of reinfection with this variant, as compared to the original virus and other variants of concern (VOCs). Safe and effective vaccines are widely acknowledged to be a crucial solution in bringing the pandemic to an end, as vaccines reduce disease severity and curb virus transmission. As of May 2021, three COVID-19 vaccines have been authorized, including two mRNA vaccines (mRNA-1273 (Moderna) and BNT162b2 (Pfizer/BioNTech)) and one viral vector vaccine (AD26.COV2.S (Johnson & Johnson’s Janssen)). By November 30, 2021, the day before the first appearance of the Omicron variant in the US, over 70.8 % of US residents have received at least one dose of the vaccine [11].

The relatively high vaccination rate, however, failed to prevent surges of cases from the Omicron variant. Although Phase III clinical trials have announced strong vaccines efficacy, for example, 94.0–95.0 % efficacy against symptomatic infection and 100 % efficacy in preventing severe-critical disease [3], [40], there is increasing skepticism toward the real-world effectiveness of vaccines [32]. Researchers claimed that clinical trials might overestimate the level of vaccine protection compared to real-world settings since clinical trials have enrolled mostly younger, healthy adults and lacked sufficient evidence against new variants [41]. In addition, with insufficient vaccination coverage and less meticulous control over logistics and vaccine administration, vaccine effectiveness may be further attenuated [52]. Emerging studies have reported lower vaccine effectiveness compared to clinical vaccine efficacy. For example, a retrospective cohort study using vaccination data in Israel indicated that a single-dose vaccine against new infections is 51 % effective 13–24 days after immunization [13]. Moreover, signs of waning vaccine effectiveness over time and new variants escaping vaccine protection have been observed [5], [18], [50]. One study stated that protection against infections from the Omicron variant following two doses of BNT162b2 dropped from 65.5 % 2 to 4 weeks after immunization to 8.8 % after 25 or more weeks [2]. Fortunately, effectiveness against severe illness and death was still high. Post-licensure studies have proven that the rollout of vaccination campaigns was associated with substantial reductions in COVID-19 deaths and hospital admissions [18], [23], [32]. One cohort study found BNT162b2 to be effective against the Omicron variant, preventing hospitalization by 70 % [15].

Before the successful deployment of vaccinations, non-pharmaceutical interventions (NPIs) such as social distancing were considered to be one of the most effective ways to contain the dissemination of the COVID-19 virus [26]. Previous studies have proven a significantly positive relationship between human mobility and COVID-19 cases [22], [28], [42], [59]. However, the progress of the vaccination campaign has led to a relaxation in NPIs. On May 13, 2021, the Centers for Disease Control and Prevention (CDC) announced that people who are fully vaccinated could stop wearing masks and maintaining social distancing in most settings. By July 4, 2021, almost all US states had entirely lifted restrictions on travel, eased mask mandates, and gone back to business as usual. Whether the lifting of these restrictions would erode vaccine effectiveness and exacerbate the pandemic has aroused concerns. Some studies stated that vaccinations alone are not enough to curb virus transmission and NPIs are required until sufficient immunity is achieved [39].

Although sufficient studies are focusing on vaccine effectiveness, several research gaps still exist. First, when analyzing vaccine effectiveness, few studies have taken NPIs into account empirically [47]. Some of them although jointly considered the effects of vaccination and NPIs [1], [16], [37], [39], [46], [49], these studies are primarily simulation- or mathematical-based, with the assumption that vaccine efficacy is dependent on NPIs. However, there is a lack of empirical analyses based on population-representative real-world data to verify this assumption. One exception is [21], which empirically investigated the impact of NPIs on COVID-19 vaccine effectiveness in controlling the pandemic. However, this study did not account for the temporal evolution of the vaccination rate and failed to adequately control for the impact of confounding factors. Second, most post-licensure studies were based on surveillance data [2], [13], [50], among which exogenous factors like socioeconomic status, housing types, political ideology, and mobility-related information were missing and cannot be well controlled, which may induce biases into final estimations. Third, although some studies have quantified the effectiveness of vaccines against the Omicron variant [2], [15], few of them have compared the difference before and during the Omicron outbreak at a national level, and limited studies have delineated how the effectiveness varied over time [35].

In this study, we aim to fill these gaps by comprehensively examining the time-varying relationships among vaccination, human mobility, and COVID-19 health outcomes (assessed via case rate and case-fatality rate), controlling for confounders such as racial make-up, industry types, socioeconomics, demographics, and partisanship. Statistical methods, including cross-sectional generalized additive models and time-varying structural equation modeling, were applied to nationwide county-level data throughout the pandemic until February 28, 2022. Our study contributes to the current vaccine trials and emerging observational studies by 1) empirically comparing the differences in case rate and case-fatality rate before and during the Omicron surge; 2) disentangling direct, indirect, and total effects among vaccination, human mobility, and COVID-19 health outcomes via the mediation analysis; 3) investigating time-varying effects throughout the pandemic; and 4) considering underlying socioeconomic, demographic, racial/ethnic, and partisan disparities. Findings are expected to provide scientific evidence and policy suggestions on how to respond to the current COVID-19 pandemic and future epidemics that adequately account for vaccination, NPIs, and social equality.

2. Research design

2.1. Data and variables

Endogenous variables: Three types of endogenous variables are considered, including COVID-19 health outcomes, vaccination, and human mobility. Detailed descriptions and statistics are reported in Table 1. COVID-19 health outcomes were assessed via the number of COVID-19 cases per 100,000 population (abbreviated as case rate) and the number of COVID-19 deaths per 100 confirmed cases (abbreviated as case-fatality rate). County-level statistics documented a much higher daily case rate (Mean: 101.681 vs. 30.799) and lower case-fatality rate (Mean: 0.860 vs. 1.444) during the Omicron surge compared to those before the Omicron surge.

Table 1.

Summary of county-level variables.

		Description	Mean	St.d.	Min.	Median	Max.
Dependent Variables
Case rate		Average daily COVID-19 cases per 100,000 population before the outbreak of Omicron a	30.799	9.206	0.000	30.744	73.778
New case rate		Average daily COVID-19 cases per 100,000 population during the outbreak of Omicron a	101.681	48.516	0.000	100.458	2068.395
Case-fatality rate		Average COVID-19 deaths per 100 confirmed cases before the outbreak of Omicron	1.444	0.954	0.000	1.313	10.870
New case-fatality rate		Average COVID-19 deaths per 100 confirmed cases during the outbreak of Omicron	0.860	0.745	0.000	0.694	15.000
Fully vaccinated rate		Percentage of people fully vaccinated by 2021/12/16, in %	47.693	11.983	0.000	46.900	95.000
At-least-one-dose rate		Percentage of people with at least one dose by 2021/12/16, in %	53.844	15.892	0.000	53.400	95.000
Booster rate		Percentage of people who are fully vaccinated and have received a booster (or additional) dose by 2021/12/16, in %	29.895	9.127	0.000	29.800	95.000
Human mobility		Average daily mobility flow during the outbreak of Omicron, in 104	2.117	5.365	0.005	0.663	104.966
Exogenous Variables
Racial/ethnic groups	White	The percentage of Non-Hispanic Whites, in %	77.320	19.763	0.693	84.694	100.000
	African American	The percentage of African Americans, in %	8.177	13.651	0.000	2.152	87.226
	Asian	The percentage of Asians, in %	1.305	2.428	0.000	0.617	36.467
	Hispanic	The percentage of Hispanics/Latinos, in %	9.620	14.295	0.000	4.097	99.174
	Minority	The percentage of other minorities including American Indian and Alaska Native alone, Native Hawaiian or other Pacific Islander, two or more races, and others, in %	6.320	8.447	0.000	3.896	94.782
Industry types	Finance	The percentage of finance and insurance, real estate, and rental and leasing, in %	4.563	1.924	0.000	4.303	20.141
	Technology	The percentage of professional, scientific, and technical services, in %	3.737	2.669	0.000	3.106	52.900
	Administration	The percentage of administration, business support, management of companies, and waste management services, in %	3.226	1.394	0.000	3.177	15.686
	Manufacture	The percentage of manufacturing industry, in %	12.287	7.140	0.000	11.368	46.394
	Retail	The percentage of retail trade and wholesale trade, in %	13.593	2.654	1.270	13.695	42.424
	Information	The percentage of information, in %	1.327	0.797	0.000	1.247	11.609
	Utility	The percentage of transportation, warehousing, and utilities, in %	5.518	1.944	0.000	5.298	21.849
	Education	The percentage of educational services, in %	9.306	3.238	0.000	8.691	36.123
	Health Care	The percentage of health care and social assistance, in %	13.996	3.376	0.000	13.995	38.154
	Recreation & Food	The percentage of accommodation, food, arts, entertainment, and recreation services, in %	8.298	3.513	0.000	7.958	40.490
	Agriculture	The percentage of agriculture, forestry, fishing, hunting, construction, and mining, in %	14.133	7.801	0.896	12.087	66.748
Socioeconomics	Median Income	The median household income (in 2019 Inflation-Adjusted Dollars), in $103/household	53.528	14.010	21.504	51.946	142.299
	High Educated	The percentage of residents with education attainment equal to/higher than Bachelor, in %	22.080	9.535	0.000	19.644	77.557
	Without Vehicle	The percentage of households with no vehicle available, in %	6.152	3.575	0.000	5.600	77.000
	Without Insurance	The percentage of people with no health insurance coverage, in %	9.295	4.941	0.674	8.381	40.907
	Multi-unit House	The percentage of housing in structures with 10 or more units, in %	4.732	5.709	0.000	3.000	89.400
	Mobile Home	The percentage of mobile homes, in %	12.566	9.171	0.000	10.800	52.600
	Crowd Home	The percentage of occupied housing units with more people than rooms, in %	2.320	1.948	0.000	1.900	33.800
	Group Quarters	The percentage of persons in group quarters, in %	3.456	4.367	0.000	2.000	55.700
Demographics	Population Density	Population density, in 102 persons/sq. mile	2.751	18.559	0.001	0.447	720.192
	Urbanized Population	The percentage of residents in urbanized areas, in %	18.509	33.276	0.000	0.000	100.000
	Age over 65	The percentage of residents 65 years and over, in %	18.992	4.586	6.624	18.636	56.714
	Age under 18	The percentage of residents under 18 years, in %	22.142	3.396	7.269	22.184	41.795
	Male	The percentage of male, in %	50.066	2.259	42.813	49.653	72.720
Partisanship	Democrat	The percentage of Democrats in 2020 presidential candidate vote totals, in %	33.173	15.853	3.091	29.981	89.256
	Republican	The percentage of Republicans in 2020 presidential candidate vote totals, in %	65.039	16.043	8.730	68.310	96.182
Weather	Temperature	Average daily temperature from 2021/12/17 to 2022/02/28, in degrees F	34.628	12.251	0.522	34.442	73.085
	Precipitation	Average daily precipitation from 2021/12/17 to 2022/02/28, in inch	0.048	0.046	0.000	0.039	0.431

Notes:

• a.
  Data sources are as follows: Case and death were from COVID-19 Data Repository by Johns Hopkins University [17]. Vaccination rates were from CDC COVID Data Tracker [11]. Racial groups, industry types, socioeconomics, and demographics were from the Census Bureau’s 2015–2019 ACS 5-year estimates; Weather was from the US National Centers for Environmental Information; Partisanship was from the 2020 presidential election results.
• b.
  Before the outbreak of Omicron: 2021/05/15–2021/12/16; During the outbreak of Omicron: 2021/12/17–2022/02/28.
• c.
  Samples comprise 2919 contiguous US counties. Utah and Georgia were excluded due to a large proportion of missing data.
• d.
  Variables in Italic were excluded from the models due to the high multicollinearity.

Vaccination rates were from CDC COVID Data Tracker [11]. Here we used the percentage of people who are fully vaccinated (have a second dose of a two-dose vaccine or one dose of a single-dose vaccine) based on the county where the recipient lives to represent the county-level vaccination rate. For a robustness check, we also considered other vaccination measures, including the percentage of people with at least one dose (abbreviated as “at-least-one-dose rate”) and the percentage of people who are fully vaccinated and have received a booster (or additional) dose (abbreviated as “booster rate”). We used human mobility as a proxy for people’s adherence to NPIs [58]. Human mobility was calculated using data from SafeGraph [45], a data company that aggregates anonymized location-based service (LBS) data from ∼19 million smartphone devices observed per day across the US. SafeGraph data has emerged as a key data source for tracking population movements during the COVID-19 pandemic in the US [12], [20], [31]. Previous studies have validated SafeGraph data by comparing it with other mobility datasets, such as Google [12] and PlaceIQ [56]. Their results showed that SafeGraph was highly consistent with other mobility data sources. Specifically, we used the Neighborhood Patterns dataset provided by SafeGraph to extract the footfall data aggregated by county to represent human mobility.

Exogenous variables/controls: Exogenous variables included racial/ethnic groups, industry types, demographics, socioeconomics, partisanship, and weather conditions. Variables were selected based on evidence from previous studies [14], [38] and the CDC social vulnerability index [10], which uses 15 variables grouped into four themes, including socioeconomic status, household composition & disability, minority status & language, and housing type & transportation, to reflect the community’s ability to prevent people suffering from disaster. Variables and data were chosen to meet regression assumptions. The generalized variance inflation factor (${\mathit{GVIF}}^{1/2\bullet df}$, where $\mathit{df}$ is the number of coefficients in the variable-wise subset) [19] was calculated to test the multicollinearity, and ${\mathit{GVIF}}^{1/2\bullet df}$ greater than $\surd 5$ (equivalent to $\mathit{VIF}<5$ for non-categorical variables) were excluded. We chose to use GVIF because our model includes categorical variables (e.g., state effects), which may not be handled appropriately by traditional VIF. The detailed statistics and descriptions of final exogenous variables included in the models were shown in Table 1.

2.2. Methodology

2.2.1. Cross-sectional generalized additive models

We first examined underlying determinants of average case rate and case-fatality rate in a cross-sectional setting. The whole course of the pandemic was divided into two parts: before the Omicron surge (2021/05/15–2021/12/16), and during the Omicron surge (2021/12/17–2022/02/28). 2021/05/15 was chosen as the starting point of the observation period, as by then, the second dose of the vaccine had become available to all population groups in the US, and thus vaccine accessibility is not a significant factor to impact COVID-19-related outcomes. 2021/12/17 was set as the beginning of the outbreak of the Omicron variant in the US since it is the time when the percentage of the Omicron variant exceeded 50 % [25]. 2022/02/28 was selected as the end of the observation considering the henceforward large-scale adoption of self-test kits, which would lead to an underestimated case number.

We acknowledge that drawing clear boundaries between the Omicron variant and other variants was challenging, as they coexisted for a considerable period, which means that some new cases during the Omicron surge may still be caused by other variants. However, we deemed those cases negligible due to the Omicron variant's predominance and substantially higher reproductive number [33], [34]. Another concern is that the timing of the Omicron wave's arrival varied across counties. To demonstrate the county-level accuracy, we defined the outbreak of the Omicron variant in each county as the time when the increasing rate of new cases doubled, given that the reproductive number of the Omicron variant is twice that of the Delta variant [33], [34]. Our analysis revealed that 86.54 % of the counties showed signs of an Omicron outbreak between 2021/12/17 and 2021/12/24 (see Appendix Fig. A1), supporting our selection of the Omicron outbreak start time at the county level.

Fig. A1.

Temporal evolution of county-level new case rate. Each blue curve represents one county. The red curve represents the average trend. The black dashed line represents the date 2021/12/17. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

In our study, four cross-sectional models were fitted, with the county-level average case rate and case-fatality rate before and during the Omicron surge as dependent variables, controlling for effects from vaccination, mobility, and various exogenous variables. The four dependent variables were calculated as follows:

$$C_i=100000\frac{{\sum }_{t_0}^{T_0}{c}_{i,t}}{\left(T_0-t_0\right)P_i},C_i^{\prime }=100000\frac{{\sum }_{T_0}^T{c}_{i,t}}{\left(T-T_0\right)P_i}$$ (1)

$$F_i=100\frac{{\sum }_{t_0}^{T_0}{f}_{i,t}}{{\sum }_{t_0}^{T_0}{c}_{i,t-l}},F_i^{\prime }=100\frac{{\sum }_{T_0}^T{f}_{i,t}}{{\sum }_{T_0}^T{c}_{i,t-l}}$$ (2)

where $c_{i,t}$ is the number of new confirmed cases in county i on day t; $l$ is the time lag between infection and death, which is set as 21 days in this study [4],$f_{i,t}$ is the number of new deaths in county i on day t;$P_i$ is the total population of county i; $t_0$ is the start time of the pandemic (2021/05/15); $T_0$ is the start time of the outbreak of the Omicron variant (2021/12/16); $T$ is the end time of the observation (2022/02/28); $C_i,C_i^{\prime },F_i,F_i^{\prime }$ denotes the four dependent variables, i.e. case rate before the Omicron surge, case rate during the Omicron surge, case-fatality rate before the Omicron surge, and case-fatality rate during the Omicron surge, respectively.

In addition to the four models related to COVID-19 health outcomes, we built two other cross-sectional models using the average vaccination rate and human mobility during the Omicron surge as dependent variables, controlling for effects from other exogenous variables. The effects of vaccination on human mobility were also controlled when modeling human mobility.

All cross-sectional models were fitted under the generalized additive model (GAM) framework. GAM [57] is a semi-parametric model with a linear predictor involving a series of additive non-parametric smooth splines of covariates. A noticeable advantage of GAM is its capability and flexibility to handle different nonlinear effects [57], such as state random effects and spatial autocorrelations in our study [55]:

$$g\left(H_i\right)={\beta }_0^{\left(0\right)}+{\delta }^{(0)}{V}_i+{\gamma }^{(0)}{M}_i+\sum _{r=1}^R{\beta }_r^{\left(0\right)}{X}_{r,i}+ti\left(X_{p,i}\times X_{q,i}\right)+R_i+E_O$$ (3)

$$g\left(V_i\right)={\beta }_0^{\left(1\right)}+\sum _{r=1}^R{\beta }_r^{\left(1\right)}{X}_{r,i}+ti\left(X_{p,i}\times X_{q,i}\right)+R_i+E_P$$ (4)

$$g\left(M_i\right)={\beta }_0^{\left(2\right)}+{\delta }^{(2)}{V}_i+\sum _{r=1}^R{\beta }_r^{\left(2\right)}{X}_{r,i}+ti\left(X_{p,i}\times X_{q,i}\right)+R_i+E_M$$ (5)

$$\frac{\left(Y_i-{\mu }_i\right)}{{\sigma }_i}t\left({\vartheta }_i\right),Y_i\in H_i,M_i,V_i$$ (6)

where $H_i$ is one of the dependent variables from $C_i,C_i^{\prime },F_i,F_i^{\prime }$; ${\beta }_0^{\left(0-2\right)}$ are the overall intercepts; $g(.)$ is the link function; ${\beta }_r^{\left(0-2\right)}$ are the coefficients of the r^th control variable $X_{r,i}$, and R is the number of controls; ${\delta }^{(0)}$ is the coefficient of the vaccination rate $V_i$; ${\gamma }^{(0)}$ is the coefficient of human mobility $M_i$; $X_p$ and $X_q$ are the pairs of independent variables nonlinearly interplaying with each other; $ti\left(.\right)$ is marginal nonlinear smoother excluding the basic functions associated with the main effects (in our cases, the spatial coordinate interaction was fitted through this term); $R_i$ is the nonlinear random effect of county i; $E_O,$ $E_P,$ $E_M$are the error terms. Due to the non-normality (heavy-tailed distributions) of those county-level variables, all dependent variables are assumed to follow scaled t-family with ${\mu }_{\mathrm{i}}$ as expectation, ${\sigma }_{\mathrm{i}}$ as variance, and ${\vartheta }_{\mathrm{i}}$ as the degree of freedom. ${\mu }_{\mathrm{i}}$ is determined by a linear predictor, and ${\sigma }_{\mathrm{i}}$ and ${\vartheta }_{\mathrm{i}}$ are estimated alongside smoothing parameters.

2.2.2. Time-varying structural equation modeling

Cross-sectional models provide an estimate of relationships at an average level. However, when variables are averaged over a long period, important temporal dynamics in endogenous variables may be overlooked, potentially resulting in biased relationship estimates and a limited understanding of how these relationships evolve over time. Moreover, separately estimated models cannot fully disentangle the extent to which human mobility accounts for the relationship between vaccination and COVID-19 health outcomes. To this end, we constructed a set of structural equation modeling (SEM) under specifications similar to cross-sectional models, but with time-varying coefficients that consider the fluctuations of all variables as their monthly averages shifted throughout the pandemic with a weekly resolution. Mediation analysis was employed to understand the underlying mechanism by which vaccination influences COVID-19 health outcomes through human mobility, controlling for other exogenous variables. Specifically, we assumed human mobility ($M_{i,t}$) could directly influence COVID-19 health outcomes (${\overline{H}}_{i,t:t+30}$), vaccination ($V_{i,t}$) could influence COVID-19 health outcomes both directly and indirectly via human mobility, and other controls could exert influence on each of the three endogenous variables. The conceptual diagram for the SEM was shown in Fig. 1 and the equations were as Eqs. (7), (8), (9):

$$g\left({\overline{H}}_{i,t:t+30}\right)={\beta }_{0,t}^{\left(0\right)}+I(t){\delta }_t^{\left(0\right)}{V}_{i,t}+{\gamma }_t^{\left(0\right)}{M}_{i,t}+\sum _{r=1}^R{\beta }_{r,t}^{\left(0\right)}{X}_{r,t,i}+R_{i,t}+E_{O,t}$$ (7)

$$g\left(V_{i,t}\right)=I(t)({\beta }_{0,t}^{\left(1\right)}+\sum _{r=1}^R{\beta }_{r,t}^{\left(1\right)}{X}_{r,t,i}+R_{i,t}+E_{P,t})$$ (8)

$$g\left(M_{i,t}\right)={\beta }_{0,t}^{\left(2\right)}+I(t){\delta }_t^{\left(2\right)}{V}_{i,t}+\sum _{r=1}^R{\beta }_{r,t}^{\left(2\right)}{X}_{r,t,i}+R_{i,t}+E_{M,t}$$ (9)

where ${\overline{H}}_{i,t:t+30}$ is the average case rate or case-fatality rate from day t to day t + 30 in county i; $t$ is the day index rolling in a weekly step, i.e. $t=1,8,15,...$; ${\beta }_{0,t}^{\left(0-2\right)}$ are the overall intercepts on day t; $I(t)$ is the indicator of whether the time exceeds the initial rollout time of vaccines in the US; ${\beta }_{r,t}^{\left(0-2\right)}$ are coefficients of the r^th control variable $X_{r,t,i}$ by day t; $V_{i,t}$ is the cumulative fully vaccinated rate in county i by day t and ${\delta }_t^{\left(0\right)}$ is its coefficient; ${\gamma }_t^{\left(0\right)}$ is the coefficient of human mobility $M_{i,t}$ by day t; $R_{i,t}$ is the fixed effect of county i on day t; $E_{O,t},$ $E_{P,t},$ $E_{M,t}$are error terms.

Fig. 1.

Conceptual Diagram for structural equation modeling.${\mathit{c}}_{\mathit{E}\mathit{M}},{\mathit{c}}_{\mathit{E}\mathit{P}},{\mathit{c}}_{\mathit{E}\mathit{O}}$ mean the variances of the error terms of mediators, predictor, and outcomes; ${\mathit{c}}_{\mathit{r}\mathit{g}},{\mathit{c}}_{\mathit{i}\mathit{t}},$${\mathit{c}}_{\mathit{s}\mathit{e}},$${\mathit{c}}_{\mathit{d}},$${\mathit{c}}_{\mathit{p}},{\mathit{c}}_{\mathit{w}},{\mathit{c}}_{\mathit{s}\mathit{e}}$ mean the variances of controls.

Following the general rules of SEM [26], [30], we hypothesized there are three different types of effects, including:

• (1)Direct effects: Vaccination $\to$ COVID-19 health outcomes: ${\delta }_t^{\left(0\right)}$; Mobility $\to$ COVID-19 health outcomes: ${\gamma }_t^{\left(0\right)}$.
• (2)Indirect effects: Vaccination $\to$ Mobility $\to$ COVID-19 health outcomes: ${\delta }_t^{\left(2\right)}\bullet {\gamma }_t^{\left(0\right)}$;
• (3)Total effects: Vaccination $\to$ COVID-19 health outcomes: ${\delta }_t^{\left(0\right)}+{\delta }_t^{\left(2\right)}\bullet {\gamma }_t^{\left(0\right)}$.

The model parameters were estimated using maximum likelihood estimation. Since indirect effects are calculated by products of estimation, their estimated distributions tend to be nonnormal. Thus, 95 % CIs of indirect effects and total effects were obtained through 500 bootstraps. Finally, several SEM fit measures were employed, including Root means the square error of approximation (RMSEA), Tucker-Lewis index (TLI), and Comparative fit index (CFI) to evaluate the SEM goodness-of-fit with the rules of thumb guidelines including CFI $\ge$0.95, TLI $\ge$0.95, and RMSEA $\le$ 0.06.

3. Results

3.1. Spatiotemporal distribution

Temporal evolutions of vaccination, mobility, and COVID-19 health outcomes are visualized in Fig. 2 . Moving across the pandemic, we noticed the epidemiological situation in the US has been characterized by five distinct waves in terms of case and death rate (Fig. 2 (a)), starting in March 2020, June 2020, October 2020, July 2021, and December 2021, respectively. The last two waves also correspond to the outbreak of the Delta variant and the Omicron variant (Fig. 2 (c)), among which the Omicron variant has led to a dramatic spike, particularly in the new case rate. In addition, the temporal patterns of new case rate (Fig. 2 (a)) and case-fatality rate (Fig. 2 (d)) were non-conforming. The highest case-fatality rate occurred in May 2020, decreased afterward, and stayed at a low rate regardless of the continual explosion of new cases.

Fig. 2.

Temporal evolution of nationwide (a) cases rate (/100,000) and death rate(/100,000); (b) vaccination rate (%) and new vaccinated doses; (c) % of SARS-Cov-2 variants; (d) Mobility flow (10,000) and case-fatality rate (%). Translucent curves are daily and bold curves are weekly.

As for vaccination, starting from December 2020, the nationwide vaccination rate has steadily increased but with a varying rate of increase. We observed two peaks in the number of newly administered doses, occurring in April 2021 and December 2021, respectively. Regarding human mobility, we found that mobility plummeted steeply in March 2020 until reaching its nadir in late-April 2020, followed by rapid recovery to near-unperturbed status in July 2020 (Fig. 2 (d)). This can be explained by the announcement and lifting of stay-at-home orders at the early stage of COVID-19. Afterward, mobility remained relatively stable until the end of observation.

Spatial distributions of vaccination, mobility, and COVID-19 health outcomes were mapped in Fig. 3 . Spatial distributions of new case rates and new case-fatality rates were different. States located in the junction of South and Midwest and part of the Southwest exhibited the highest new case rate, while Florida, East North Central, and West South Central exhibited the highest new case-fatality rate. Such a difference indicates that the underlying determinants of case rate and case-fatality rate are diverse. In addition, the spatial distribution of the vaccination rate did not show a distinguishable reverse pattern compared with the new case rate. California, Northeast, and states like Arizona and New Mexico exhibited the highest fully vaccinated rate, while some of these regions such as Arizona and California showed higher new case rates. This implies vaccination may not significantly inhibit virus transmission at least in some regions. Last, regarding human mobility, we found high concentrations of mobility located in the West Coast, Northeast, and South during the Omicron surge.

Fig. 3.

Spatial distribution of county-level (a) new case rate; (b) new case-fatality rate (%); (c) fully vaccinated rate (%); and (d) mobility flow (10,000). (a), (b), and (d) are plotted based on daily average data during the Omicron surge (2021/12/17–2022/02/28). (c) is plotted based on data by the eve of the Omicron surge (by 2021/12/16).

3.2. Outcomes of cross-sectional regressions on COVID-19 health outcomes

Standardized outputs of cross-sectional models related to COVID-19 health outcomes were reported in Table 2 . Contrary to our assumption, our results documented that mobility flow did not show significant associations with case rate and case-fatality rate. In addition, results revealed a counterintuitive relationship between vaccination rate and COVID-19 health outcomes. Counties with greater percentages of fully vaccinated residents presented significantly higher case rates during the Omicron surge but did not show significant signs before the Omicron outbreak. However, a significantly negative association between vaccination rate and case-fatality rate was found both before and after the Omicron outbreak. These findings suggest that vaccination may not be effective in preventing infections but does show effectiveness against COVID-19-related deaths.

Table 2.

Estimations of COVID-19 health outcomes before and during the outbreak of the Omicron variant.

Parametric coefficients
Variables	Case rate (Before the Omicron)	Case-fatality rate (Before the Omicron)	New case rate (During the Omicron)	New case-fatality rate (During the Omicron)
(Intercept)	0.030 (−0.113, 0.173)	0.007 (−0.153, 0.166)	−0.007 (−0.101, 0.088)	−0.005 (−0.164, 0.154)
Human Mobility	0.005 (−0.029, 0.039)	−0.019 (−0.056, 0.017)	0.007 (−0.016, 0.030)	−0.028. (−0.062, 0.005)
Fully Vaccinated Rate	0.024 (−0.014, 0.063)	−0.081*** (−0.123, −0.040)	0.150*** (0.124, 0.177)	−0.096*** (−0.136, −0.057)
Racial/ethnic groups
Asian	−0.046* (−0.087, −0.004)	0.091*** (0.047, 0.135)	−0.039** (−0.067, −0.011)	0.059** (0.019, 0.100)
African American	0.122*** (0.057, 0.188)	0.187*** (0.117, 0.257)	−0.083*** (−0.125, −0.042)	−0.010 (−0.071, 0.051)
Hispanic	0.084*** (0.039, 0.129)	0.031 (−0.017, 0.079)	0.043* (0.008, 0.078)	−0.061. (−0.113, 0.010)
Minorities	0.100*** (0.052, 0.147)	0.015 (−0.036, 0.066)	0.071*** (0.044, 0.098)	−0.038. (−0.078, 0.003)
Industry types
Recreation & Food	0.060*** (0.026, 0.095)	−0.002 (−0.039, 0.035)	0.041*** (0.017, 0.065)	0.020 (−0.015, 0.055)
Health Care	0.100*** (0.068, 0.133)	0.054** (0.020, 0.089)	0.086*** (0.064, 0.109)	−0.019 (−0.051, 0.014)
Retail	0.037** (0.009, 0.065)	0.021 (−0.009, 0.051)	0.018. (−0.001, 0.037)	−0.000 (−0.028, 0.028)
Utilities	0.048** (0.020, 0.077)	0.037* (0.006, 0.068)	0.028** (0.009, 0.048)	0.044** (0.015, 0.073)
Education	−0.033. (−0.070, 0.005)	−0.027 (−0.067, 0.013)	−0.006 (−0.031, 0.019)	−0.009 (−0.046, 0.028)
Manufacture	0.078*** (0.036, 0.119)	0.006 (−0.039, 0.050)	0.048*** (0.019, 0.076)	0.033 (−0.008, 0.074)
Scientific	0.027 (−0.022, 0.076)	−0.012 (−0.065, 0.041)	0.001 (−0.037, 0.040)	0.038 (−0.011, 0.087)
Administration	0.061*** (0.028, 0.094)	0.001 (−0.034, 0.037)	−0.002 (−0.024, 0.019)	0.009 (−0.023, 0.041)
Information	0.006 (−0.022, 0.035)	0.041 (−0.010, 0.071)	−0.001 (−0.020, 0.019)	0.008 (−0.021, 0.036)
Socioeconomics
Median Income	−0.248*** (−0.306, −0.190)	−0.072* (−0.134, −0.010)	−0.054** (−0.093, −0.015)	−0.146*** (−0.203, −0.089)
No Vehicle	0.025 (−0.027, 0.077)	0.031 (−0.024, 0.087)	0.039. (−0.004, 0.073)	0.034 (−0.017, 0.085)
No Insurance	−0.089*** (−0.135, −0.044)	0.094*** (0.045, 0.143)	−0.074*** (−0.108, −0.040)	0.107*** (0.057, 0.156)
Multi-unit House	0.049** (0.003, 0.100)	−0.022 (−0.077, 0.033)	0.040. (−0.006, 0.075)	−0.059. (−0.109, 0.008)
Mobile Home	−0.065** (−0.107, −0.023)	−0.011 (−0.058, 0.037)	0.020 (−0.009, 0.049)	−0.030 (−0.073, 0.012)
Crowd Home	0.097*** (0.053, 0.141)	0.037 (−0.008, 0.082)	0.009 (−0.023, 0.040)	−0.016 (−0.058, 0.026)
Group Quarters	0.005 (−0.041, 0.052)	0.001 (−0.049, 0.051)	0.017 (−0.014, 0.048)	−0.019 (−0.065, 0.027)
Demographics
Male	−0.087*** (−0.130, −0.044)	−0.031 (−0.078, 0.015)	−0.025 (−0.055, 0.005)	0.040. (−0.005, 0.084)
Age over 65	−0.156*** (−0.208, −0.103)	0.151*** (0.095, 0.207)	−0.173*** (−0.208, −0.138)	0.225*** (0.174, 0.277)
Age under 18	0.059. (−0.000, 0.119)	−0.035 (−0.099, 0.028)	0.003 (−0.036, 0.042)	0.073* (0.016, 0.130)
Population Density	−0.019 (−0.062, 0.025)	−0.027 (−0.074, 0.020)	0.008 (−0.021, 0.037)	0.015 (−0.027, 0.058)
Urbanized Population	−0.005 (−0.049, 0.038)	−0.037 (−0.084, 0.009)	0.002 (−0.028, 0.032)	0.059** (0.015, 0.102)
Partisanship
Democrat	−0.473*** (−0.545, −0.401)	−0.137*** (−0.214, −0.060)	0.000 (−0.047, 0.048)	−0.118*** (−0.187, −0.049)
Weather
Temperature	−0.024 (−0.118, 0.071)	0.031 (−0.071, 0.133)	0.139*** (0.076, 0.202)	−0.133** (−0.229, −0.037)
Precipitation	−0.003 (−0.042, 0.035)	0.026 (−0.015, 0.068)	0.026. (−0.001, 0.052)	−0.019 (−0.059, 0.020)
Smooth terms
	e.d.f.	e.d.f.	e.d.f.	e.d.f.
s(STFIPS)	39.650***	39.869***	41.232***	42.028***
ti(Latitude, Longitude)	9.433***	8.185**	10.953***	11.386***
Model fit
R-sq.(adj)	0.586	0.485	0.497	0.438

Notes: Coefficients are standardized, and all interpretations are based on the unit of variable’s standard deviation. Robust 95% confidence intervals are in parentheses. Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1. s() means the spline function. ti() means the marginal nonlinear interaction function.

Regarding racial/ethnic groups, holding others constant and using White populations as a reference, counties with more Asians were associated with significantly lower case rates but higher case-fatality rates across the pandemic. Counties with greater Hispanic populations and more minorities showed higher case rates across the pandemic but did not show significant differences in case-fatality rates. African Americans were the only race showing different infection and death patterns before and during the Omicron surge. Before the Omicron surge, counties with a higher percentage of African Americans exhibited higher case rates and case-fatality rates. However, during the Omicron surge, these counties had lower case rates and lost significance in the relationship with case-fatality rates.

For industry types, we found counties with higher proportions of recreation and food services, health care services, utilities, and manufacturing presented higher case rates throughout the pandemic. However, their relationships with case-fatality rates were mostly insignificant except for health care services, which showed a significantly positive relationship with case-fatality rates before the Omicron surge. Utility services were the only industry type that consistently showed positive relationships with both case rates and case-fatality rates across the whole pandemic, which is plausible considering utility services mostly belong to “essential” services and have a high degree of exposure to human interaction.

Regarding socioeconomics, counties with higher median household income constantly exhibited lower case rates and case-fatality rates across the pandemic. Counties with lower percentages of health insurance coverage constantly presented lower case rates but higher case-fatality rates throughout the pandemic. Housing types only presented significance in the case rate before the Omicron surge. For example, counties with higher proportions of high-density housing showed higher case rates during the pre-Omicron period, which indicates at the early stage, physical interaction among crowd spaces is one of the main drivers of virus transmission. Among demographics, age composition showed the most significance. Counties with more elders (65+) continuously presented lower case rates but higher case-fatality rates across the pandemic. Last, partisanship also presented significant relationships with COVID-19 health outcomes. Counties with more Democrats showed lower case-fatality rates across the pandemic; however, they only presented significantly lower case rates before the Omicron and henceforward lost significance.

Rankings of the standardized estimates (only considering those of significance) also provide interesting findings. Before the Omicron surge, the percentage of Democrats presented the strongest (negative) relationship with the case rate. During the Omicron surge, the percentage of elders presented the strongest (negative) relationship with the case rate. As for the case-fatality rate, before the Omicron surge, the percentage of African American presented the strongest (positive) relationship with the case-fatality rate. During the Omicron surge, the percentage of elders presented the strongest (positive) relationship with the case-fatality rate. These substantial differences further corroborate that the segments of populations bearing a disproportionate burden are substantially different before and during the Omicron surge.

It is noteworthy that the significantly positive relationship between the new case rate and the fully vaccinated rate deserves further discussion. As the relationship cannot prove causality, it is inappropriate here to interpret that vaccination will lead to an increase in the new case rate during the Omicron surge. One explanation is the waning effectiveness of vaccines. People with full vaccination may receive vaccines long before the Omicron surge which has already lost effectiveness to some extent. In addition, counties with higher vaccination coverage may be more aggressive in lifting NPIs, putting people at a higher risk of exposure to the new variant [48]. The lower effectiveness against Omicron infections and the higher risk of exposure jointly lead to a positive relationship, which is documented in our models.

3.3. Robustness check

For robustness check, the fully vaccinated rate in Table 2 was replaced by other vaccination measures, including the percentage of people with a booster, the percentage of people with at least one dose, and the percentage of people fully vaccinated in the past two, four, and six months before the Omicron surge, and the coefficients of vaccination measures and human mobility were extracted and reported in Table 3 . The reason that we considered more recent vaccination rates is that several studies have documented waning vaccine effectiveness with a duration of 5–6 months [41], [50], [51].

Table 3.

Robustness check of different vaccination measures.

Dependent variable: new case rate (During the Omicron surge)
	Booster rate	At-least-one-dose rate	Fully vaccinated rate (past 6 months)	Fully vaccinated rate (past 4 months)	Fully vaccinated rate (past 2 months)
Coeff.	−0.005 (−0.037, 0.026)	0.147*** (0.118, 0.175)	0.092*** (0.054, 0.130)	0.075*** (0.032, 0.118)	0.055* (0.007, 0.104)
Dependent variable: new case-fatality rate (During the Omicron surge)
	Booster rate	At-least-one-dose rate	Fully vaccinated rate (past 6 months)	Fully vaccinated rate (past 4 months)	Fully vaccinated rate (past 2 months)
Coeff.	0.027 (−0.020, 0.073)	−0.083*** (−0.124, −0.041)	−0.110*** (−0.166, −0.055)	−0.115*** (−0.177, −0.052)	−0.136*** (−0.206, −0.066)

Notes: Coefficients are standardized, and all interpretations are based on the unit of variable’s standard deviation. Robust 95 % confidence intervals are in parentheses. Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1. We only considered the robustness check on new case rate and case-fatality rate during the Omicron surge considering the chronological condition.

As shown in Table 3, the relationships between vaccination rate and COVID-19 health outcomes lost their significance when using booster rate as the vaccination measure. This could be attributed to the fact that the booster rate was still relatively low at the onset of the Omicron surge. As per Table 1, the average county-level booster rate was only 29.895 % before the Omicron outbreak. Therefore, the impact of boosters in curbing Omicron transmission remains insignificant at the early stage of the Omicron outbreak. On the other hand, when using at-least-one-dose rate or fully vaccinated rate in the past two, four, and six months as the vaccination measure, signs and significances remained consistent with Table 2. However, as the backward length decreased, the positive relationships between vaccination rate and case rate weakened, while the negative relationships between vaccination rate and case-fatality rate strengthened, which affirms waning vaccine effectiveness over time.

We also examined the impact of various controls on model estimation and presented the coefficients of vaccination rate and human mobility in Table 4 . We excluded different groups of controls from the fully-controlled models to analyze the coefficient change. The “vanilla” model, which only included state effects and weather, and removed all other exogenous variables, was established as the baseline. As shown, the significance and signs of the vaccination rate remained stable across all models. However, the effects of human mobility were only significant when no controls were included. Despite testing the GVIF to manage multicollinearity, strong connections between human mobility and controls persisted (refer to Table 5 ). The loss of significance in human mobility suggested that other controls were more effective in explaining the dependent variables in comparison to human mobility. Nonetheless, it also indicated that, at least in a univariate context, human mobility was significantly and positively related to the new case rate and negatively related to the case-fatality rate during the Omicron outbreak.

Table 4.

Robustness check of different controls.

Dependent variable: new case rate (During the Omicron surge)
	Vanilla	-Racial	-Socioeconomic	-Industry	-Demographic	-Partisanship
Human Mobility	0.110*** (0.078, 0.142)	0.014 (−0.022, 0.049)	0.017 (−0.018, 0.052)	0.032. (−0.004, 0.067)	0.033. (−0.003, 0.068)	0.024 (−0.011, 0.059)
Fully Vaccinated Rate	0.225*** (0.190, 0.261)	0.241*** (0.201, 0.280)	0.210*** (0.171, 0.249)	0.234*** (0.194, 0.274)	0.162*** (0.121, 0.202)	0.194*** (0.156, 0.232)
Dependent variable: new case-fatality rate (During the Omicron surge)
	Vanilla	-Racial	-Socioeconomic	-Industry	-Demographic	-Partisanship
Human Mobility	−0.095*** (−0.127, −0.063)	−0.014 (−0.050, 0.023)	−0.020 (−0.057, 0.017)	−0.012 (−0.049, 0.025)	−0.014 (−0.051, 0.023)	−0.013 (−0.050, 0.025)
Fully Vaccinated Rate	−0.202*** (−0.237, −0.166)	−0.096*** (−0.137, −0.055)	−0.126*** (−0.167, −0.084)	−0.102*** (−0.144, −0.060)	−0.059** (−0.101, −0.017)	−0.108*** (−0.148, −0.068)

Notes: Coefficients are standardized, and all interpretations are based on the unit of variable’s standard deviation. Robust 95 % confidence intervals are in parentheses. Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1. Different groups of controls are as follows: Vanilla: only consider state random effects and weather; -Racial: only controls belong to racial/ethnic groups excluded; - Socioeconomic: only controls belong to socioeconomics excluded; - Industry: only controls belong to industry types excluded; - Demographic: only controls belong to demographics excluded; - Partisanship: only controls belong to partisanship excluded.

Table 5.

Estimations of vaccination rate and human mobility (During the Omicron surge).

Parametric coefficients
Variables	Human mobility	Fully vaccinated rate
(Intercept)	−0.018 (−0.064, 0.029)	0.005 (−0.110, 0.120)
Fully Vaccinated Rate	0.040 (0.012, 0.068)**
Racial/ethnic groups
Asian	0.164 (0.134, 0.195)***	0.032 (−0.004, 0.068).
African American	0.062 (0.020, 0.105)**	−0.208 (−0.261, −0.156)***
Hispanic	0.008 (−0.028, 0.044)	0.124 (0.079, 0.169)***
Minorities	0.008 (−0.021, 0.037)	0.087 (0.052, 0.122)***
Industry types
Recreation & Food	−0.010 (−0.036, 0.016)	0.045 (0.014, 0.076)**
Health Care	0.023 (−0.001, 0.047).	0.093 (0.064, 0.121)***
Retail	0.032 (0.011, 0.052)**	0.053 (0.028, 0.078)***
Utilities	0.022 (0.001, 0.044)*	−0.007 (−0.033, 0.018)
Education	−0.036 (−0.063, −0.009)**	−0.006 (−0.039, 0.026)
Manufacture	0.012 (−0.018, 0.041)	−0.022 (−0.058, 0.015)
Scientific	0.027 (−0.009, 0.062)	0.017 (−0.026, 0.059)
Administration	0.054 (0.031, 0.078)***	0.011 (−0.017, 0.039)
Information	0.036 (0.015, 0.058)***	0.022 (−0.003, 0.048).
Socioeconomics
Median Income	−0.086 (−0.127, −0.046)***	0.236 (0.187, 0.285)***
No Insurance	0.025 (−0.010, 0.060)	−0.115 (−0.158, −0.071)***
No Vehicle	−0.106 (−0.143, −0.070)***	0.017 (−0.027, 0.062)
Multi-unit House	0.121 (0.084, 0.158)***	0.015 (−0.029, 0.060)
Mobile Home	−0.061 (−0.092, −0.031)***	−0.052 (−0.089, −0.015)**
Crowd Home	−0.006 (−0.036, 0.025)	0.073 (0.034, 0.111)***
Group Quarters	−0.043 (−0.077, −0.010)*	−0.035 (−0.075, 0.005).
Demographics
Male	0.010 (−0.023, 0.043)	0.063 (0.024, 0.102)**
Age over 65	−0.028 (−0.066, 0.009)	0.172 (0.127, 0.217)***
Age under 18	0.019 (−0.022, 0.061)	−0.038 (−0.088, 0.012)
Population Density	0.090 (0.047, 0.132)***	−0.110 (−0.160, −0.060)***
Urbanized Population	0.269 (0.237, 0.301)***	−0.019 (−0.057, 0.018)
Partisanship
Democrat	−0.026 (−0.074, 0.023)	0.630 (0.573, 0.687)***
Weather
Temperature	0.122 (0.072, 0.172)***	0.014 (−0.066, 0.094)
Precipitation	0.012 (−0.016, 0.039)	−0.006 (−0.040, 0.029)
Smooth terms
	e.d.f.	e.d.f.
s(STFIPS)	31.533***	40.820***
ti(Latitude, Longitude)	13.064***	6.449
Model fit
R-sq.(adj)	0.548	0.625

Notes: Coefficients are standardized, and all interpretations are based on the unit of variable’s standard deviation. Robust 95 % confidence intervals are in parentheses. Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1. s() means the spline function. ti() means the marginal nonlinear interaction function. For other vaccination measures, estimation results are recorded in Appendix Table A1.

3.4. Outcomes of cross-sectional regressions on mobility and vaccination

We also modeled vaccination coverage and human mobility to understand disparities in receiving vaccines and restricting travel, as well as the relationship between mobility and vaccination. The estimation results are reported in Table 5. As shown, in the human mobility model, we found counties with higher vaccination rates also exhibited greater mobility. This is plausible since traveling to get vaccines may increase human mobility, and those who are fully vaccinated may also travel more due to the relief from the virus threat. Substantial vaccination inequalities can also be observed in Table 5. Counties with fewer African Americans and more Hispanics and other minorities exhibited higher vaccination rates. Counties with more recreation and food services, health care services, and retails also presented higher vaccination rates. As for socioeconomics, counties with high socioeconomic statuses presented greater vaccination rates. Counties with fewer mobile homes and more crowded homes also presented higher vaccination rates. Among demographics, counties with higher percentages of males and elders and lower population density presented higher vaccination rates. Last, partisanship presented the strongest (positive) relationship with the vaccination rate. Counties with higher percentages of Democrats presented much higher vaccination rates, indicating that political ideology may be the key determinant of the willingness to vaccinate.

Note that obscure patterns were found when we compared Table 5 with Table 2. Counties showing better adherence to NPIs or with higher vaccination rates did not always present lower case rates and case-fatality rates. For example, counties with higher percentages of Hispanics and other minorities and higher percentages of health care services presented significantly higher vaccination rates but continuously presented higher case rates across the pandemic. Counties with higher percentages of elders presented substantially higher vaccination rates but consistently presented higher case-fatality rates. These paradoxes imply that vaccine effectiveness may be weakened when crossing with socioeconomic and demographic status. High-risk and vulnerable populations are still bearing a larger brunt despite their higher vaccination rates.

3.5. Outcomes of temporal structural equation models

To further disentangle the time-varying relationships among vaccination, mobility, COVID-19 health outcomes, and various controls, we built a set of rolling SEMs with coefficients varying across the pandemic. In the main text, we focused on the temporal evolutions of standardized effects among endogenous variables (Fig. 4 ). Other effects, such as the time-varying relationships between controls and endogenous variables, were delineated in Appendix Fig. A2. Several important findings can be summarized as follows:

Fig. 4.

Temporal evolution of standardized effects among mobility flow, fully vaccinated rate, and COVID-19 health outcomes. Coefficients are standardized. Each spot, with robust 95 % CI as the error bar, represents coefficients of one cross-sectional SEM estimated using one-month average data with a 7-day rolling step. Only coefficients with P-values greater than 0.05 are plotted. Each curve denotes the temporal evolution smoothed by the spline function. Detailed statistics are reported in Appendix Table A2.

Fig. A2.

Temporal evolution of standardized indirect effects from controls to new case rate (a) and case-fatality rate (b) via mobility flow and fully vaccinated rate. Each spot, with robust 95 % CI as the error bar, represents coefficients of one cross-sectional SEM estimated using one-month average data with a 7-day rolling step. Only coefficients with P-values greater than 0.05 are plotted. Each curve denotes the temporal evolution smoothed by the spline function.

• (1)Mobility directly exerted positive effects on case rate and case-fatality rate across the pandemic (blue curves). Strong positive relationships were observed particularly with case rate, exhibiting several salient peaks occurring in March 2020, June 2020, December 2020, April-June 2021, and December 2021. These peaks are in line with the waves of epidemic outbreaks (Fig. 2 (a)), implying mobility consistently led to more infections during each wave of outbreak.
• (2)Vaccination rate directly presented a significantly negative relationship with case rate before November 2021 but changed to significantly positive afterward (purple curves in Fig. 4 (a)), implying vaccines did help decrease infections at least before the outbreak of Omicron variant; however, it failed to curb the transmission of the Omicron potentially due to waning vaccine effectiveness or the Omicron variant escaping vaccine protection. We also found noticeable fluctuation regarding the relationship between vaccination rate and case rate. Three local minimums of vaccine effectiveness against SARS-CoV-2 infections were observed in February 2021, June 2021, and December 2021, corresponding to the three main outbreaks of variants.
• (3)Vaccination rate indirectly exerted a positive effect on the case rate via human mobility (green curves in Fig. 4 (a)). Such indirect effects were significant during April–July 2021 (Mean: 8.862 %, 95 % CI: 5.509, 12.215) and November–December 2021 (Mean: 13.709 %, 95 % CI: 0.171, 27.247). On average, human mobility has averagely caused a 10.276 % (95 % CI: 6.257, 14.294) decrease in vaccine effectiveness against SARS-CoV-2 infections. Such findings highlight the need for continued adherence to NPIs to achieve high vaccine effectiveness.
• (4)Regarding the case-fatality rate, the vaccination rate has directly presented a significantly negative relationship with the case-fatality rate since June 2020 (purple curves in Fig. 4 (b)). Unlike the relationship with case rate, the relationship between vaccination rate and case-fatality rate remained negative even during the Omicron surge, although a reduction in vaccine effectiveness was also observed. Such findings indicated that vaccination did help lower disease severity and averted fatality among infected populations.
• (5)The indirect effect of vaccination rate on case-fatality rate via human mobility was weaker and showed less significance compared with case rate. However, we still observed a 6.57 % (95 % CI: −8.58, 21.72) decrease, albeit not significant, in vaccine effectiveness against case-fatality rate during the outbreak of the Omicron variant that should be sourced to human mobility. These findings further affirm the importance of sticking to NPIs before achieving herd immunity.

4. Discussion, conclusion, and limitations

Leveraging over two years of nationwide data, our study documented a salient structural inequality in COVID-19-related outcomes, which is consistent with various previous studies [7], [8], [9], [27], [35], [36], [38], [43], [44], [53]. Counties with higher median household income continuously exhibited lower case rates, lower case-fatality rates, higher vaccination rates, and better adherence to NPIs across the pandemic. Racial minorities, high-risk and socially vulnerable population groups, and those engaged in health care services, though with significantly higher vaccination rates, still suffered a larger burden of infection or death risk. In addition, the percentage of Democrats presented the strongest associations with vaccination rate (positive) and case rate (negative) before the Omicron surge, indicating that political ideology may be one of the root causes of observed disparities in the US. These persistent inequalities indicate additional efforts are still needed to overcome the ideological and structural barriers that perpetuate these inequities. Countermeasures may include assigning additional resources to protect high-risk essential workers, maintaining targeted NPIs, prioritizing persons in socially vulnerable groups during vaccination, and engaging faith leaders to mitigate medical mistrust and vaccine hesitancy.

We found the outbreak caused by the Omicron variant is substantially different from previous waves of the outbreak. A much higher case rate but lower case-fatality rate is documented, affirming that the Omicron variant is more contagious but less virulent [6]. Nuances also existed between the population groups who were infected or died before and during the Omicron surge. We observed the mitigation of racial, socioeconomic, and partisan disparities in COVID-19 health outcomes during the Omicron surge. Explanations for such attenuation abound, including a break of structural barriers by the high transmissible Omicron variant, successful implementation of policies tailoring to socially vulnerable populations, and earlier achievement of herd immunity among disadvantaged groups due to their greater infection and death tolls at the early stage of the pandemic.

Contrary to evidence for a strong clinical efficacy of COVID-19 vaccines [3], [40], our population-level regression-based estimates documented that COVID-19 vaccines did not show significant effectiveness against infections from the Omicron variant. Time-varying mediation analysis further substantiated that the effectiveness of vaccines against infections varied substantially throughout the pandemic. The effectiveness against infections was broadly significant before the Omicron surge. However, pronounced reductions in effectiveness were observed coinciding with the surging prevalence of each new variant, along with a waning trend over a 5–6-month duration. During the Omicron surge, the relationship between vaccination rate and case rate even became significantly positive. Fortunately, the relationship between vaccination rate and case-fatality rate stayed negative during the Omicron surge, although a reduction in effectiveness was also observed. Similar conclusions were documented in recent cohort studies [2], [15], [23], [50], [51], [54]. Vaccination is still crucial, but its effectiveness may need to be enhanced by developing boosters with new antigenic composition against the Omicron variant.

Another main conclusion from our mediation analysis is that human mobility plays an important role in COVID-19 health outcomes, particularly regarding COVID-19 infections. County-level human mobility broadly presented significantly positive relationships with infection rates over the whole pandemic. Meanwhile, human mobility mediates the direct effects of vaccines on infections, averagely accounting for a 10.276 % decrease in vaccine effectiveness. Such mediation effect was even higher during the Omicron surge, leading to a 13.709 % decrease in effectiveness. Findings are in line with recent studies claiming that mobility was a significant determinant of COVID-19 infections during the Omicron surge in most US counties [24], [23]. These findings imply that sole reliance on vaccination as a primary strategy to mitigate COVID-19 needs to be carefully re-examined. Targeted NPIs need to be put in place alongside vaccination until herd immunity is achieved. Such course correction is paramount, especially considering the emerging evidence on real-world vaccine failure against infections from the Omicron variant and the likelihood of future variants [2].

Several limitations are recognized and deserve further research. First, most variables used in our study were calculated at an aggregate level. Thus, conclusions drawn from this study should not be extrapolated to individuals due to the potential ecological fallacy. Also, considering the modifiable area unit problem, a county-level analysis may gloss over details existing at a more localized level. Finer-grained studies to check result robustness are warranted when related data become accessible [29]. Second, our analysis relies on population-level measures to draw associations and cannot prove causality. The establishment of causality is challenging in an observational setting and across nationwide coverage due to various unobserved confounding effects. Third, the number of cases, deaths, and vaccination rate might not accurately reflect real situations due to the adoption of self-test kits, inequitable distribution of healthcare resources, and different jurisdictional reporting rules and issues in data collection and synchronization. Finally, it should be noted that aggregating all different activities to assess human mobility may not be the most appropriate approach, given the varying impacts of different activities on COVID-19 health outcomes. For instance, trips to crowded public spaces like restaurants and bars may pose a higher risk of infection. Therefore, future research should consider categorizing human mobility by activities to more accurately capture the activity-specific relationships between human mobility and COVID-19 infections.

Appendix A.

See Fig. A1 , Table A1 , Table A2 , Fig. A2 .

Table A1.

Estimations of booster rate and at-least-one-dose rate.

Parametric coefficients
Variables	Booster rate	At-least-one-dose rate
(Intercept)	−0.035 (−0.228, 0.158)	0.042 (−0.104, 0.189)
Racial/ethnic groups
Asian	0.065 (0.034, 0.097)***	0.021 (−0.013, 0.055)
African American	−0.217 (−0.264, −0.171)***	−0.180 (−0.231, −0.130)***
Hispanic	−0.087 (−0.126, −0.047)***	0.116 (0.074, 0.159)***
Minorities	−0.076 (−0.107, −0.045)***	0.089 (0.056, 0.123)***
Industry types
Recreation & Food	−0.069 (−0.096, −0.042)***	0.055 (0.025, 0.084)***
Health Care	0.011 (−0.014, 0.036)	0.075 (0.048, 0.102)***
Retail	−0.009 (−0.030, 0.013)	0.036 (0.013, 0.060)**
Utilities	−0.032 (−0.054, −0.010)**	−0.008 (−0.032, 0.016)
Education	−0.019 (−0.047, 0.010)	−0.007 (−0.038, 0.023)
Manufacture	−0.026 (−0.058, 0.006)	−0.029 (−0.063, 0.006)
Scientific	−0.015 (−0.052, 0.023)	0.019 (−0.022, 0.059)
Administration	−0.005 (−0.029, 0.020)	0.013 (−0.014, 0.040)
Information	−0.004 (−0.026, 0.018)	0.016 (−0.007, 0.040)
Socioeconomics
Median Income	0.064 (0.021, 0.108)**	0.203 (0.156, 0.250)***
No Insurance	0.034 (−0.005, 0.074).	0.027 (−0.016, 0.069)
No Vehicle	−0.027 (−0.066, 0.011)	−0.099 (−0.140, −0.057)***
Multi-unit House	−0.060 (−0.098, −0.021)**	0.026 (−0.016, 0.068)
Mobile Home	−0.055 (−0.087, −0.022)***	−0.028 (−0.063, 0.008)
Crowd Home	0.015 (−0.018, 0.049)	0.069 (0.033, 0.105)***
Group Quarters	−0.015 (−0.050, 0.020)	−0.037 (−0.075, 0.001).
Demographics
Male	−0.011 (−0.045, 0.023)	0.068 (0.031, 0.105)***
Age over 65	0.226 (0.187, 0.265)***	0.146 (0.103, 0.189)***
Age under 18	−0.066 (−0.109, −0.022)**	−0.013 (−0.060, 0.035)
Population Density	−0.085 (−0.128, −0.041)***	−0.104 (−0.151, −0.056)***
Urbanized Population	−0.038 (−0.070, −0.005)*	0.003 (−0.033, 0.038)
Partisanship
Democrat	0.264 (0.214, 0.315)***	0.554 (0.500, 0.609)***
Weather
Temperature	−0.129 (−0.209, −0.050)**	0.030 (−0.052, 0.112)
Precipitation	−0.005 (−0.035, 0.026)	−0.008 (−0.041, 0.025)
Smooth terms
	e.d.f.	e.d.f.
s(STFIPS)	43.752***	42.531***
ti(Latitude, Longitude)	7.428	7.013
Model fit
R-sq.(adj)	0.718	0.667

Notes: Coefficients are standardized, and all interpretations are based on the unit of variable’s standard deviation. Robust 95 % confidence intervals are in parentheses. Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1. s() means the spline function. ti() means the marginal nonlinear interaction function.

Table A2.

Time-varying direct, indirect, and total effects among COVID-19 health outcomes, vaccination, and human mobility.

Outcome: Case rate; Mediator: Human mobility; Predictor: Fully vaccinated rate
Date	Total Effect Estimate	Direct Effect Estimate	Percentage
2/21/2021	−0.040 (−0.067, −0.013)	−0.040 (−0.067, −0.013)	0.000
2/28/2021	−0.053 (−0.076, −0.030)	−0.053 (−0.076, −0.030)	0.000
3/7/2021	−0.074 (−0.098, −0.051)	−0.073 (−0.094, −0.052)	1.750
3/14/2021	−0.088 (−0.127, −0.049)	−0.086 (−0.121, −0.051)	2.125
3/21/2021	−0.099 (−0.142, −0.056)	−0.099 (−0.136, −0.061)	0.882
3/28/2021	−0.109 (−0.139, −0.079)	−0.109 (−0.135, −0.083)	0.531
4/4/2021	−0.121 (−0.157, −0.085)	−0.122 (−0.153, −0.091)	0.569
4/11/2021	−0.124 (−0.158, −0.090)	−0.127 (−0.158, −0.097)	2.428
4/18/2021	−0.124 (−0.151, −0.097)	−0.130 (−0.151, −0.109)	4.660
4/25/2021	−0.134 (−0.210, −0.059)	−0.141 (−0.209, −0.073)	4.795
5/2/2021	−0.115 (−0.148, −0.083)	−0.126 (−0.148, −0.104)	8.750
5/9/2021	−0.123 (−0.173, −0.073)	−0.135 (−0.180, −0.090)	8.593
5/16/2021	−0.125 (−0.144, −0.107)	−0.125 (−0.144, −0.107)	0.000
5/23/2021	−0.098 (−0.167, −0.029)	−0.109 (−0.171, −0.047)	10.437
5/30/2021	−0.093 (−0.127, −0.060)	−0.104 (−0.124, −0.085)	10.557
6/6/2021	−0.085 (−0.115, −0.056)	−0.095 (−0.122, −0.068)	10.713
6/13/2021	−0.065 (−0.078, −0.052)	−0.076 (−0.094, −0.059)	15.313
6/20/2021	−0.064 (−0.098, −0.030)	−0.076 (−0.112, −0.040)	15.567
6/27/2021	−0.066 (−0.097, −0.034)	−0.083 (−0.100, −0.066)	20.774
7/4/2021	−0.066 (−0.091, −0.041)	−0.082 (−0.103, −0.061)	19.359
7/11/2021	−0.080 (−0.106, −0.055)	−0.092 (−0.116, −0.069)	13.027
7/18/2021	−0.100 (−0.136, −0.064)	−0.105 (−0.139, −0.072)	5.109
7/25/2021	−0.118 (−0.152, −0.085)	−0.118 (−0.152, −0.085)	0.000
8/1/2021	−0.133 (−0.148, −0.118)	−0.133 (−0.148, −0.118)	0.000
8/8/2021	−0.148 (−0.176, −0.121)	−0.148 (−0.176, −0.121)	0.000
8/15/2021	−0.154 (−0.193, −0.115)	−0.154 (−0.193, −0.115)	0.000
8/22/2021	−0.181 (−0.204, −0.157)	−0.181 (−0.204, −0.157)	0.000
8/29/2021	−0.189 (−0.234, −0.143)	−0.189 (−0.234, −0.143)	0.000
9/5/2021	−0.194 (−0.216, −0.173)	−0.194 (−0.216, −0.173)	0.000
9/12/2021	−0.183 (−0.212, −0.154)	−0.183 (−0.212, −0.154)	0.000
9/19/2021	−0.188 (−0.247, −0.129)	−0.188 (−0.247, −0.129)	0.000
9/26/2021	−0.170 (−0.199, −0.141)	−0.170 (−0.199, −0.141)	0.000
10/3/2021	−0.138 (−0.152, −0.123)	−0.138 (−0.152, −0.123)	0.000
10/10/2021	−0.095 (−0.114, −0.076)	−0.095 (−0.114, −0.076)	0.000
10/17/2021	−0.065 (−0.097, −0.033)	−0.065 (−0.097, −0.033)	0.000
10/24/2021	−0.045 (−0.069, −0.022)	−0.045 (−0.069, −0.022)	0.000
10/31/2021	−0.046 (−0.075, −0.018)	−0.046 (−0.075, −0.018)	0.000
11/7/2021	−0.051 (−0.086, −0.015)	−0.051 (−0.086, −0.015)	0.000
11/14/2021	−0.033 (−0.055, −0.012)	−0.035 (−0.053, −0.016)	3.727
11/28/2021	0.102 (0.071, 0.132)	0.073 (0.043, 0.102)	40.270
12/5/2021	0.198 (0.171, 0.226)	0.161 (0.125, 0.197)	23.152
12/12/2021	0.206 (0.177, 0.235)	0.176 (0.146, 0.206)	17.101
12/19/2021	0.202 (0.158, 0.246)	0.181 (0.148, 0.213)	11.714
12/26/2021	0.177 (0.158, 0.196)	0.177 (0.158, 0.196)	0.000
1/3/2022	0.148 (0.111, 0.184)	0.148 (0.111, 0.184)	0.000
1/10/2022	0.130 (0.080, 0.181)	0.130 (0.080, 0.181)	0.000
1/17/2022	0.101 (0.097, 0.105)	0.101 (0.097, 0.105)	0.000
1/24/2022	0.084 (0.031, 0.137)	0.084 (0.031, 0.137)	0.000
1/31/2022	0.083 (0.029, 0.137)	0.083 (0.029, 0.137)	0.000
2/7/2022	0.076 (0.026, 0.125)	0.076 (0.026, 0.125)	0.000
2/14/2022	0.055 (0.034, 0.076)	0.055 (0.034, 0.076)	0.000
Outcome: Case−fatality rate; Mediator: Human mobility; Predictor: Fully vaccinated rate
Date	Total Effect Estimate	Direct Effect Estimate	Percentage
3/21/2021	0.029 (0.008, 0.051)	0.029 (0.008, 0.051)	0.000
3/28/2021	0.041 (0.012, 0.069)	0.041 (0.012, 0.069)	0.000
4/11/2021	0.063 (0.044, 0.082)	0.063 (0.044, 0.082)	0.000
4/18/2021	0.050 (0.020, 0.080)	0.050 (0.020, 0.080)	0.000
4/25/2021	0.048 (0.021, 0.075)	0.048 (0.021, 0.075)	0.000
5/9/2021	0.045 (0.010, 0.080)	0.045 (0.010, 0.080)	0.000
6/27/2021	0.051 (0.000, 0.102)	0.051 (0.000, 0.102)	0.000
8/8/2021	−0.021 (−0.040, −0.001)	−0.029 (−0.043, −0.015)	28.304
9/12/2021	−0.053 (−0.082, −0.023)	−0.053 (−0.082, −0.023)	0.000
9/19/2021	−0.049 (−0.071, −0.026)	−0.049 (−0.071, −0.026)	0.000
9/26/2021	−0.045 (−0.079, −0.011)	−0.045 (−0.079, −0.011)	0.000
10/3/2021	−0.073 (−0.097, −0.050)	−0.073 (−0.097, −0.050)	0.000
10/17/2021	−0.065 (−0.100, −0.030)	−0.065 (−0.100, −0.030)	0.000
10/24/2021	−0.089 (−0.123, −0.054)	−0.089 (−0.123, −0.054)	0.000
10/31/2021	−0.074 (−0.140, −0.009)	−0.074 (−0.140, −0.009)	0.000
11/7/2021	−0.086 (−0.116, −0.055)	−0.086 (−0.116, −0.055)	0.000
11/14/2021	−0.090 (−0.119, −0.061)	−0.090 (−0.119, −0.061)	0.000
11/21/2021	−0.079 (−0.099, −0.060)	−0.079 (−0.099, −0.060)	0.000
11/28/2021	−0.108 (−0.125, −0.091)	−0.108 (−0.125, −0.091)	0.000
12/5/2021	−0.149 (−0.191, −0.108)	−0.149 (−0.191, −0.108)	0.000
12/12/2021	−0.151 (−0.184, −0.118)	−0.151 (−0.184, −0.118)	0.000
12/19/2021	−0.124 (−0.156, −0.093)	−0.124 (−0.156, −0.093)	0.000
12/26/2021	−0.095 (−0.117, −0.074)	−0.095 (−0.117, −0.074)	0.000
1/10/2022	−0.012 (−0.024, −0.000)	−0.029 (−0.033, −0.026)	59.129
2/7/2022	−0.046 (−0.074, −0.018)	−0.046 (−0.074, −0.018)	0.000
2/14/2022	−0.045 (−0.078, −0.012)	−0.045 (−0.078, −0.012)	0.000

Notes: Coefficients are standardized, and all interpretations are based on the unit of variable’s standard deviation. Robust 95 % confidence intervals are in parentheses. All reported effects are with P-value < 0.01.

Data availability

Data will be made available on request.