Geographic accessibility predicts pediatric preventive care utilization, including vaccine uptake. However, spatial inequities in the pediatric coronavirus disease 2019 (COVID-19) vaccination rollout remain underexplored. We assessed the spatial accessibility of vaccination sites and analyzed predictors of vaccine uptake.
In this cross-sectional study of pediatric COVID-19 vaccinations from the US Vaccine Tracking System as of July 29, 2022, we described spatial accessibility by geocoding vaccination sites, measuring travel times from each Census tract population center to the nearest site, and weighting tracts by their population demographics to obtain nationally representative estimates. We used quasi-Poisson regressions to calculate incidence rate ratios, comparing vaccine uptake between counties with highest and lowest quartile Social Vulnerability Index scores: socioeconomic status (SES), household composition and disability (HCD), minority status and language (MSL), and housing type and transportation.
We analyzed 15 233 956 doses administered across 27 526 sites. Rural, uninsured, white, and Native American populations experienced longer travel times to the nearest site than urban, insured, Hispanic, Black, and Asian American populations. Overall Social Vulnerability Index, SES, and HCD were associated with decreased vaccine uptake among children aged 6 months to 4 years (overall: incidence rate ratio 0.70 [95% confidence interval 0.60–0.81]; SES: 0.66 [0.58–0.75]; HCD: 0.38 [0.33–0.44]) and 5 years to 11 years (overall: 0.85 [0.77–0.95]; SES: 0.71 [0.65–0.78]; HCD: 0.67 [0.61–0.74]), whereas social vulnerability by MSL was associated with increased uptake (6 months–4 years: 5.16 [3.59–7.42]; 5 years–11 years: 1.73 [1.44–2.08]).
Pediatric COVID-19 vaccine uptake and accessibility differed by race, rurality, and social vulnerability. National supply data, spatial accessibility measurement, and place-based vulnerability indices can be applied throughout public health resource allocation, surveillance, and research.
The coronavirus disease 2019 pandemic has disproportionately impacted marginalized children and adults across the United States. Geographic accessibility predicts preventive care utilization, including vaccine uptake, especially among children. However, spatial inequities in the pediatric coronavirus disease 2019 vaccination rollout remain underexplored.
In this study of 15 253 956 doses across 27 526 sites, vaccine uptake and vaccination site accessibility differed significantly by race, rurality, and social vulnerability. Spatial accessibility and social vulnerability measures can help equitably allocate scarce pediatric resources, including vaccinations.
The coronavirus disease 2019 (COVID-19) pandemic has had devastating and disproportionate impacts on the well-being of marginalized and minoritized communities across the United States.1–4 Yet, crucial pandemic mitigation efforts, including widespread access to testing, clinical trials, treatments, and vaccinations have not been equitably distributed to socially and clinically vulnerable adults.5–13 Children from marginalized and minoritized communities have also faced disparate impacts across the COVID-19 care continuum, including inequities in rates of COVID-19 infection and COVID-19-related hospitalization, ICU admission, complication (eg, multisystem inflammatory syndrome), mortality, and loss of a primary caregiver.14–19
In October 2021, the US Food and Drug Administration (FDA) authorized emergency use of the Pfizer-BioNTech COVID-19 vaccination for children and adolescents aged >5 years, with additional approval of both the Pfizer-BioNTech and Moderna vaccines in June 2022 for children aged >6 months.20,21 Although there has been a dearth of disaggregated data on the pediatric COVID-19 vaccination rollout, white and Asian American children had higher vaccination rates than Black and Hispanic children during the first 2 months of vaccination rollout.18,22 Parents of Black and Hispanic children reported greater vaccine hesitancy for their children than parents of white children.23,24 Geographic variability also persists; although 76% of parents in rural areas identified their child’s pediatrician as a trusted source of vaccination information, nearly 40% (compared with only 8% in urban areas) noted their pediatrician did not recommend a COVID-19 vaccine.25
Spatial accessibility is a crucial predictor of health care access, utilization, and outcomes.26–28 Importantly, children who must travel farther are less likely to access needed preventive, acute, and pandemic-related care,29–31 including childhood vaccinations,32 suggesting that pediatric preventive care may be especially sensitive to geographic access barriers. Previous work has highlighted stark geographic inequities in access to COVID-19 testing, clinical trials, monoclonal antibody treatment, and oral antiviral “test-to-treat” sites for adults.5,7,8,33 Our team has also developed methods for identifying “vaccine deserts” and has modeled relationships between vaccine uptake and place-based vulnerability indices among adults.11,34,35 However, crucial research gaps remain among children.
Inequities in pediatric vaccine access and uptake have been theorized as reflecting multifactorial and structural underpinnings, including differential convenience (ie, proximity and affordability), confidence (ie, trust in vaccine safety, medical providers, and policies/politics), and complacency (ie, perceptions of risk and urgency).36–38 Yet, studies to date have largely not characterized pediatric vaccination site locations, disparities in site accessibility, or the association of place-based vulnerability indices with pediatric vaccine uptake.
With these gaps in mind, we analyzed national childhood COVID-19 vaccine distribution and dispensing data by vaccine type and target age group. We sought to describe tract-level vaccination site accessibility and county-level vaccine uptake. We hypothesized that accessibility would vary by sociodemographic characteristics, and vaccine uptake would differ significantly by measures of social vulnerability. Overall, our study aimed to provide policy-relevant insights by refining novel methods for the identification of priority populations and geographic areas using vaccine distribution and dispensing data.
Methods
This cross-sectional study follows the Strengthening the Reporting of Observational Studies in Epidemiology reporting guideline and was exempt from institutional review board approval in accordance with 45 Code of Federal Regulations, Section 46. Statistical analyses were performed using R, version 4.2.3 (R Foundation for Statistical Computing, Vienna, Austria). Data sources are briefly summarized in Supplemental Table 3; of note, publicly available data sources reflected the most recent or most appropriate releases available during data collection.
Study Sample
We obtained COVID-19 vaccine distribution locations from the Centers for Disease Control and Prevention’s (CDC) Vaccine Tracking System (VTrckS).39 VTrckS is the most comprehensive US vaccination supply and distribution data source because it is the ordering platform for all 64 public health jurisdictions and for pharmacies and practices receiving federally purchased vaccines. This includes the Vaccines for Children program for low-income families, in which an estimated 86% of US pediatricians participate, and the Federal Retail Pharmacy Program, through which 46% of doses were administered 3 months into the pediatric COVID-19 vaccine rollout.32,40–43 We used National Drug Codes, which identify vaccines on the basis of specific FDA approval indications44 to track distribution of the Pfizer-BioNTech vaccine for children 6 months through 4 years old, the Moderna vaccine for children 6 months through 4 years old, and the Pfizer-BioNTech vaccine for children 5 through 11 years old.
We extracted data for all VTrckS distribution sites in the contiguous United States that had dispensed pediatric COVID-19 vaccines as of July 29, 2022. Vaccination sites were already geocoded to their street address when obtained from VTrckS and had been validated by the organizations administering vaccines themselves on the basis of their receipt of VTrckS deliveries. To determine doses administered for each National Drug Code, we subtracted each day’s reported supply from the previous day’s supply level. We considered decreases as doses administered and increases as a resupply. This approach to estimate vaccine administration levels from supply data has been validated through identification of vaccine deserts and quasi-experimental association of supply changes with vaccination rates and clinical outcomes.11,45
Analysis #1: Geographic Accessibility of Pediatric COVID-19 Vaccination Sites by Rurality, Race, Ethnicity, and Uninsurance
We used vaccination site geolocations within the VTrckS data set to identify corresponding Census tract Federal Information Processing System codes. We linked our data with tract-level population demographics for children <18 years old from the 2017–2021 US Census American Community Survey46 and tract-level rurality from the 2010 US Department of Agriculture Rural–Urban Commuting Area codes.47
For each vaccination age group, we modeled spatial accessibility with a travel time-based approach, which is widely used in spatial access to care research,48,49 is superior to distance-based approaches,11,49 and has been validated previously by our team.5,7,8,11,33 We calculated travel times using a friction surface and Dijkstra’s algorithm.11 In brief, the friction surface divides the United States into a grid of 1 km2 cells, each with a transversal time influenced by factors such as road networks and public transportation availability.11,50 Travel times were estimated using Dijkstra’s algorithm to calculate the time from each cell to the nearest pediatric vaccine facility.11,51 This approach assumes that individuals take the most efficient possible ground route (eg, car, train), and that road networks and public transport systems operate at optimal conditions. Our methods do not account for the probabilistic nature of real-world travel (eg, traffic patterns), but are a computationally efficient way to determine approximate travel time to multiple nearby sites.
Each tract’s travel time was defined as the median drive time. We then calculated the population-weighted proportion of each demographic subgroup residing in a census tract within x minutes of the nearest site, including the proportion living in a vaccine desert >30 minutes away (defined based on federal standards for access to primary care52 ). We derived demographics (ie, uninsurance, race, ethnicity) using American Community Survey data and stratified by rurality using Rural-Urban Commuting Area codes. Lastly, we characterized subgroup travel times as the median and interquartile range across all tracts by weighting with the subgroup N from each tract.
Analysis #2: County-Level Pediatric COVID-19 Vaccination Uptake by Rurality and Social Vulnerability
We aggregated vaccine supply and doses administered to the county level, because most sites serve a larger catchment area than their census tract. Pediatric vaccine uptake was defined as the number of doses administered divided by the number of children <5 years old or 5 to 11 years old within each county. Age-based population denominators were derived from vintage 2019 bridged-race postcensal population estimates,53 mirroring methods from a recent CDC study of vaccine coverage.54
Building upon related research,2,4,35,55 we assessed the relationship between area-level social vulnerability and pediatric vaccination uptake in urban and rural areas. We defined county-level social vulnerability using the 2020 CDC Social Vulnerability Index (SVI)56 and county-level rurality using the 2013 US Department of Agriculture Rural-Urban Continuum Codes.57 The SVI is a composite index derived from US Census data across 4 domains: socioeconomic status (SES), household composition and disability (HCD), minority status and language (MSL), and housing type and transportation (HTT). Index values range from 0 (least vulnerable) to 1 (most vulnerable); we coalesced SVI domains into quartiles for analysis.
We used quasi-Poisson regressions to examine the association of each SVI domain with county-level vaccination uptake per capita. Vaccine uptake was overdispersed (the variance exceeded the mean, requiring an alternative to Poisson models); we tested both quasi-Poisson and negative binomial models with similar primary findings, prioritizing quasi-Poisson models herein to allow greater influence of counties with the most doses administered.58 Models included state fixed effects to account for heterogeneity in state vaccine rollouts and pandemic policy choices, and were weighted by the number of people less <5 years old or 5 to 11 years old in the county. In secondary analyses, we stratified tracts by rurality. We reported incidence rate ratios with 95% confidence intervals to reflect the relative increase or decrease in vaccine uptake between the highest and lowest quartile counties by overall social vulnerability and each SVI domain.
Results
From October 2021 to July 2022, we identified 27 526 unique vaccination providers who administered 271 589 doses of the Pfizer-BioNTech 6 months–4 years vaccine, 6270 doses of the Moderna 6 months–4 years vaccine, and 14 956 097 doses of the Pfizer-BioNTech 5 years–11 years vaccine.
Accessibility of Nearest Vaccination Site by Race, Ethnicity, and Rurality
The Pfizer-BioNTech 5 years–11 years vaccine approved first by the FDA was more widely available across the country, with fewer vaccine desert tracts that faced 1-way drive times >30 minutes to the nearest site (Fig 1). Overall, 2.0% of the US population, 2.7% of uninsured children, 10.5% of rural children, 13.2% of American Indian and Alaska Native (AIAN) children, 2.0% of white children, 2.2% of Hispanic children, and 1.2% of Black children lived >30 minutes from the nearest 5 years–11 years vaccine site. In contrast, 13.7% of the US population, 65.9% of rural children, 15.3% of uninsured children, 25.3% of AIAN children, 14.5% of white children, 11.8% of Hispanic children, and 9.0% of Black children lived >30 minutes from the nearest 6 months–4 years vaccine site.
Geospatial visualization of 1-way travel time to the nearest COVID-19 vaccination site by age group. (A) 6 months to 4 years old; (B) 5 years to 11 years old.
Geospatial visualization of 1-way travel time to the nearest COVID-19 vaccination site by age group. (A) 6 months to 4 years old; (B) 5 years to 11 years old.
Shorter travel times were centered around large urban areas (Fig 1). Rural children had longer drive times than urban children across all demographic subgroups and both vaccine age groups, with particularly pronounced differences in accessibility of 6 months–4 years vaccination sites (Fig 2). Compared with white children (weighted median travel time 4.8 minutes [weighted interquartile range 2.2–16.5]), AIAN children (13.5 [2.9–63.8]) lived farther from the nearest 6 months–4 years vaccination site, whereas Asian (2.2 [1.2–3.9]), Hispanic (2.7 [1.5–5.4]), and Black (3.4 [1.8–6.5]) children lived closer. Similarly, compared with white children (weighted median: 2.3 [1.2–6.5]), AIAN children (4.1 [1.4–22.1]) lived farther from the nearest 5 years–11 years vaccination site, whereas Asian American (4.1 [1.4–22.1]), Hispanic (1.3 [0.8–2.4]), and Black (1.6 [0.9–2.9]) children lived closer.
One-way travel time to the nearest COVID-19 vaccination site by uninsurance, race, ethnicity, rurality, and age group. (A) 6 months to 4 years old; (B) 5 years to 11 years old.
One-way travel time to the nearest COVID-19 vaccination site by uninsurance, race, ethnicity, rurality, and age group. (A) 6 months to 4 years old; (B) 5 years to 11 years old.
Social Vulnerability and County-Level Vaccine Uptake
The 6 months–4 years and 5 years–11 years vaccines were administered in 27.6% (N = 866, out of 3142 total US counties) and 77.5% (N = 2435) of US counties, respectively. Descriptive characteristics of the 3105 US counties included in our analysis (after excluding 37 counties in Alaska, Hawaii, or with missing data for a key variable) are presented in Table 1. The number of children increased by SVI quartile, with more children living in higher vulnerability counties (analysis of variance P < .001). A larger proportion of the highest SVI counties were rural compared with the other 3 quartiles (quartile 4 68.3% versus quartile 1 61.2%, quartile 2 60.1%, quartile 3 61.1%; χ2 test P < .001).
Descriptive Analysis of County-Level Pediatric COVID-19 Vaccination Site Accessibility and Vaccination Uptake by SVI Quartile
| Q1, Least Vulnerable (N = 779) | Q2 (N = 776) | Q3 (N = 774) | Q4, Most Vulnerable (N = 775) | Overall (N = 3104) | |
|---|---|---|---|---|---|
| Number of people <5 y old | |||||
| Mean (SD) | 3090 (6170) | 5800 (13 100) | 6880 (18 600) | 9300 (33 500) | 6260 (20 600) |
| Median [min, max] | 967 [11, 67 500] | 1630 [140, 127 000] | 1700 [16, 2 760 000] | 1730 [29, 580 000] | 1490 [11, 580 000] |
| Number of people 5–11 y old | |||||
| Mean (SD) | 4790 (9530) | 8580 (19 200) | 9900 (26 600) | 13 200 (47 400) | 9100 (29 300) |
| Median [min, max] | 1490 [19, 98 700] | 2340 [30, 179 000] | 2460 [26, 411 000] | 2470 [45, 840 000] | 2220 [19, 840 000] |
| Rurality,aN (%) | |||||
| Urban | 303 (38.8%) | 310 (39.9%) | 301 (38.9%) | 246 (31.7%) | 1160 (37.4%) |
| Rural | 477 (61.2%) | 466 (60.1%) | 473 (61.1%) | 529 (68.3%) | 1945 (62.6%) |
| SVIb domain, mean (SD) | |||||
| SESc | 0.15 (0.11) | 0.39 (0.15) | 0.62 (0.14) | 0.84 (0.11) | 0.50 (0.29) |
| HCDd | 0.24 (0.19) | 0.40 (0.22) | 0.59 (0.23) | 0.77 (0.20) | 0.50 (0.29) |
| MSLe | 0.27 (0.18) | 0.40 (0.23) | 0.53 (0.25) | 0.79 (0.19) | 0.49 (0.29) |
| HTTf | 0.21 (0.18) | 0.43 (0.22) | 0.58 (0.22) | 0.77 (0.18) | 0.50 (0.29) |
| Drive time to nearest 6 mo–5 y site, N (%) | |||||
| <15 min | 107 (13.7%) | 115 (14.8%) | 102 (13.2%) | 76 (9.8%) | 400 (12.9%) |
| 15–<30 min | 165 (21.2%) | 163 (21.0%) | 137 (17.7%) | 93 (12.0%) | 558 (18.0%) |
| 30–60 min | 230 (29.5%) | 252 (32.5%) | 267 (34.5%) | 290 (37.4%) | 1039 (33.5%) |
| 60–<120 min | 141 (18.1%) | 158 (20.4%) | 194 (25.1%) | 250 (32.3%) | 743 (23.9%) |
| 120+ min | 136 (17.5%) | 88 (11.3%) | 74 (9.6%) | 66 (8.5%) | 364 (11.7%) |
| Drive time to nearest 5 y–11 y site, N (%) | |||||
| <15 min | 297 (38.1%) | 311 (40.1%) | 302 (39.0%) | 251 (32.4%) | 1161 (37.4%) |
| 15–<30 min | 246 (31.5%) | 301 (38.8%) | 308 (39.8%) | 335 (43.2%) | 1190 (38.3%) |
| 30–60 min | 158 (20.3%) | 117 (15.1%) | 127 (16.4%) | 145 (18.7%) | 547 (17.6%) |
| 60–<120 min | 68 (8.7%) | 44 (5.7%) | 33 (4.3%) | 41 (5.3%) | 186 (6.0%) |
| 120+ min | 11 (1.4%) | 3 (0.4%) | 4 (0.5%) | 3 (0.4%) | 21 (0.7%) |
| 6 mo–5 y vaccination doses per 1000 peopleg | |||||
| Mean (SD) | 5.27 (15.6) | 7.15 (32.1) | 5.35 (19.4) | 3.53 (17.6) | 5.33 (22.1) |
| Median [min, max] | 0 [0, 206] | 0 [0, 583] | 0 [0, 386] | 0 [0, 420] | 0 [0, 583] |
| 5 y–11 y vaccination doses per 1000 peopleh | |||||
| Mean (SD) | 27.6 (24.9) | 23.4 (19.5) | 23.1 (19.6) | 24.9 (20.9) | 24.7 (21.4) |
| Median [min, max] | 18.0 [0, 166] | 16.9 [0, 144] | 17.3 [0, 188] | 18.5 [0, 280] | 17.8 [0, 280] |
| Q1, Least Vulnerable (N = 779) | Q2 (N = 776) | Q3 (N = 774) | Q4, Most Vulnerable (N = 775) | Overall (N = 3104) | |
|---|---|---|---|---|---|
| Number of people <5 y old | |||||
| Mean (SD) | 3090 (6170) | 5800 (13 100) | 6880 (18 600) | 9300 (33 500) | 6260 (20 600) |
| Median [min, max] | 967 [11, 67 500] | 1630 [140, 127 000] | 1700 [16, 2 760 000] | 1730 [29, 580 000] | 1490 [11, 580 000] |
| Number of people 5–11 y old | |||||
| Mean (SD) | 4790 (9530) | 8580 (19 200) | 9900 (26 600) | 13 200 (47 400) | 9100 (29 300) |
| Median [min, max] | 1490 [19, 98 700] | 2340 [30, 179 000] | 2460 [26, 411 000] | 2470 [45, 840 000] | 2220 [19, 840 000] |
| Rurality,aN (%) | |||||
| Urban | 303 (38.8%) | 310 (39.9%) | 301 (38.9%) | 246 (31.7%) | 1160 (37.4%) |
| Rural | 477 (61.2%) | 466 (60.1%) | 473 (61.1%) | 529 (68.3%) | 1945 (62.6%) |
| SVIb domain, mean (SD) | |||||
| SESc | 0.15 (0.11) | 0.39 (0.15) | 0.62 (0.14) | 0.84 (0.11) | 0.50 (0.29) |
| HCDd | 0.24 (0.19) | 0.40 (0.22) | 0.59 (0.23) | 0.77 (0.20) | 0.50 (0.29) |
| MSLe | 0.27 (0.18) | 0.40 (0.23) | 0.53 (0.25) | 0.79 (0.19) | 0.49 (0.29) |
| HTTf | 0.21 (0.18) | 0.43 (0.22) | 0.58 (0.22) | 0.77 (0.18) | 0.50 (0.29) |
| Drive time to nearest 6 mo–5 y site, N (%) | |||||
| <15 min | 107 (13.7%) | 115 (14.8%) | 102 (13.2%) | 76 (9.8%) | 400 (12.9%) |
| 15–<30 min | 165 (21.2%) | 163 (21.0%) | 137 (17.7%) | 93 (12.0%) | 558 (18.0%) |
| 30–60 min | 230 (29.5%) | 252 (32.5%) | 267 (34.5%) | 290 (37.4%) | 1039 (33.5%) |
| 60–<120 min | 141 (18.1%) | 158 (20.4%) | 194 (25.1%) | 250 (32.3%) | 743 (23.9%) |
| 120+ min | 136 (17.5%) | 88 (11.3%) | 74 (9.6%) | 66 (8.5%) | 364 (11.7%) |
| Drive time to nearest 5 y–11 y site, N (%) | |||||
| <15 min | 297 (38.1%) | 311 (40.1%) | 302 (39.0%) | 251 (32.4%) | 1161 (37.4%) |
| 15–<30 min | 246 (31.5%) | 301 (38.8%) | 308 (39.8%) | 335 (43.2%) | 1190 (38.3%) |
| 30–60 min | 158 (20.3%) | 117 (15.1%) | 127 (16.4%) | 145 (18.7%) | 547 (17.6%) |
| 60–<120 min | 68 (8.7%) | 44 (5.7%) | 33 (4.3%) | 41 (5.3%) | 186 (6.0%) |
| 120+ min | 11 (1.4%) | 3 (0.4%) | 4 (0.5%) | 3 (0.4%) | 21 (0.7%) |
| 6 mo–5 y vaccination doses per 1000 peopleg | |||||
| Mean (SD) | 5.27 (15.6) | 7.15 (32.1) | 5.35 (19.4) | 3.53 (17.6) | 5.33 (22.1) |
| Median [min, max] | 0 [0, 206] | 0 [0, 583] | 0 [0, 386] | 0 [0, 420] | 0 [0, 583] |
| 5 y–11 y vaccination doses per 1000 peopleh | |||||
| Mean (SD) | 27.6 (24.9) | 23.4 (19.5) | 23.1 (19.6) | 24.9 (20.9) | 24.7 (21.4) |
| Median [min, max] | 18.0 [0, 166] | 16.9 [0, 144] | 17.3 [0, 188] | 18.5 [0, 280] | 17.8 [0, 280] |
Min, minimum; max, maximum; Q, quartile.
Rurality was determined from the US Office of Management and Budget categorization of the US Department of Agriculture Economic Research Service’s 2013 Rural–Urban Commuting Codes, with a code of 1 to 2 (ie, metropolitan areas) classified as urban and a code of 3 to 12 (ie, nonmetropolitan areas) classified as rural.
SVI is an aggregate of all 4 domains, each calculated on the basis of variables from the 2016–2020 5-year US Census American Community Survey data.
SES domain includes income, poverty, employment, and education variables.
HCD domain includes dependent children <18 y of age, persons 65 and older, single-parent households, and people with disabilities.
MSL domain includes race, ethnicity, and English language proficiency variables.
HTT domain includes housing structure, crowding, and vehicle access variables.
Six-month to 5-year doses per capita were estimated by dividing the cumulative number of doses administered within each county divided by the county’s number of children aged <5 years on the basis of vintage 2019 bridged-race postcensal population estimates.
Five-year to 11-year doses per capita were estimated by dividing the cumulative number of doses administered within each county divided by the county’s number of children 5 years to 11 years old on the basis of vintage 2019 bridged-race postcensal population estimates.
In adjusted quasi-Poisson models presented in Table 2, greater social vulnerability by overall SVI, SES, and HCD was associated with decreased uptake of the 6 months–4 years vaccines (overall: incidence rate ratio 0.70 [95% confidence interval 0.60–0.81]; SES: 0.66 [0.58–0.75]; HCD: 0.38 [0.33–0.44]) and the 5 years–11 years vaccine (overall: 0.85 [0.77–0.95]; SES: 0.71 [0.65–0.78]; HCD: 0.67 [0.61–0.73]). In contrast, greater social vulnerability by MSL was associated with increased uptake in both vaccine groups (6 months–4 years: 5.16 [3.59–7.42]; 5 years–11 years: 1.73 [1.44–2.08]). For the 6 months–4 years vaccines, findings by SVI, SES, and MSL remained significant in urban areas but not rural areas, and were robust for HCD in both rural and urban areas. For the 5 years–11 years vaccine, findings by overall SVI were no longer significant after stratification by rurality, only significant in urban areas for SES, and significant across both urban and rural areas for HCD and MSL.
Pediatric COVID-19 Vaccination Uptake in the Most and Least Socially Vulnerable Counties by Rurality
| Pfizer + Moderna, 6 mo–5 y | Pfizer, 5 y–11 y | |||||
|---|---|---|---|---|---|---|
| Q1g | Q4g | IRR [95% CI]h | Q1 | Q4 | IRR [95% CI] | |
| Overall SVIa | ||||||
| Total | 9.2 (47.4) | 6.5 (33.2) | 0.70 [0.60–0.81] | 473 (22.1) | 404 (15.3) | 0.85 [0.77–0.95] |
| Urbanb | 10.8 (112.1) | 7.7 (80.3) | 0.72 [0.58–0.89] | 499 (40.8) | 436 (29.8) | 0.88 [0.74–1.04] |
| Rural | 0.3 (44.8) | 0.2 (28.8) | 0.64 [0.34–1.20] | 329 (18.7) | 328 (16.4) | 1.00 [0.85–1.17] |
| SESc | ||||||
| Total | 9.7 (50.3) | 6.4 (33.0) | 0.66 [0.58–0.75] | 534 (19.5) | 380 (14.6) | 0.71 [0.65–0.78] |
| Urban | 11.0 (115.0) | 7.9 (82.0) | 0.72 [0.59–0.86] | 563 (36.2) | 411 (28.4) | 0.73 [0.64–0.84] |
| Rural | 0.3 (43.9) | 0.2 (30.0) | 0.68 [0.35–1.32] | 360 (21.2) | 324 (16.5) | 0.90 [0.76–1.06] |
| HCDd | ||||||
| Total | 10.6 (84.0) | 4.0 (31.9) | 0.38 [0.33–0.44] | 532 (18.4) | 354 (18.4) | 0.67 [0.61–0.73] |
| Urban | 12.6 (123.1) | 5.4 (52.5) | 0.43 [0.34–0.53] | 563 (34.7) | 388 (29.3) | 0.69 [0.60–0.80] |
| Rural | 0.3 (39.8) | 0.1 (15.8) | 0.40 [0.22–0.72] | 399 (18.8) | 316 (14.5) | 0.79 [0.69–0.90] |
| MSLe | ||||||
| Total | 2.0 (16.8) | 10.1 (86.7) | 5.16 [3.59–7.42] | 284 (24.7) | 492 (18.5) | 1.73 [1.44–2.08] |
| Urban | 2.8 (29.0) | 11.0 (115.0) | 3.97 [2.05–7.69] | 300 (60.9) | 507 (34.5) | 1.69 [1.12–2.55] |
| Rural | 0.2 (32.6) | 0.1 (14.7) | 0.45 [0.20–1.02] | 265 (14.8) | 392 (22.9) | 1.48 [1.25–1.76] |
| HTTf | ||||||
| Total | 8.2 (41.9) | 8.8 (44.9) | 1.07 [0.93–1.24] | 526 (23.7) | 477 (16.5) | 0.91 [0.83–1.00] |
| Urban | 9.4 (97.4) | 10.5 (108.9) | 1.12 [0.91–1.38] | 578 (44.8) | 519 (32.3) | 0.90 [0.77–1.04] |
| Rural | 0.4 (53.2) | 0.2 (33.7) | 0.63 [0.40–1.01] | 276 (16.9) | 356 (15.1) | 1.29 [1.12–1.49] |
| Pfizer + Moderna, 6 mo–5 y | Pfizer, 5 y–11 y | |||||
|---|---|---|---|---|---|---|
| Q1g | Q4g | IRR [95% CI]h | Q1 | Q4 | IRR [95% CI] | |
| Overall SVIa | ||||||
| Total | 9.2 (47.4) | 6.5 (33.2) | 0.70 [0.60–0.81] | 473 (22.1) | 404 (15.3) | 0.85 [0.77–0.95] |
| Urbanb | 10.8 (112.1) | 7.7 (80.3) | 0.72 [0.58–0.89] | 499 (40.8) | 436 (29.8) | 0.88 [0.74–1.04] |
| Rural | 0.3 (44.8) | 0.2 (28.8) | 0.64 [0.34–1.20] | 329 (18.7) | 328 (16.4) | 1.00 [0.85–1.17] |
| SESc | ||||||
| Total | 9.7 (50.3) | 6.4 (33.0) | 0.66 [0.58–0.75] | 534 (19.5) | 380 (14.6) | 0.71 [0.65–0.78] |
| Urban | 11.0 (115.0) | 7.9 (82.0) | 0.72 [0.59–0.86] | 563 (36.2) | 411 (28.4) | 0.73 [0.64–0.84] |
| Rural | 0.3 (43.9) | 0.2 (30.0) | 0.68 [0.35–1.32] | 360 (21.2) | 324 (16.5) | 0.90 [0.76–1.06] |
| HCDd | ||||||
| Total | 10.6 (84.0) | 4.0 (31.9) | 0.38 [0.33–0.44] | 532 (18.4) | 354 (18.4) | 0.67 [0.61–0.73] |
| Urban | 12.6 (123.1) | 5.4 (52.5) | 0.43 [0.34–0.53] | 563 (34.7) | 388 (29.3) | 0.69 [0.60–0.80] |
| Rural | 0.3 (39.8) | 0.1 (15.8) | 0.40 [0.22–0.72] | 399 (18.8) | 316 (14.5) | 0.79 [0.69–0.90] |
| MSLe | ||||||
| Total | 2.0 (16.8) | 10.1 (86.7) | 5.16 [3.59–7.42] | 284 (24.7) | 492 (18.5) | 1.73 [1.44–2.08] |
| Urban | 2.8 (29.0) | 11.0 (115.0) | 3.97 [2.05–7.69] | 300 (60.9) | 507 (34.5) | 1.69 [1.12–2.55] |
| Rural | 0.2 (32.6) | 0.1 (14.7) | 0.45 [0.20–1.02] | 265 (14.8) | 392 (22.9) | 1.48 [1.25–1.76] |
| HTTf | ||||||
| Total | 8.2 (41.9) | 8.8 (44.9) | 1.07 [0.93–1.24] | 526 (23.7) | 477 (16.5) | 0.91 [0.83–1.00] |
| Urban | 9.4 (97.4) | 10.5 (108.9) | 1.12 [0.91–1.38] | 578 (44.8) | 519 (32.3) | 0.90 [0.77–1.04] |
| Rural | 0.4 (53.2) | 0.2 (33.7) | 0.63 [0.40–1.01] | 276 (16.9) | 356 (15.1) | 1.29 [1.12–1.49] |
Values provided in this table were obtained from quasi-Poisson regression models to predict the county-level number of vaccine doses administered per 1000 people <5 years old or 5 years to 11 years old by SVI quartile, adjusted for state fixed effects and population-weighted by either the county population <5 years old or 5 years to 11 years old. Statistically significant values at a threshold of P < .05 are denoted in boldface. CI, confidence interval; IRR, incidence rate ratio; Q1, least vulnerable quartile; Q4, most vulnerable quartile.
SVI is an aggregate of all 4 domains, each calculated on the basis of variables from the 2016–2020 5-year US Census American Community Survey data.
Urban and rural characteristics were determined from the US Office of Management and Budget categorization of the US Department of Agriculture Economic Research Service’s 2013 Rural–Urban Commuting Codes, with a code of 1 to 2 (ie, metropolitan areas) classified as urban and a code of 3 to 12 (ie, nonmetropolitan areas) classified as rural.
SES domain includes income, poverty, employment, and education variables.
HCD domain includes dependent children <18 years of age, persons 65 and older, single-parent households, and people with disabilities.
MSL domain includes race, ethnicity, and English language proficiency variables.
HTT domain includes housing structure, crowding, and vehicle access variables.
Q1 and Q4 are reported as vaccine doses administered per 1000 people (SE).
Incidence rate ratios represent the contrast for quartile 4/quartile 1, with 95% confidence intervals in brackets.
Discussion
In this national study of vaccination sites and doses ordered through VTrckS as of July 2022, we found substantial geographic variation in the accessibility and uptake of pediatric COVID-19 vaccines. Urban, insured, Hispanic, Black, and Asian American populations had shorter drive times to the nearest pediatric vaccination site than rural, uninsured, non-Hispanic, white, and especially AIAN populations. Counties with the highest vulnerability by overall SVI, SES, and HCD had lower uptake of both vaccines, whereas counties with the highest vulnerability by MSL had increased uptake of both vaccines. Our study contributes to the growing body of literature on differential access to COVID-19 testing, treatment, and vaccination by race/ethnicity, rurality, and social vulnerability by demonstrating that these trends persist for pediatric populations’ vaccine access and utilization.
In general, our findings corroborate and expand upon previous work on pediatric COVID-19 vaccination. Eleven weeks after launch of the pediatric COVID-19 vaccination program, a CDC analysis demonstrated that 92% of children aged 5 years to 11 years lived within 5 miles of a pediatric vaccine provider.42 Our travel time-based findings add demographic breakdowns to quantify nationwide accessibility by race, ethnicity, and rurality. In the same study, first-dose coverage was lowest in high SVI areas but improved over time42 ; this temporal trajectory is mirrored by our findings that 6 months–4 years vaccine accessibility and uptake were far worse than the 5 years–11 years vaccine. In another CDC analysis, counties with at least 1 active vaccine provider, urban counties, and low SVI counties had higher vaccination coverage among children aged 5 years to 11 years54 ; our decomposition of specific SVI domains, stratification by rurality, and inclusion of children aged 6 months to 4 years expands upon these findings. By the end of 2022, vaccine coverage among children aged 6 months to 4 years remained demonstrably worse than older groups, with only 10.1% completing their first dose nearly 5 months after age eligibility expansion.59 Our findings corroborate these trends and add a concerning layer of geographic inequity by SVI and SVI subdomains.
From an implementation standpoint, our findings point toward the need to leverage tools like the Vaccine Equity Planner (https://vaccineplanner.org/) in public health preparedness efforts. Designed to help public health officials, clinicians, employers, researchers, and the lay public simulate potential interventions and spatial inequities, the Vaccine Equity Planner identifies age-specific vaccine deserts and high-vulnerability areas to identify opportunities for improving equitable vaccination accessibility.34,55 Our findings also elevate concerns regarding unique challenges with vaccinating younger age groups. This may reflect slow “diffusion of innovation” in the early weeks after FDA approval of the 6 months–4 years vaccination and increased parental hesitancy about vaccinating younger children.60,61
Our study also reaffirms that geographic accessibility is not the standalone barrier to improving vaccine uptake. Pediatric vaccination coverage has been demonstrably higher among white children aged 5 years to 11 years compared with most other racial/ethnic groups.62 Yet, interestingly, we found that high vulnerability by the SVI–MSL domain predicted markedly greater vaccination uptake. This may reflect a concerning recapitulation: Greater vaccination among privileged groups even when sites are located within marginalized communities, because of low vaccine confidence and other access barriers disproportionately impacting marginalized populations.63,64 Moreover, early racial and spatial inequities in the pediatric vaccination rollout may grow further as pandemic-related coverage and reimbursement expansions are rescinded.65
To narrow these gaps, evidence-based, “low-tech/high-touch” approaches such as interventions that directly change behavior (eg, behavioral nudges, reminders, and employer/school vaccine requirements),66 educate parents about vaccine importance and safety,67 and partner with trusted community messengers to build vaccine confidence should be prioritized.12,68,69 Our methods and findings may be useful for prioritizing equity in the rollout of promising new interventions like nirsevimab for respiratory syncytial virus,70,71 targeting future outbreak response efforts,72 and surveilling population-level disparities in chronic pediatric conditions.73–75
Strengths and Limitations
There are important strengths to our approach. Existing research on COVID-19 vaccination uptake has largely used public state or federal data sources, which are often not age-stratified and have heterogeneous data quality/missingness, or survey data, which are prone to self-report and recall biases.76–78 In contrast, our data are obtained from a primary source. Moreover, because vaccine uptake outcomes were aggregated to the population level rather than the individual or facility level, our approach lends itself well to policy relevance and place-based allocation strategies.55,72,79
There are also limitations to consider. First, our vaccine uptake denominator should be interpreted carefully; we measured uptake at vaccine site locations rather than residential addresses. We mitigated this concern by aggregating to county-level uptake, which better reflects the population of children seeking care at each site. Second, our numerator for dose administration does not differentiate first-dose vaccination from course completion. Third, our dose administration estimates do not account for expired doses removed from the supply. However, our approach has been quasi-experimentally validated through association with COVID-19 vaccination rates, cases, and deaths.45 Fourth, doses were geolocated to their primary distribution location; secondary dissemination (eg, hospitals receiving and distributing doses to satellite clinics) may not be fully captured. Fifth, although our quasi-Poisson regression models with rurality stratification, pediatric population weighting, and state fixed effect adjustment were a theory-informed approach to identify factors associated with vaccine uptake, they do not intrinsically account for spatial autocorrelation; however, previous work identifying similar factors associated with COVID-19 vaccine deserts using spatial lag models support our overall findings.11 Lastly, although Vaccines.gov and VTrckS are the most authoritative source for US vaccine supply data, they do not include military treatment facilities, Veteran’s Affairs hospitals, and some federally qualified health centers.
Conclusions
Our study found meaningful variability in the spatial accessibility and uptake of pediatric COVID-19 vaccinations by race, ethnicity, rurality, uninsurance, and SVI domains. We refined a novel data pipeline and geospatial methods for intervention monitoring and targeting, especially in resource-constrained scenarios where equity and maximum societal benefit must be assured. Data quality limitations have long hampered efforts to improve childhood vaccine equity and coverage80 ; supply data and vulnerability indices offer a pragmatic opportunity for local, state, and federal partners to take targeted action.81 Our approaches can be applied broadly to emerging pathogen and disease response efforts.
The COVID-19 pandemic, and outbreaks of nearly every endemic infectious disease in recent memory, pose their greatest threats to our most structurally vulnerable populations. Coordinated responses to emerging pathogens at hyperlocal, regional, and national levels must prioritize health equity, because of distributive justice, because of the many intersections of health and social conditions, and because our health is inextricably linked to the health of those around us more than ever during a pandemic.
Acknowledgments
We thank Andrew Strumpf, MS, for his assistance with configuring a portion of the statistical analysis.
Dr Khazanchi conceptualized and designed the study, conducted the initial analyses, drafted the initial manuscript, and critically reviewed and revised the manuscript; Dr Rader conceptualized and designed the study, coordinated data collection, advised the analytic approach, and critically reviewed and revised the manuscript; Drs Cantor, McManus, Bravata, Weintraub, Whaley, and Brownstein conceptualized the study, and critically reviewed and revised the manuscript for important intellectual content; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
Statistical code is available upon request from Dr Khazanchi ([email protected]). Publicly available coronavirus disease 2019 vaccine data can be downloaded from data.cdc.gov. Drs Khazanchi and Rader had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
An abstract version of this work was presented at IDWeek 2023 in Boston, Massachusetts, on October 13, 2023.
FUNDING: Dr Khazanchi is supported by the Emerging Pathogens Innovator Award sponsored by the Emerging Pathogens and Epidemic Response Cluster at Boston Children’s Hospital. The funding source had no role in the design or conduct of this study. The content is solely the responsibility of the authors.
CONFLICT OF INTEREST DISCLOSURES: Dr Khazanchi reported previously serving as a health equity consultant to the New York City Department of Hygiene and Mental Health’s Coalition for Ending Racism in Clinical Algorithms and the Office of the Chief Medical Officer; currently serving as a strategic advisory board member for the Rise to Health Coalition; currently serving as a member of The Lancet Commission on Antiracism in Solidarity; and receiving unrelated grant funding from the Joel and Barbara Alpert Endowment for Children of the City and Brigham & Women’s Hospital, all outside the submitted work. Dr McManus reported previously owning stock in Gilead Sciences, Inc. The remaining authors have indicated they have no conflicts of interest relevant to this article to disclose.
- AIAN
American Indian and Alaska Native
- CDC
Centers for Disease Control and Prevention
- COVID-19
coronavirus disease 2019
- FDA
US Food and Drug Administration
- HCD
household composition and disability
- HTT
housing type and transportation
- MSL
minority status and language
- SES
socioeconomic status
- SVI
Social Vulnerability Index
- VTrckS
Vaccine Tracking System
Comments