Main

The US Food and Drug Administration first granted emergency use authorizations for vaccines against coronavirus disease 2019 (COVID-19) on 8 December 2020. On 19 April 2021, all adult US residents became eligible to receive a COVID-19 vaccine; at this point, one quarter (87 million) of the US population was fully vaccinated; 46 million people required a second shot; 70–100 million adults who would need to be vaccinated to achieve herd immunity remained unvaccinated; and around 3 million daily doses were administered1. Scarcity due to production capacity limitations and logistical constraints had become less intense than in previous months, but equitable allocation remained critical. Allocations are determined by the Centers for Disease Control and Prevention’s (CDC) 64 immunization grantees (50 states, the District of Columbia (DC), five large cities and eight territories—referred to, collectively, as jurisdictions). The CDC requested that all jurisdictions provide allocation plans by 31 October 2020 (ref. 2). To assist planners throughout the allocation process, we analyzed these plans and subsequent updates to understand to what extent they reflect a novel proposal by the National Academies of Science, Engineering and Medicine (NASEM) to promote equity not only across, but also within, phases of vaccine distribution. A summary of the main findings and limitations of the study is provided in Table 1.

Table 1 Policy summary

COVID-19 vaccine allocation relates to two main processes: providing available doses to jurisdictions according to their population or other metrics3 and then, within jurisdictions, distributing vaccines among specific populations. Allocation frameworks seek to integrate a multitude of factors. Typically, they center on a risk-based approach that aims to reduce mortality and limit the spread of infections. A central question is to what extent ongoing implementation efforts align with, or stand in conflict with, commitments to mitigate inequities, particularly those affecting economically disadvantaged racial and ethnic groups who have experienced, and continue to experience, disproportionate effects of COVID-19 (refs. 4,5,6,7,8,9).

The CDC’s Advisory Committee on Immunization Practice (ACIP) traditionally provides guidance to jurisdictions in the case of pandemics. ACIP’s overarching ethical framework for allocating COVID-19 vaccines notes that allocation strategies ‘should aim to both reduce existing disparities and avoid creating new disparities’10, echoing an earlier articulation of the committee’s scientific and ethical principles11. A similar emphasis is found in initial proposals in the academic literature12; in influential high-level policy advice by NASEM4, which is tasked by the CDC and the National Institutes of Health with assisting ACIP in developing an equitable allocation framework; as well as in the World Health Organization’s (WHO) Strategic Advisory Group of Experts on immunization13. In this study, our concept of equity is directly aligned with NASEM’s articulation. That is, we understand it as integrating both horizontal equity (requiring treating the same those who have the same needs)14 and, importantly, vertical equity (requiring treating differently those with different needs). Specifically, as NASEM exemplifies, this requires prioritizing communities that have been disproportionately impacted by COVID-19 due to persistent structural and systemic disadvantage and racism that contribute to avoidable shortfalls in health outcomes more generally4,12,15.

One way equity can be addressed is through the sequence of priority groups, which often simultaneously reflect a desire to prevent harms (such as death, hospitalization and infection) and promote equity (as risks are not distributed equally across income and racial or ethnic groups). Figure 1 shows ACIP’s and NASEM’s respective frameworks, depicting similarities and differences. Both frameworks prioritize healthcare workers equally; however, people with comorbidities are prioritized differently. Because structural racism worsens health and economic opportunity, these groups comprise larger shares of economically disadvantaged minorities than the general population5,16,17,18,19,20. NASEM places incarcerated and homeless populations before children and non-frontline critical risk workers, whereas ACIP does not address these groups specifically, instead grouping them with the general population.

Fig. 1: Priority groups under NASEM and ACIP frameworks.
figure 1

Group sizes take into account overlaps with preceding groups, so as not to double-count people who fall into more than one priority group. Group sizes vary between NASEM and ACIP because they are defined and constructed differently. NASEM’s priority groups cover 82% of the US population; ACIP’s cover 64%, or 80% of the population over 16 years of age. (for methods, see https://vaccineallocation.ariadnelabs.net/assets/Vaccine_Allocation_Planner_for_COVID19_Methods.pdf; depiction: Ariadne Labs) #Includes first responders and teachers. ^Includes pregnancy and smoking (not included in NASEM).

Importantly, NASEM also recommended the use of an additional measure to promote equity. Within “each population group, vaccine access should be prioritized for geographic areas identified through CDC’s Social Vulnerability Index (SVI) or another more specific index.”2 An index such as SVI is a statistical measure tied to a geographic area that captures the relative average advantage and disadvantage of people living there, by integrating relevant metrics such as income, educational attainment and housing quality21,22. Such indices can capture population groups for which the protection offered by vaccines is both more necessary and more valuable, as they are typically more dependent on regular income, less able to socially distance and more likely to contract and spread the infection6,7. In addition to increasing the epidemiological benefit of vaccines, disadvantage indices also address health inequities, which ACIP, NASEM and others recognize as important4,8,12,19. NASEM notes that an index such as SVI incorporates the variables that are most linked to the disproportionate impact of COVID-19 on people of color and other vulnerable populations4. To address this disparate impact, NASEM recommends setting aside 10% of federally available vaccines for vulnerable communities (as determined by SVI), to be added to the allocations that would otherwise be offered proportionately by population size23. Jurisdictions should, furthermore, make special efforts to deliver vaccines to residents in high-SVI areas (conceptualized as the 25% most disadvantaged)4. CDC staffers noted that the SVI could be integrated into Tiberius, a newly developed software. Developed by a private company, the platform is intended to assist jurisdictions with determining the shares of vaccines to be sent to particular areas within their boundaries and to track administration21. SVI-weights can be added to the allocation formulas such that more disadvantaged areas receive larger amounts. In response to data demonstrating disproportionately lower vaccine receipt by minorities after the first month of the rollout, the CDC also expressly recognized the index’s utility for monitoring coverage rates and for identifying communities where focused efforts might be required to reduce inequity24. Similarly, the Biden–Harris administration’s national COVID-19 strategy recommended that jurisdictions should use the SVI or other indices to describe how they have or will provide equitable vaccine access25. Public health planning aside, assessing coverage rates by, for example, SVI deciles can also support disparate impact monitoring, a legal concept focused on determining whether policies negatively affect a protected group, even if they do not have that express intention and do not directly use information about that group26,27,28. To ascertain the extent to which allocation guidance incorporates disadvantage indices and related measures, we, therefore, analyzed jurisdictions’ plans.

Results

The initial search (7–14 November 2020) yielded a total of 63 summaries (98.4% of all jurisdictions) and 47 full guidance documents (73.4% of all jurisdictions, including all states)29. Subsequent searches (8–14 December 2020; 21 December 2020–1 January 2021; 7–19 January 2021; and 20–30 March 2021) yielded one additional summary (total n = 64, 100%) and five additional full guidance documents (total n = 52, 81.3%). Twenty-four jurisdictions had one or more updates to full guidance documents, and, for 35, one or more supplemental documents were identified; the current review includes data for all 64 jurisdictions. Tables 2 and 3 and Figs. 2 and 3 summarize the findings from all searches. Because NASEM recommended the use of a disadvantage index, and to reduce complexity in general, we focused on indices but provide data on zip code use for further context.

Table 2 Jurisdictions’ use of disadvantage indices and zip codes for prioritizing vaccine allocation and use of Tiberius software
Table 3 Central verbatim sections illuminating jurisdictions’ uses of disadvantage indices and zip codes in combination with proxies for disadvantage, to promote equitable vaccine allocation
Fig. 2: Jurisdictions’ use of disadvantage indices and zip codes for prioritizing vaccine allocation and use of Tiberius software—a geographical depiction.
figure 2

Underline indicates that the state (including DC) is among the 20 with the largest share of disadvantaged communities. For construction of high- and low-vulnerability jurisdictions, see Supplementary Table 2.

Fig. 3: 56 Jurisdictions’ use of disadvantage indices, by share of population in vulnerable areas, population size and use of Tiberius software.
figure 3

Circle size is proportional to jurisdiction population. All population data includes people under 16 years of age (not currently incorporated in ACIP’s framework). Higher vulnerability = 25% or more of the jurisdiction’s population lives in census tracts with SVI scores in the most disadvantaged quartile (Extended Data Fig. 1). Phil, Philadelphia PA; SA, San Antonio TX; Chi, Chicago IL; Hou, Houston TX; NYC, New York City NY. Excludes the eight territories owing to lack of SVI data. (Depiction: Ariadne Labs). m, milllion.

A total of 37 jurisdictions report using a disadvantage index in different ways, applying to around 269 million residents (82%; Fig. 3). Twenty-nine jurisdictions (28 states and one city) refer specifically to the SVI. Eleven states and two cities refer to seven other established or internally developed indices: Community Vulnerability Index (CCVI, n = 5), Area Deprivation Index (ADI, n = 2), and five newly developed indices (Healthy Places Index, Pandemic Influenza Vulnerability Index, and three indices all called the ‘COVID Vulnerability Index’, but differing from one another). Five states refer to more than one index. Among the 20 high-disadvantage jurisdictions (where more than 25% of the local population are among the nationwide most disadvantaged group; Extended Data Fig. 1), 14 report using an index.

Fourteen jurisdictions (three cities and 11 states) describe zip code-based prioritizations combining the US Post Office geographic structure with different proxies for disadvantage. In several cases, more than one proxy was mentioned. By frequency, we found zip code in combination with COVID-19 incidence (n = 5); vaccine uptake rates (n = 4), COVID-19-associated mortality (n = 3); bespoke algorithms developed by the respective health department (n = 2); economic data (n = 2); general health data (n = 2); hospitalization rates (n = 1); and social data (n = 1). In five cases, the metric applied to the zip code was unclear. Colorado focuses on census tracts with the highest density of low-income and minority communities (Table 3 and Supplementary Data).

Thirty-six jurisdictions indicate using Tiberius (which, in principle, could enable applying SVI-weights to allocations), including 13 jurisdictions not otherwise signaling any use of a place-based measure.

Among the 34 states and three cities that refer to a disadvantage index, five different purposes can be distinguished. (Some jurisdictions indicate the intention to pursue more than a single goal; see overviews in Table 3 and full data in Supplementary Table 1).

In direct alignment with NASEM’s recommendation, 25 total jurisdictions (17 using SVI, seven using zip code and one using census tracts) use language that makes it clear that a place-based measure not only has public health relevance but also can be used for promoting equity. Expressly, they refer to the need to address existing inequities across racial, ethnic and income groups that are associated with poverty, deprivation and differential COVID-19 impact and burden in health and economic terms, that live in underserved areas or that belong to a historically or systematically marginalized population (AR, CA, CHI, CO, CT, DC, Houston, IN, IL, LA, MA, MI, MD, MN, NC, ND, NH, NM, NV, NYC, OH, Philadelphia, TN, TX and WI)12,30. Specific policy responses are found in jurisdictions specifying increased allocations30 or larger shares of appointments for people from more disadvantaged areas. Tennessee, the first state to mirror NASEM’s approach at the state level, reserves 5% of its Moderna vaccine allocation for high-SVI areas. New Hampshire adds 10% of its allocation to communities disproportionately impacted, drawing on SVI and CCVI. Massachusetts allocates 20% additional vaccines to communities with disproportionate COVID-19 burden and high social vulnerability. Connecticut commits to administering at least 25% of available vaccine supply to high-SVI areas. California reserves 40% of vaccines for communities in the first quartile of its index and recommends that a similar share be reserved in appointments. North Carolina reserves 30% of vaccines for purposes including equitable access for racial and ethnic minorities and requests that 40% of daily vaccinations be reserved and filled with individuals from historically marginalized populations first. Arkansas and Illinois use the CCVI to apply unspecified weights to increase allocations for more disadvantaged communities. Indiana, Michigan, Minnesota, North Dakota, Ohio and Wisconsin indicate the same, using SVI.

Eighteen jurisdictions plan to use an index to identify priority populations (AL, AK, FL, GA, Houston, KS, MD, NY, OR, PA, RI, SC, TX, VA, VT, WA and Philadelphia), which might entail increased vaccine or appointment allocations or earlier placement in the sequence of priority groups or other targeted activities.

Fifteen jurisdictions plan to use an index for promoting access: planning locations of dispensing sites (CT, LA, MI, NC, NH, NJ, Philadelphia and SD) or outreach or communication strategies (AK, AZ, CT, LA, MA, MD, MI, NC, NY, VT, WA and Philadelphia).

Finally, four states states (CA, MI, NC and OH) use an index to monitor vaccine receipt. North Carolina set the goal that minority populations should receive vaccines at least proportionate to their population share, and Michigan aims to have no differences across racial, ethnic or SVI groups.

Discussion

We highlight four aspects regarding our appraisal of the data and their broader implications: (1) variation among jurisdictions in the adoption of disadvantage indices; (2) variations in the types of uses to which such measures are put; (3) plans for the uptake of the Tiberius software; and (4) the importance of monitoring. Additionally, we note implications for adoption beyond the United States.

By March 2021, 6 months after first being asked to publish their formal allocation frameworks, 37 jurisdictions and most states (n = 34) had adopted a disadvantage index. Including zip code-based measures, roughly two-thirds of jurisdictions (n = 43) were using a place-based measure of disadvantage. The dominance of disadvantage indices over zip code approaches is likely explained by NASEM’s recommendation to use the SVI or a similar index and by the fact that combining zip codes with proxy measures for disadvantage incurs a justificatory burden for the metric chosen. Although the rapid pace of adoption is remarkable, it is not universal across jurisdictions. Place-based disadvantage measures are not the only way equity could be addressed, and jurisdictions might have alternative strategies. But scrutiny of efforts to implement ways of allocating vaccines in ways that reduce inequities will likely increase. For example, even if all states were to set aside a 10% reserve of their allotted vaccines as additional amounts for the most disadvantaged quartile, under the NASEM framework populations of color would be offered vaccines below their population share until the beginning of phase 3, except for the very first phase (see Extended Data Fig. 3, simulation for ACIP framework ongoing)3.

An important use of place-based disadvantage measures relates to the expression among vaccine workers that ‘Vaccines don’t save lives. Vaccinations save lives.’31 Setting aside larger shares of vaccines alone can be meaningless for reducing inequity if these steps are not matched with genuine and proactive efforts to make vaccines available in conveniently located and trusted settings. However, currently, only 18 jurisdictions indicate using a disadvantage index for planning the location of vaccination sites or communication and outreach efforts.

Using a rigorous measure of disadvantage for promoting receipt is of great importance in view of the overall policy that jurisdictions receive only new vaccine allocations once already received batches have been distributed32. Although entirely reasonable in its motivation to increase population protection, an unintended consequence of this policy can be that jurisdictions might prioritize regions where uptake is swift and virtually guaranteed33 and, conversely, might deprioritize locations with real or anticipated lower receipt. But interpreting low vaccine receipt in, for example, communities with predominantly Black, Hispanic or Indigenous populations as expressing that these groups might simply not be interested in vaccines due to personal reasons is shortsighted. Lacking trust in the healthcare system or government, and ongoing experiences of structural racism in healthcare and beyond, can be powerful barriers to vaccine uptake24,26,27,34. Likewise, given that the intention to be vaccinated among Black Americans has approximated that of white and Hispanic Americans (61% versus 69% versus 70%)35, lower rates of receipt by Black Americans could plausibly reflect insufficient opportunities to receive vaccines in a trusted setting36. Given that current incentive structures implicitly favor prioritizing allocations to geographic areas with the swiftest uptake, using place-based measures for targeted outreach, communication, appointment sign-up assistance and dispensing site planning are, therefore, critical, especially in jurisdictions with larger proportions of disadvantaged communities of color and others disengaged from healthcare systems. Such measures can also help mitigate the fact that populations who are more ‘internet savvy’ gain advantages in allocation systems that often rely on online reservations for vaccines37.

On a practical note (that can have normative implications), use of the Tiberius software is not universal among the jurisdictions. In principle, uniform adoption of a centralized platform to inform state plans could be helpful. Two main uses of Tiberius are consistent implementation of index-based prioritization by disadvantage and near-real-time transparency around vaccine allocation and receipt. However, Tiberius seems mainly to represent an opportunity missed, as it appears that policymakers engaging with the software had major concerns, including around the opacity of data integration and alignment of data representation with state-level datasets38.

As jurisdiction-level planners distribute vaccines in weekly batches, each delivery offers an opportunity to monitor the status of vaccine receipt by vulnerable populations and course correct when and where needed. Targets such as Michigan’s Zero Disparity goal (achieving no disparity in vaccination rates across racial and ethnic groups or by social vulnerability index) provide critical orientation and succinctly articulate a central notion of health equity that aims to allocate resources not per capita alone but also in ways that avoidable unfair differences in health outcomes are genuinely addressed. Monitoring matters throughout all allocation phases. It also gains in importance at the point where rationing seemingly ends (once vaccines are offered to the entire population). Making all residents eligible is not the same as getting everyone vaccinated. Scarcity of availabile vaccines decreases gradually, and the switch to the final phase (universal eligibility) marks a step change. Equity issues, therefore, continue to persist, as the general public, just as in all previous phases, is not a homogenous group and differs in their risks of getting and spreading the virus. Monitoring receipt (and intensifying, as appropriate, outreach, communication or dispensing site efforts) is especially important for jurisdictions that open eligibility to the general population but have below-average vaccination rates and above-average proportions of disadvantaged communities39.

Finally, although the United States stands out globally in the magnitude of disparities across racial, ethnic and income groups relative to the country’s wealth, it is certainly not the only country that struggles with inequities. As noted, WHO guidance expressly urges planners around the world to consider ways of allocating vaccines equitably within nations13. Our review was restricted to the United States but can raise the question of how feasible the adoption of disadvantage indices or other place-based measures would be in other countries. A particularly striking example is the United Kingdom (UK), which was one of the first countries to establish and implement such measures in health policy more broadly40. The UK’s Indices of Multiple Deprivation have been used for directly related budgetary resource allocations, and evaluations demonstrated their effectiveness at reducing disparities40. However, the UK’s allocation framework makes no use of it, despite an express acknowledgment of clear evidence demonstrating marked disparate impact in incidence and mortality across racial, ethnic and income groups41. At the same time, we note that a new vulnerability index specifically for guiding COVID-19-related allocation has been developed for India42. The different ways in which US planners are using place-based disadvantage measures could, therefore, be helpful for vaccination planning outside of the United States, particularly in countries with similarly pronounced patterns of disparities in health, wealth and COVID-19 impact.

Our study had some limitations. Jurisdictions were asked to publish initial allocation plans under a constrained schedule with only 30 days between the official request and the deadline. Although summary allocation plans were eventually available for all jurisdictions, full allocation plans were publicly available for 81% of jurisdictions as of 31 October 2020. After vaccine rollouts started, some jurisdictions stopped updating or publishing their original documents, and not every intended use of place-based measures might have been captured. At the same time, our analysis includes all states and demonstrates that an increasing number of jurisdictions have adopted NASEM’s recommendation to use a disadvantage index alongside a primarily risk-based framework with sequential subpopulations, despite disadvantage indices going unmentioned by ACIP. The data presented here cannot settle exhaustively the extent to which jurisdictions use indices primarily to reduce inequity (that is, in full alignment with NASEM’s social justice rationale) or for less normative reasons, which might be grounded more in epidemiology. Still, our analysis provides a historical benchmark in that at least 22 states, DC and three cities expressly note social justice considerations in explaining their use of an index (Table 2 and Supplementary Data), and, hence, indicate that reducing inequity through the use of place-based disadvantage measures is perceived to be both pressing and feasible relative to other important priorities.

The United States continues to face an unprecedented public health, logistical and social justice challenge in allocating vaccines43,44. In a major shift in designing rationing frameworks, most US states have recognized the need to promote equity within allocation phases through the use of place-based disadvantage indices and related measures. Although ongoing impatience with delays in receiving vaccines is understandable, within each priority group some communities remain more able to protect themselves from COVID-19 than others. Jurisdictions should explore, to the fullest extent, the potential of using disadvantage indices alongside other options to allocate vaccines equitably, not just within each of the priority populations but also now that vaccines are offered to the general population4,12.

Although the tasks at hand are urgent and dynamic, there is still time for jurisdiction planners to play a direct role in changing the course of a troubling historical trajectory of inequity. Allocation frameworks that increase the chances of more disadvantaged communities—and particularly those of color—to be offered a vaccine can help to reduce inequity and promote public health simultaneously6,7 and can be one way of mitigating the consequences of past, and, in many ways, still ongoing9,26,27,28,34,36, wrongs.

Methods

We obtained summaries of all jurisdictions’ allocation plans published by 8 November 2020, on the CDC’s dedicated website2. Where a document linked to full guidance, we included it in the analysis and additionally obtained full plans by searching jurisdictions’ health department websites through to 30 March 2021 (see description of sequential searches below; archived copies of all retrieved documents are available upon reasonable request). Note that four states (NY, PA, IL and TX) include cities that are themselves CDC jurisdictions (New York City, Philadelphia, Chicago, Houston and San Antonio). For analytic purposes, and in line with the CDC’s taxonomy, we captured these cities separately rather than including them in the states they fall within.

Given the rapid pace of updating evolving guidance, we supplemented the initial search with a web browser search. All researchers used the same browser, unlinked to any personal profiles, and searched for the health department name in combination with keywords corresponding to the terms emerging as pertinent from the initial search (see description under Step 3).

Plans and subsequent supplemental documents were analyzed using a nine-item extraction tool conceptualized by H.S., M.A.W. and L.G. and refined in discussions with A.S. and R.W. and eliciting:

  1. 1.

    Whether jurisdictions intended to use a disadvantage index or zip code-based measure (two items);

  2. 2.

    Whether indices were used for prioritizing disadvantaged groups through larger allocations and, if so, what share of what population should be prioritized and to what extent (two items); and whether indices were used to define priority groups for other uses of prioritization, for planning outreach and communication, for planning dispensing sites or for monitoring uptake (1 item each);

  3. 3.

    Whether planners indicate use of the Tiberius software (which might include prioritized allocations to disadvantaged areas via SVI weights) (one item).

Four authors (A.D., E.S., H.W. and N.N.) completed data extraction for all formal allocation plans and supplemental documents; each data source was analyzed in parallel by two analysts. E.S. and H.S. led data entry verification. Any changes in subsequent documents were reviewed collectively (H.S., E.S., H.W., N.N. and A.D.). Differences in data capture were marginal, given the simplicity of the extraction tool, and resolved by consensus.

Data acquisition and extraction

Step 1: Initial data analysis (7–14 November 2020)

Data extraction tool was used for all formal jurisdiction summary and full guidance documents.

H.W. and A.D.—Each completed data extraction for half of jurisdictions; E.S.—Verified all data entry; A.D., E.S., H.S. and H.W.—Resolved any differences in data capture.

Step 2: Full guidance updates (8–14 December 2020; 21 December 2020–1 January 2021; and 7–19 January 2021)

Monitoring for updates to jurisdictions’ full guidance documents. Data extraction tool was used for all updated documents.

A.D., E.S., H.W. and N.N.—Each monitored one-quarter of jurisdictions for updates to full guidance documents.

A.D., E.S., H.S., H.W. and N.N.—Resolved any differences in data capture.

Step 3: Supplemental document search (13–19 January 2021; and 20–30 March 2021)

An additional search was done to identify supplemental documents on the jurisdictions’ health department websites aside from the formal allocation plans that identified use of an index of disadvantage, a reserve system or the Tiberius platform. Google Incognito mode was used to search the following:

  1. 1.

    ‘[state website url]’ covid vaccine Reserve

  2. 2.

    ‘[state website url’ covid vaccine ‘Categorized priority system’

  3. 3.

    ‘[state website url’ covid vaccine Tiberius

  4. 4.

    ‘[state website url’ covid vaccine ‘vulnerability index’

  5. 5.

    ‘[state website url]’’ covid vaccine ‘deprivation index’

  6. 6.

    ‘[state website url]’ covid vaccine ‘equitable distribution’

For the March 20–30 supplemental document search, an additional search was added as below:

  1. 1.

    ‘[state health department name]’ covid vaccine Reserve

  2. 2.

    ‘[state health department name]’ covid vaccine ‘Categorized priority system’

  3. 3.

    ‘[state health department name]’ covid vaccine Tiberius

  4. 4.

    ‘[state health department name]’ covid vaccine ‘vulnerability index’

  5. 5.

    ‘[state health department name]’ covid vaccine ‘deprivation index’

  6. 6.

    ‘[state health department name]’ covid vaccine ‘equitable distribution’

  7. 7.

    ‘[state health department name]’ covid vaccine zip

  8. 8.

    ‘[state website url]’ covid vaccine zip

A.D., E.S., H.W. and N.N.—Each completed search as described above for one-quarter of the jurisdictions and completed data extraction for identified documents. A.D., E.S., H.S., H.W. and N.N.—Resolved any differences in data capture.

Step 4: Cross-checking

Finally, A.D., E.S., H.W. and N.N. switched sets of states to verify mentions in the documents found by the other individual. Additionally, searches were completed for the partner’s states as described in Step 3. At each of the four overall steps, H.S. and E.S. verified all data extraction for accurate capture.

Supplementary Table 1 provides all central verbatim sections of jurisdictions’ uses of disadvantage indices and zip codes. For the full extracted data, see Supplementary Data.

Data on quantifying shares of disadvantaged populations and the impact of statistical measures of disadvantage for adjusting allocations

We sought to classify jurisdictions by their degree of disadvantage. For this part of the analysis, our goal was to classify jurisdictions by the degree of disadvantage or vulnerability of their populations to compare that to their use of a disadvantage index. This analysis matters as, during the period of the analysis (November 2020–March 2021), vaccines were allocated to jurisdictions proportionate to population. If all jurisdictions had the same share of disadvantaged people, these groups would stand an equal chance of getting a vaccine. To the extent that this is not the case, equity issues become the more important: the larger the share of disadvantaged people in a jurisdiction is, the relatively worse chance they stand of receiving a vaccine.

To classify jurisdictions, modifying an earlier analysis that focused on the state, as opposed to jurisdiction level3, we started with the 2018 nationwide census tract SVI data, as made available on the CDC’s website (https://www.atsdr.cdc.gov/placeandhealth/svi/data_documentation_download.html). For the four states with city-level jurisdictions within them, we classified census tracts by county, as shown in Supplementary Table 2.

For the remaining 46 states and DC, we counted each census tract as within its state jurisdiction.

The SVI score is a percentile with uniform distribution over all census tracts in the country. A higher SVI means more disadvantage, so an SVI score of 0.75 or higher signifies that the census tract is in the highest quartile of disadvantage relative to a nationwide standard. We flagged each census tract as disadvantaged if its SVI ≥ 0.75, using the omnibus variable RPL_THEMES for SVI.

We then collapsed the census tracts to the jurisdiction level, summing up the total population of the jurisdiction as well as the population living in high-disadvantage SVI census tracts, using the population variable E_TOTPOP. Finally, for each jurisdiction, we calculated the percent of its total population living in high disadvantage areas, as shown in Extended Data Fig. 1.

As displayed, 20 jurisdictions have more than one-quarter of their population living in areas of high disadvantage, including all five cities and DC. (Alabama’s result is 24.538%, just under the 25% threshold.)

Using the pragmatic threshold of 25%, used in studies by the CDC with similar aims of assessing the impact on particularly disadvantaged populations6, half of jurisdictions with a high level of disadvantage are using an index (10/20) as are 56% of the non-disadvantaged (20/36).

The difference between these proportions is sensitive to the choice of threshold. Extended Data Fig. 2 displays the percent of jurisdictions using an index at a variety of threshold choices.

At the time NASEM recommended setting aside a 10% national reserve to be allocated to disadvantaged populations as captured under SVI, it was unclear what quantitative impact that this would have in terms of the numbers of doses offered to these communities. To quantify this, we simulated using SVI along a modified version of the index that reduced legal challenges and another index that likewise reduces this risk (the ADI)23. Extended Data Fig. 3 shows, on the left-hand side, the consequences of setting aside 10% at the state level (the more realistic approach; see the example of Tennessee, as noted in the manuscript) of the amount allocated to states based on population and adding this in addition to the share that a state’s worse-off quartile, as captured on the respective index, would receive. The right-hand side shows the consequences of doubling this amount to 20%, which can also give a rough idea of what a combined 10% reserve at the national level and at the state level would mean3.

The share of the minority populations that would be offered vaccines under the unadjusted NASEM framework in shown in the gray line. In the initial phase, all indices would offer disadvantaged populations vaccines above their population share, even though, in the case of the unadjusted NASEM framework, the margin is slim and considerably higher on the different indices. Around halfway through phase 1, using only the state-level 10% reserve (left-hand side illustration) on all scenarios, the share of offered vaccines drops below the population share, whereas increasing the reserve size to 20% leads to offers that are consistently above the population share. Note, also, the shares of COVID-related deaths (crude and age-adjusted) of all minority populations collectively that are shown for context on the vertical axis. Furthermore, note that the standardized assumptions made here set aside logistical complexities of implementation, which make it harder, rather than easier, to reach disadvantaged groups, and, for the purpose of illustration, that we (counterfactually) assume that everyone who is offered a vaccine will take one, to err on the side of not overstating our findings.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.