{"paper_id":"7e0476d4-8154-4a6f-afdf-aa4dfab89e11","body_text":"The impact of COVID-19 vaccination in the US: averted burden of SARS-COV-2-\nrelated cases, hospitalizations and deaths\nTeresa K. Yamana1*, Marta Galanti1, Sen Pei1, Manuela Di Fusco2, Frederick J. Angulo3, Mary \nM. Moran3, Farid Khan3, David L. Swerdlow3, Jeffrey Shaman1*\n1Department of Environmental Health Sciences, Mailman School of Public Health,\nColumbia University, 722 West 168th Street, New York, NY 10032\n2Health Economics and Outcomes Research, Pfizer, 235 East 42nd Street, NY 10017\n3Medical Development and Scientific/Clinical Affairs, Pfizer Vaccines, 500 Arcola Road, \nCollegeville PA 19426\n*Correspondence to: T.K.Y. (tky2104@cumc.columbia.edu) and  J.S. \n(jls106@cumc.columbia.edu)\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \nNOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.\n\nAbstract\nBy August 1, 2022, the SARS-CoV-2 virus had caused over 90 million cases of COVID-19 and \none million deaths in the United States.  Since December 2020, SARS-CoV-2 vaccines have \nbeen a key component of US pandemic response; however, the impacts of vaccination are not \neasily quantified.  Here, we use a dynamic county-scale metapopulation model to estimate the \nnumber of cases, hospitalizations, and deaths averted due to vaccination during the first six \nmonths of vaccine availability.  We estimate that COVID-19 vaccination was associated with \nover 8 million fewer confirmed cases, over 120 thousand fewer deaths, and 700 thousand fewer \nhospitalizations during the first six months of the campaign.  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nIntroduction\nBy August 1, 2022, SARS-CoV-2, the virus responsible for the COVID-19 pandemic, had \ncaused over 90 million cases and 1 million deaths in the United States (1).  While these numbers \nare likely affected by the widespread availability of SARS-CoV-2 vaccines, the precise impact of \nvaccination on the burden of COVID-19 disease is uncertain.  Here we use a dynamic model, \ncoupled with historical data, statistical inference methods, and hospitalization costs, to quantify \nthe clinical and economic burdens of infections, hospitalizations, and deaths averted due to \nvaccination in the US, both cumulatively and in individual states, during the first approximately \nsix months of vaccine availability when the wild type and alpha variants of SARS-CoV-2 were \nthe predominant drivers of infection.\nIn mid-December 2020, the first SARS-CoV-2 vaccine received emergency use authorization in \nthe US and was initially recommended for healthcare workers and long-term care facility \nresidents, followed by adults aged 65 years and older, adults aged 16-64 with high-risk medical \nconditions and essential workers (2). By early April 2021, the vaccine recommendation was \nextended to the general population aged 16 years and older. Subsequent steps have seen \nrecommended vaccine use for 12-15 year-olds (May 2021) and 5-11 year-olds (November 2021). \nThree different vaccines (two mRNA vaccines and one antiviral vector vaccine) with varying \nefficacy and estimates of duration of protection have been authorized for use in the US. \nHowever, vaccination delivery has been variable: it was initially limited by vaccine availability, \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nwith roughly 15 million doses provided in the first month, but reached a peak of roughly 90 \nmillion doses administered during April, 2021(1).\nBy early November 2021, 78% of the US population aged 12 years and older had received at \nleast one dose of a SARS-CoV-2 vaccine, with heterogeneous distribution across age groups \n(97% of adults aged 65+ years vs 60% of persons aged 12-18 years) and across states (<65%  in \nAL, ID, IN, LA, MS, ND, TN, WY, WV, compared with >90% in CT, MA, HI, VT, PA [2]). \nDuring the time period of vaccine rollout, variable levels of non-pharmaceutical interventions \n(NPIs), such as social distancing, closures of restaurants and bars, mask mandates and travel \nrestrictions, were implemented across states with different start and end dates. \nHere, we use a dynamic county-scale metapopulation model, previously used COVID-19 \ninference and projections (3-5), to conduct counterfactual simulations representing the effects of \nvaccination.  These simulations are used to estimate the number of cases, hospitalizations, and \ndeaths averted due to vaccination during the first six months of vaccine roll out.  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nMethods\nWe used a metapopulation model with a Susceptible-Exposed-Infected-Recovered (SEIR) \nstructure run at the county level, coupled with a data assimilation method (EAKF, the ensemble \nadjustment Kalman filter). We have previously used this framework for inference, forecasting \nand projections of influenza and SARS-CoV-2 infections at various locations and spatial scales \n(3-5). Here, we simulated SARS-CoV-2 transmission within and among the 3142 counties of the \nUnited States (see Supplementary Material).\nSpecifically, we first used the model-inference system to fit reported case counts in each county \nof the US [3] from the time of identification of the first COVID-19 cases in the United States in \nFebruary 2020, through December 14, 2020, the date of first authorized SARS-CoV-2 \nvaccination in the US. The inferred values of parameters and state variables on December 14, \n2020 served as initial conditions for the averted burden analysis. \nWe then included vaccination in the dynamical model structure using documented daily rates of \nvaccine administration (1, 6) (see Supplementary Materials).  State-level daily vaccination data \nfrom the CDC COVID Data Tracker (1) were allocated proportionally to each county based on \npopulation size. Within each county, we assumed equal probability of vaccination regardless of \nprior infection status. We modeled the vaccine as producing direct effects only, with 90% \neffectiveness against infection (7-9) – i.e. 90% of vaccinated individuals with no prior immunity \nwere fully protected while the remaining 10% receive no protection. Specifically, 90% of \nvaccinated individuals with no prior immunity were removed from the Susceptible pool and \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nplaced in the Recovered compartment 24 days after administration of the first dose. In the \nRecovered compartment, we did not distinguish between vaccinated individuals and individuals \nrecovered from infection; given uncertainty and limited data on re-infections and waning, both \nwere considered immune for the remainder of the simulation period. With the 24-day delay, the \nimpact of vaccinations on the simulation begins on January 8th. This baseline scenario, \nretrospectively fitted to case counts, enabled estimation of the daily timeseries of \nepidemiological parameters, including Rt, the time-varying reproductive number, for each county \nlocation from December 14, 2020 through June 3, 2021. \nWe ran the simulations through June 3, 2021 to focus on the impact of vaccination prior to the \npredominance of the Delta and Omicron variants (1). Given that the higher transmissibility and \nimmune escape properties of the Delta and Omicron variants require substantial additional \nmodifications of the dynamical model structure, as well as re-parametrization, we restricted our \nanalysis to the December 14, 2020 through June 3, 2021, or the pre-Delta, time period, during \nwhich the vaccine provided strong protection against infection.  To quantify the burden averted \nby vaccination, we compared the baseline vaccination scenario to 3 counterfactual no-\nvaccination scenarios simulated over the same time period.  All counterfactual scenarios \nassumed no vaccinations (or, equivalently, 0% vaccine effectiveness) but varied transmissibility \nto mimic different levels of non-pharmaceutical intervention (NPI) response in the absence of \nvaccination:\n Counterfactual Scenario 1; A no-transmission-change, no-vaccination scenario in which \nthe Rt daily time series for each location was as inferred for the baseline scenario;\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\n Counterfactual Scenario 2: A no-vaccination scenario in which Rt for each location-day \nwas increased 10% with respect to the baseline scenario; and\n Counterfactual Scenario 3: A no-vaccination scenario in which Rt for each location-day \nwas decreased 10% with respect to the baseline scenario.\nThese counterfactual scenarios represent potential population behaviors and policies that might \nhave been effected in the absence of vaccination. Scenario 3 represents increased NPIs through \npolicies and individual action; Scenario 2 represents a decrease of NPIs, perhaps due to \npandemic fatigue. We compared cumulative SARS-COV-2 cases in the 3 no-vaccination \nscenarios to the baseline scenario at national and state levels, analyzed differences in averted \ncases among states, and identified factors correlating with vaccination success.\nHospitalizations and Deaths \nTo calculate hospitalizations and deaths in the counterfactual scenarios, we made the assumption \nthat excess cases would have continued to lead to hospitalizations and deaths at the same overall \nrate as they did in each state during the summer and fall of 2020, prior to vaccine availability.  \nWe applied a state-specific pre-vaccine Case Hospitalization Rate (CHR) and a Case Fatality \nRate (CFR) multiplier to the total number of averted cases in each scenario. CHR was calculated \nas the number of COVID-19 hospitalizations divided by cases during August 1 – December 14, \n2020. Hospitalization data were compiled from the HHS dataset (10) and cases from the Johns \nHopkins Center for Systems Science and Engineering (JHU CSSE) COVID-19 data set (11, 12).  \nAugust 2020 was the first full month with all states reporting daily COVID-19 hospitalizations. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nCFR was calculated as the number of COVID-19 deaths divided by the number of cases in each \nstate during August 1 - December 14, 2020, using the JHU CSSE COVID-19 Data (11). We \nexcluded deaths and cases prior to August 1, 2020, for consistency with the hospitalization data \nset, and because both the ascertainment rate (fraction of true infections that are reported as \nconfirmed cases) and the infection fatality rate (fraction of true infections that resulted in death) \nwere unstable during the initial wave of the pandemic (4).   \nHospitalization Costs\nWe calculated averted hospitalization costs by multiplying the distribution of estimated COVID-\n19 associated hospitalizations averted by the distribution of costs per hospitalization episode, \nobtained from the US-based Premier Healthcare COVID-19 claims database (13). The median \n(interquartile range Q1-Q3) cost per hospitalization episode was $12,046 ($6,309-$25,361).\nResults\nInitialization\nAt the start of the simulation period, December 14, 2020, it was estimated 74.1% (95% credible \ninterval: 70.2 – 78.6) of the US population was susceptible, 0.8% (95% CrI 0.6-1.2%) exposed, \n0.8% (95% CrI 0.6-1.0%) infectious and 24.3% (95% CrI 19.2-28.6%) recovered.  Figure 1 \nshows the distribution of the estimated epidemiological parameters across states at the beginning \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nof vaccine administration. The median estimated susceptible fraction, corresponding to the \nfraction of the population that had not yet been infected since the beginning of the pandemic \nvaried by state and ranged from 58% (95% CrI 56%-61%) in North Dakota to 94% (93%-95%) \nin Vermont. The susceptible fraction was highest in northwestern and northeastern states.  The \ntime dependent reproductive number ranged from median 0.8 (0.7-1.4 95% CrI) in Minnesota to \n2.0 (1.7-2.3 95% CrI) in Tennessee. The CFR prior to the start of vaccination varied from 0.5% \nin Alaska (95% CrI 0.3-0.7%) to 2.3% (95% CrI 2.1-2.6%) in Rhode Island.  The CHR in the \nsame period ranged from 3.8% (95% CrI 3.6-4.0%) in Alaska to 20.7% in Kentucky (95% CrI \n20.5-20.9%). While there was a modest reduction in CHR at the national scale from the pre-\nvaccine period (8.8% August 1, 2020 – December 14, 2020) to the analysis period (7.8% \nDecember 15, 2020 – June 2, 2021), we did not observe consistent population level differences \nin CFR at the national level (1.5% during both time periods) nor to CHR and CFR at the state \nlevel.  \nFigure 1 Caption: Upper Left: Population susceptibility, S (proportion of the population \nnot yet infected), at the start of vaccine administration; Upper Right: Time-varying \nreproductive number, Rt, at the start of vaccine administration; Lower Left: State-specific \ncase hospitalization rate, CHR; Lower Right: State-specific case fatality rate, CFR. Color \nscales show the median values.\nModel results\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nBetween December 14 and June 3, 2021, the baseline model estimated 16.1 million (95% CrI \n15.1 – 18.3 million) total cases, 1.4 million (95% CrI 1.3 – 1.6 million) hospitalizations, and \n246.7 thousand (95% CrI 230.4 – 279.6 thousand) deaths. These estimates were generally \nconsistent with data reported by CDC COVID Data Tracker (1).\nThe time series of Rt resulting from fitting the baseline scenario from December 14, 2020 \nthrough June 4, 2021 is shown at the state and national level in Supplementary Figure S2. Note \nthat Rt in this analysis refers to the time-varying basic reproductive number, not to be confused \nwith the effective reproductive number Reff(t), which is Rt multiplied with the fractional \nsusceptible population.  \nBy June 4, 2021, 51% of the population in the US had received at least one dose of vaccine (1). \nVaccine coverage differed widely by location, ranging from 35% of the population in Mississippi \nup to 74% in Vermont. The weekly number of vaccinations administered increased over time: \ninitially at less than 5 million vaccinated per week but reaching a peak of 14 million vaccinated \nper week in April when vaccination was extended to the general population aged 16 and older \n(Figure S2).\nTable 1 reports the cumulative averted COVID-19 cases, deaths, hospitalizations and \nhospitalization cost savings for the 3 scenarios, while Figure 2 shows the counterfactual \nscenarios and trends.\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nTable 1: Total cumulative COVID-19 cases, deaths, hospitalizations and hospitalization cost \nsavings.\nScenario 1\nNo change in \ntransmission\nScenario 2\n10% higher \ntransmission\nScenario 3\n10% lower \ntransmission\nCases averted 8.1 million 17.0 million -1.6 million\nmedian, (95% CrI) (-4.8, 26.3 million) (1.5, 32.0 million) (-8.8, 18.2 million)\nDeaths averted 123.2 thousand 260.1 thousand -25.1 thousand\nmedian, (95% CrI) (-74.3, 403.0 thousand) (23.0, 489.7 thousand) (-134.8, 278.8 thousand)\nHospitalizations averted 0.7 million 1.5 million -0.1 million\nmedian, (95% CrI) (-0.4, 2.3 million) (0.1, 2.8 million) (-0.8, 1.6 million)\nHospitalization cost \nsavings $7.0 billion $17.3 billion -$0.9 billion\nmedian, (95% CrI) ($-11.9, 112.0 billion) ($0.9, 170.3 billion) (-$44.7, 70.1 billion)\nFigure 2 Caption: Modeled total COVID-19 Cases in Counterfactual Scenarios 1 (top \npanel), 2 (middle panel) and 3 (bottom panel) in the United States.  The black line \npresents observed cases, the blue line indicates the median counterfactual projection, and \nthe blue shaded area shows the 95% credible interval\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nIn the scenario with no change in transmission, we estimated that vaccination averted 8.1 million \ncases at the national level (median value, 95% CrI :[-4.8, 26.3] millions cases), 123.2 thousand \ndeaths (median value, 95% CrI. :[-74.3, 403.0] thousand deaths) and 0.7 million hospitalizations  \n(median value, 95% CrI :[-0.4, 2.3] millions hospitalizations). The median cost savings \nassociated with averted hospitalization was $7.0 billion (median value, 95% CrI:[-$11.9, \n$112.0]) (see Table 1). \nIncreasing Rt by 10% with no vaccination in Counterfactual Scenario 2 roughly doubled the \nmedian cases averted nationally whereas decreasing Rt by 10% with no vaccination in \nCounterfactual Scenario 3 considerably reduced the averted burden during the approximately 6 \nmonths of analysis (Figure 2 and Table 1).  In effect, the decreased Rt, representing increased \nNPIs, initially offsets the effects of no vaccination during the first 3 months when a more limited \npercentage of the population is effectively vaccinated.  However, this effect decreases in mid-\nMarch as vaccination rates climb in the baseline scenario, and by May more cases are produced \nper day in Counterfactual Scenario 3 due to the absence of vaccination.\nIn all three counterfactual scenarios, the majority of averted cases occurred between April and \nJune 2021 (Figure 2). For individual states, the median estimates of cases averted ranged from \nroughly1000 to 6400 cases per hundred thousand population (Figure 3). Median cumulative \naverted hospitalizations varied from 74 to 752 per hundred thousand and median cumulative \naverted deaths varied from 16 to 128 per hundred thousand.  Higher averted case burden \ncorrelated with higher vaccination rate (R2=0.16) and higher population susceptibility at the \nbeginning of the vaccination campaign (R2=0.21).\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nFigure 3 Caption: Total per capita averted cases (a), hospitalizations (b) and deaths (c) \nin each state between December 14, 2020 and June 2, 2021. The x-axis is the percent \npopulation vaccinated by June 2, 2021, and the y-axis is the averted \ncases/hospitalizations/deaths per 100,000 people.  Each state is represented by a dot; the \ncolor scale of the dots indicate the estimated fraction of population susceptible at the \nbeginning of vaccine rollout.\nDiscussion\nEvaluating the population-level impact of COVID-19 vaccination through mathematical \nmodeling can provide useful insights to policy makers. Here, we leveraged a validated dynamical \nmodeling approach, previously used for research and operationally to simulate county-level \nCOVID-19 transmission, to quantify the additional burden of disease in alternate scenarios \nwithout vaccination. Our analyses show that under unchanged NPI levels, COVID-19 \nvaccination in the US cumulatively prevented 8.3 million cases, 681 thousand hospitalizations \nand 118 thousand deaths in the first 6 months of implementation. States with high vaccination \ncoverage such as Maine averted as many as 6,000 cases per 100,000 individuals.\nThese simulations confirm findings from other modeling studies, set in the US and elsewhere, \nthat have found a substantial impact of vaccination in terms of averted burden of disease. \nShoukat et al  found that vaccination was fundamental for reducing the spring/summer wave in \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nNYC by reducing cases by one third and hospitalization and deaths by half during December \n2020 through July 2021 (14). Vilches et al showed that vaccination may have averted more than \n14 million cases, 241 thousand deaths and 1.1 million hospitalizations in the US by late June \n2021(15). Moghadas et al. found an even stronger effect of vaccination with 26 million cases, 1.2 \nmillion hospitalizations and 279,000 deaths averted through the end of June 2021 (16). These \nthree studies are in general agreement with our findings and indicate the US would have \nexperienced a substantial wave of infections beginning in March/April 2021 in the absence of a \nvaccine (14-16).  Large effects were found also in Israel by Haas et al. where two thirds of \nhospitalizations and deaths were averted with vaccination in the first four months of vaccine \nimplementation (17). \nOur study augments prior research in this field by providing further geographical granularity. \nThe state-level analyses provide a dynamic picture revealing trends and differences in the public \nhealth response to the COVID-19 pandemic, which may be informative for state and local \npolicymakers. Additionally, our study presents estimates of cost savings associated with vaccine-\npreventable severe disease (i.e. hospitalizations). It showed that the benefits of vaccination due to \nreduced hospitalization loads translated into cost-savings in the billions of dollars. Vaccination \nmay also lessen other societal impacts associated with the pandemic (e.g. work productivity \nloss). The total economic impact may therefore be even greater than reported, and further studies \nelucidating those impacts are warranted.\nOur study also adds to the existing literature by considering 3 counterfactual scenarios, all \nwithout vaccinations, but with varying Rt, that mimic different possible population responses to \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\ndisease spread in the absence of a vaccine. The first counterfactual scenario is designed to \nquantify what the SARS-COV-2 related burden would have been without vaccinations if the \npopulation had maintained the same NPI measures as occurred with vaccination. The other two \ncounterfactual scenarios are designed to explore the uncertainties of these estimates, as it is \ndifficult to anticipate the public policy and population behavior response in the absence of a \nvaccine. Specifically, Counterfactual Scenario 2 represents a stronger relaxation of NPIs, \npossibly due to pandemic fatigue in the absence of an available vaccine, while Counterfactual \nScenario 3 represents a reinforcement of NPIs during the 6 months of projections, assuming that \nthe population would have responded with increased measures to control transmission. \nCounterfactual 3 shows that in the early months of the vaccine rollout, an increase in NPIs could \nhave produced an even greater reduction of disease compared to vaccination as it occurred.  \nHowever, while increased NPIs may have slowed transmission in the short term (the first months \nof vaccine rollout), those measures would not have been as effective as vaccination once the \nAlpha variant became established in the United States (Figure 2).  The benefits of vaccination are \nseen in the difference between Counterfactual Scenario 3 and the baseline curve during the last \nmonth of simulation. \nAll 3 scenarios show that vaccination benefits were limited during the early months of vaccine \nrollout, and that most of the averted burden was realized in the last 2 months of the analyzed \nperiod. The winter peak of COVID-19 cases was reached in the US during mid-January 2021 just \nwhen the first vaccinations started to become effective. Vaccine availability constraints during \nthe first months of the campaign restricted administration to portions of the population with \nincreased risk of exposure and severe disease. It was not until April 2021 that vaccination was \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nrecommended for the general population aged >16 years. The combination of an initially slower \nrate of vaccination and a decreasing trend in transmission, with some states having a significant \nproportion of the population no longer susceptible to infection, narrowed the overall averted \nburden in the first months of 2021. Some exceptions occurred in states with larger initial \nsusceptible fractions (e.g. Vermont); for these states the averted burden per hundred thousand \nwas already significant in the early months of the vaccination campaign. \nBy March 2021, the Alpha variant, a SARS-CoV-2 strain with increased transmissibility relative \nto the wild type, became the predominant circulating serotype (1). This variant, combined with \nprogressive relaxation of NPIs in most states, likely produced the increase of Rt inferred at this \ntime.  Simultaneously, the impact of vaccination, seen in the divergence between the baseline \nscenario and the no-vaccination scenario case curves, becomes much more evident at the national \nlevel (Figure 2).\nA limitation of this analysis is that it relies on assumptions about whether and how the \nparameters inferred from the true observed course of the pandemic would have changed in the \nabsence of a vaccine.  Our primary counterfactual, Scenario 1, assumed that the parameters – \nincluding the disease transmission rates and the case ascertainment rate – would have been the \nsame with or without a vaccine.  We explored some of the sensitivity to this assumption by \naltering the time-varying reproductive number in Counterfactual Scenarios 2 and 3.  However, \nthese are very simplified representations, and one could just as well imagine dramatically \ndifferent counterfactual scenarios.\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nWe limit our analysis to a relatively short projection time: the first six months of the vaccination \ncampaign. Our estimates are therefore not generalizable to the entire period of the vaccination \ncampaign. In subsequent months, booster doses, the expansion of the Pfizer vaccine to children \naged 5-11, waning immunity, and the establishment of the more virulent Delta and immune-\nevading Omicron variants have made estimation of vaccine effects more challenging.  These \nlater phases of the pandemic driven by new variants led to tens of millions of Covid-19 \ninfections.  Nevertheless, these early averted cases were crucial, as this period was prior to the \nwidespread availability of antiviral medication, and with substantially lower population \nimmunity against severe outcomes.  \nAdditional assumptions should also be noted. The model structure is parsimonious and does not \nexplicitly represent certain factors including population age structure, breakthrough infections or \nreinfections. We used a constant case hospitalization rate (CHR) and case fatality rate (CFR) for \neach state, computed based on COVID-19 outcomes during the 6 months before vaccination, to \ncalculate counterfactual hospitalization and deaths in all scenarios.  These choices ignore \ndifferences in age-specific behavior and probability of severe outcomes.    \nWe also note that the full effect of COVID-19 vaccination on hospitalizations and deaths derives \nfrom two effects: those averted due to averted cases; and those averted due to improved \noutcomes in vaccinated individuals if infected.  The estimates of averted hospitalizations and \ndeaths in this analysis are restricted to the effect of averted cases and do not include reductions in \nthe probability of hospitalization and death among the vaccinated if infected. Each of the \napproved COVID-19 vaccines has been shown to be highly effective in preventing severe \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\noutcomes in individuals infected by SARS-CoV-2.  Here, we assumed that the contribution of \nthe second effect was relatively small compared to the first, as the vaccines were shown to be \nhighly effective at preventing infections in the short term after inoculation and against the strains \ncirculating at the time of the study (18-20).\nIn conclusion, our analysis shows that COVID-19 vaccination reduced the burden of disease. \nBase case results indicate that COVID-19 vaccination was associated with over 8 million fewer \nconfirmed cases, over 120 thousand fewer deaths, and 700 thousand fewer hospitalizations in the \nfirst six months of the campaign. As such, COVID-19 vaccines represented a critical component \nof the public health response to the COVID-19 pandemic in the US.\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nAcknowledgements\nThis study was sponsored by Pfizer Inc.  TKY, MG, SP and JS are employees of Columbia \nUniversity, which received funding from Pfizer in connection with the development of this study \nand of this manuscript.  JS and Columbia University disclose partial ownership of SK Analytics. \nJS discloses consulting for BNI. MDF, FJA, MMM, and FK are employees of Pfizer and may \nhold stock or stock options. DS was employed at Pfizer at the time this work was conducted \nand he may own stock or stock options.\n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\nReferences\n1. CDC. CDC Covid Data Tracker 2021 [Available from: https://covid.cdc.gov/covid-data-\ntracker/.\n2. Dooling K. The Advisory Committee on Immunization Practices’ updated interim \nrecommendation for allocation of COVID-19 vaccine—United States, December 2020. MMWR \nMorbidity and mortality weekly report. 2021;69.\n3. Li R, Pei S, Chen B, Song Y, Zhang T, Yang W, et al. Substantial undocumented \ninfection facilitates the rapid dissemination of novel coronavirus (SARS-CoV-2). Science. \n2020;368(6490):489-93.\n4. Pei S, Yamana TK, Kandula S, Galanti M, Shaman J. Burden and characteristics of \nCOVID-19 in the United States during 2020. Nature. 2021;598(7880):338-41.\n5. Pei S, Kandula S, Yang W, Shaman J. Forecasting the spatial transmission of influenza in \nthe United States. Proceedings of the National Academy of Sciences. 2018;115(11):2752-7.\n6. Kinsey M. Covid19 vaccine timeseries 2021 [Available from: \nhttps://github.com/mattkinsey/covid19-vaccine-timeseries.\n7. Polack FP, Thomas SJ, Kitchin N, Absalon J, Gurtman A, Lockhart S, et al. Safety and \nEfficacy of the BNT162b2 mRNA Covid-19 Vaccine. New England Journal of Medicine. \n2020;383(27):2603-15.\n8. Baden LR, El Sahly HM, Essink B, Kotloff K, Frey S, Novak R, et al. Efficacy and \nSafety of the mRNA-1273 SARS-CoV-2 Vaccine. New England Journal of Medicine. \n2020;384(5):403-16.\n9. Sadoff J, Gray G, Vandebosch A, Cárdenas V, Shukarev G, Grinsztejn B, et al. Safety \nand Efficacy of Single-Dose Ad26.COV2.S Vaccine against Covid-19. New England Journal of \nMedicine. 2021;384(23):2187-201.\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\n10. Health UDo, Services H. COVID-19 reported patient impact and hospital capacity by \nfacility 2020 [Available from: https://healthdata.gov/Hospital/COVID-19-Reported-Patient-\nImpact-and-Hospital-Capa/g62h-syeh.\n11. CSSE J. COVID-19 Data Repository by the Center for Systems Science and Engineering \n(CSSE) at Johns Hopkins University 2020 [Available from: \nhttps://github.com/CSSEGISandData/COVID-19.\n12. Dong E, Du H, Gardner L. An interactive web-based dashboard to track COVID-19 in \nreal time. The Lancet Infectious Diseases. 2020;20(5):533-4.\n13. Di Fusco M, Shea KM, Lin J, Nguyen JL, Angulo FJ, Benigno M, et al. Health outcomes \nand economic burden of hospitalized COVID-19 patients in the United States. Journal of \nMedical Economics. 2021;24(1):308-17.\n14. Shoukat A, Vilches TN, Moghadas SM, Sah P, Schneider EC, Shaff J, et al. Lives saved \nand hospitalizations averted by COVID-19 vaccination in New York City: a modeling study. The \nLancet Regional Health-Americas. 2022;5:100085.\n15. Vilches TN, Moghadas SM, Sah P, Fitzpatrick MC, Shoukat A, Pandey A, et al. \nEstimating COVID-19 Infections, Hospitalizations, and Deaths Following the US Vaccination \nCampaigns During the Pandemic. JAMA network open. 2022;5(1):e2142725-e.\n16. Moghadas S, Sah P, Fitzpatrick MC, Shoukat A, Pandey A, Vilches TN, et al. COVID-19 \ndeaths and hospitalizations averted by rapid vaccination rollout in the United States. medRxiv. \n2021.\n17. Haas EJ, McLaughlin JM, Khan F, Angulo FJ, Anis E, Lipsitch M, et al. Infections, \nhospitalisations, and deaths averted via a nationwide vaccination campaign using the Pfizer–\nBioNTech BNT162b2 mRNA COVID-19 vaccine in Israel: a retrospective surveillance study. \nThe Lancet Infectious Diseases. 2021.\n18. Thompson MG, Burgess JL, Naleway AL, Tyner HL, Yoon SK, Meece J, et al. Interim \nEstimates of Vaccine Effectiveness of BNT162b2 and mRNA-1273 COVID-19 Vaccines in \nPreventing SARS-CoV-2 Infection Among Health Care Personnel, First Responders, and Other \nEssential and Frontline Workers - Eight U.S. Locations, December 2020-March 2021. MMWR \nMorbidity and mortality weekly report. 2021;70(13):495-500.\n19. Shrotri M, Krutikov M, Palmer T, Giddings R, Azmi B, Subbarao S, et al. Vaccine \neffectiveness of the first dose of ChAdOx1 nCoV-19 and BNT162b2 against SARS-CoV-2 \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\ninfection in residents of long-term care facilities in England (VIVALDI): a prospective cohort \nstudy. The Lancet Infectious Diseases. 2021;21(11):1529-38.\n20. Saciuk Y, Kertes J, Mandel M, Hemo B, Shamir Stein N, Ekka Zohar A. Pfizer-\nBioNTech vaccine effectiveness against Sars-Cov-2 infection: Findings from a large \nobservational study in Israel. Preventive Medicine. 2022;155:106947.\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint \n\n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 25, 2022. ; https://doi.org/10.1101/2022.09.23.22280281doi: medRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}