Article Text
PDF
Abstract
Introduction The COVID-19 pandemic triggered one of the largest global health and economic crises in recent history. COVID-19 vaccination (CV) has been the central tool for global health and macroeconomic recovery, yet estimates of CV’s global health and macroeconomic value remain scarce.
Methods We used regression analyses to measure the impact of CV on gross domestic product (GDP), infections and deaths. We combined regression estimates of vaccine-averted infections and deaths with estimates of quality-adjusted life years (QALY) losses, and direct and indirect costs, to estimate three broad value components: (i) QALY gains, (ii) direct and indirect costs averted and (iii) GDP impacts. The global value is the sum of components over 148 countries between January 2020 and December 2021 for CV generally and for Pfizer-BioNTech specifically.
Results CV’s global value was US$5.2 (95% CI US$4.1 to US$6.2) trillion, with Pfizer-BioNTech’s vaccines contributing over US$1.9 (95% CI US$1.5 to US$2.3) trillion. Varying key parameters results in values 10%–20% higher or lower than the base-case value. The largest value component was GDP impacts, followed by QALY gains, then direct and indirect costs averted. CV provided US$740 of value per dose, while Pfizer-BioNTech specifically provided >US$1600 per dose. We estimated conservative benefit-cost ratios of 13.9 and 30.8 for CV and Pfizer-BioNTech, respectively.
Conclusions We provide the first estimates of the broad value of CV incorporating GDP, QALY and direct and indirect cost impacts. Through December 2021, CV produced significant health and economic value, represented strong value for money and produced significant macroeconomic benefits that should be considered in vaccine evaluation.
- COVID-19
- global health
- vaccines
- health economics
- health policy
Data availability statement
Data not subject to license restrictions are available in a public open access repository. The IHME data are not available due to license restrictions. All public data, Stata and Python code, and supplementary results used and generated by this study are freely and publicly available in the following GitHub repository: https://github.com/DataforDecisionsLLC/The-global-health-and-economic-value-of-COVID-19-vaccination.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
Only three studies have evaluated the global health and/or macroeconomic impact of COVID-19 vaccination (CV), but (i) none translated global mortality or morbidity impacts into summary health measures such as quality-adjusted life years (QALYs), (ii) studies that estimated the global economic impacts of CV did not assess its impacts on the key macroeconomic indicator, gross domestic product (GDP) and (iii) none of the global studies provided a single comprehensive value measure integrating both health and macroeconomic values, or estimated the value-for-money of CV relating its benefits to its costs.
WHAT THIS STUDY ADDS
Our study addresses all three of the above deficits in the literature and in our understanding of the value of CV, finding that CV provided >US$5 trillion in value globally through December 2021, that GDP gains constituted the largest value component (US$2.6 trillion), QALY gains constituted the second largest component (US$2.1 trillion) and direct and indirect costs averted constituted the smallest component (US$0.4 trillion).
CV provided a value of US$741 per dose, which corresponds to a very strong benefit-cost ratio of almost 14.
The GDP benefits alone yielded a benefit-cost ratio of 6.5, indicating that CV’s costs are justified many times over by GDP benefits alone.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
This study demonstrates the significant value and value-for-money of CV, and the substantial economic and health benefits it provides.
It supports investments in, and prioritisation of, global vaccination to end the pandemics; clarifies that addressing pandemics is as much an economic imperative as a health imperative; and reinforces the need to value pandemic vaccines’ economic benefits alongside their health benefits to achieve optimal levels of global investment.
Introduction
The COVID-19 pandemic has been one of the largest pandemics and global economic crises of the past century. COVID-19 vaccination (CV) has been claimed the single most important policy tool allowing global health and macroeconomic recovery.1 2 Yet there is limited research quantifying the broad impact of CV on the global economy or global health.
We found only three studies that quantified the global health and/or economic impact of CV. Watson et al3 estimated the impact of CV solely in terms of averted global COVID-19-related deaths. Bell et al4 and Yang et al5 measured both the health (ie, averted global infections and/or deaths) and global economic impacts of CV. Yang et al5 estimated CV economic impacts solely in terms of averted direct and indirect costs; however, the study did not incorporate the value of unpaid work into its indirect cost estimates. And Bell et al4 estimated CV’s economic impacts on direct costs and global trade, the latter of which was based on hypothetical CV scenarios as opposed to actual CV data. None of the global studies that estimated the health impacts of CV translated these effects into quality-adjusted life years (QALYs) or disability-adjusted life years (DALYs) gains. And those studies that estimated the global economic impacts of CV did not evaluate such impacts on gross domestic product (GDP). The valuation methods used in the existing literature do not allow direct comparison of the relative magnitudes of global economic and health values. Additionally, none of the global studies derived a single comprehensive value measure spanning its health and macroeconomic values, nor an estimate of the value-for-money (VfM) of CV relating its benefits to its costs. We also found no published estimates of the global health or macroeconomic impact of Pfizer-BioNTech vaccines specifically.
Yet, value assessments are essential for decisions regarding coverage of CV within population vaccination programmes by non-commercial vaccine payers such as national governments, multilaterals like the World Bank and non-profits like Gavi who are motivated by public goals such as population health and well-being. And appropriate value assessments should adopt a societal perspective, reflecting both health sector impacts of CV and broader societal impacts, including on the macroeconomy and productivity.6 We make progress in this direction and add to the existing literature by quantifying CV’s broad VfM, reflecting its contribution to global GDP recovery, QALY gains and averted direct and indirect costs.
Methods
Overview
We retrospectively assess the VfM of non-commercial payers’ (NCP) spending on population CV programmes, adopting a societal perspective, and estimating three components of the broad health and economic value (‘broad value’) of CV: (i) health impacts in terms of QALY gains, (ii) averted direct and indirect costs of COVID-19 infections and deaths and (iii) GDP impacts. We estimate these values for each country or territory in our sample (hereafter, ‘country’) and for the globe, and for all vaccine brands combined and for the Pfizer-BioNTech vaccine specifically.
Our focus on the VfM of NCP spending on population CV programmes allows us to ignore CV-related R&D costs, manufacturing costs and vaccine manufacturer profits. In a market economy, it is not the NCP’s responsibility to account for these. Instead, it is the vaccine manufacturer’s burden as a commercial entity to develop a vaccine that is sufficiently valuable to NCPs that negotiated prices cover these costs while allowing for profits. R&D costs, whether private or public, are also irrelevant to NCP coverage decisions because such costs will have already been incurred regardless of whether or not NCPs decide to cover CV. They are therefore not incremental to the coverage decision and drop out of a comparison of vaccination and no vaccination scenarios.
To estimate these broad values, we perform a regression analysis to measure the impact of CV doses on GDP, COVID-19 infections (‘infections’) and COVID-19 deaths (‘deaths’). We combine estimates of vaccine-averted infections and deaths from these regressions with estimates of QALY losses, direct costs per non-fatal infection and death and indirect costs per non-fatal infection to estimate QALY gains and costs averted by vaccination. We value QALY gains at per capita full income. The broad value of CV (VoCV) is the sum of monetised QALY gains, GDP gains and averted direct and indirect costs. The global VoCV sums up country-level broad values, which are themselves the sum of a country’s period-specific broad values.
Box 1 describes the steps of the analysis. The online supplemental appendix provides further details on data sources, variable construction and methods.
Supplemental material
Analysis steps for the valuation
Estimate infection-, death- and PCGDP regressions.
Compute averted infections, deaths and PCGDP losses from vaccination, relative to a no-vaccination counterfactual, using estimates from (1).
Compute averted non-fatal cases from vaccination by taking the difference between averted infections and deaths from (2).
Estimate QALYs per fatal case and per non-fatal case.
Estimate direct costs per case.
Estimate indirect costs (unpaid work) per non-fatal case.
Estimate population-level QALY gains by combining averted deaths from (2), averted non-fatal cases from (3) and QALY losses per fatal and non-fatal cases from (4).
Estimate population-level averted direct costs by combining averted deaths from (2), averted non-fatal cases from (3) and direct costs per case from (5).
Estimate population-level averted indirect costs by combining averted non-fatal cases from (3) and indirect costs per non-fatal case from (6).
Compute full income.
Compute the monetary value of population-level QALY gains by multiplying population-level QALY gains from (7) by full income from (10).
Compute population-level vaccination benefits by summing the population-level monetary value of QALY gains from (11), averted direct costs from (8) and averted indirect costs from (9).
Compute benefit per dose by dividing population-level vaccination benefits from (12) by vaccine doses.
Assume each vaccine dose costs US$53.19.
Compute BCR as ratio of (13) to (14).
BCR, benefit-cost ratio; PCGDP, per capita gross domestic product; QALY, quality-adjusted life year.
Data
Data sources and summary statistics are provided in table 1. We rely on licensed Institute for Health Metrics and Evaluation (IHME) COVID-19 Projections data7 for vaccine doses, infections, deaths, hospitalisations and intensive care unit (ICU) beds. All other data sources are in the public domain.
Data sources and summary statistics
We conduct our analysis in Excel, Stata V.18 and Python V.3.11.
Study countries, horizon and currency
Our study population consists of 148 countries with available data and >1 mllion population, covering nearly 98% of the global population (online supplemental table S1).
The highest frequency at which some data are available is quarterly. Thus, the time-period in most of our analysis is the calendar-quarter (‘quarter’). However, in countries where only annual GDP data are available, the time-period in the GDP regression is annual.
Given data limitations, we limit the study horizon to the pre-Omicron period, spanning 2020Q1–2021Q4. Our base currency, to avoid postpandemic volatility, is 2019 US$. See online supplemental appendix for inflation and exchange rate adjustments.
Regression specification
Infections and deaths
To estimate the impact of vaccinations on infections and deaths, we estimate the following panel regression equations:
(1)
(2)
The dependent variables are per capita infections 
and per capita deaths 
in country i and quarter t.The equations have identical independent variables. The variables 
and 
denote per capita vaccine doses administered of Pfizer-BioNTech vaccines (superscripted by p) and all non-Pfizer-BioNTech brands combined (superscripted by 
), respectively. The variables 
and 
denote per capita cumulative vaccine doses administered since the start of the pandemic up to and including the last day of quarter t. We account for the impact of natural immunity with one lag of new infections, 
, and two lags of cumulative infections, 
. We include two lags of per capita vaccine doses and infections (denoted 
and 
, respectively) because the efficacy of vaccine-induced and infection-induced immunity can last longer than one-quarter, and waning suggests more recent vaccinations and infections have stronger impacts than less recent ones. The second lags of cumulative doses and infections allow the full histories of these variables to have causal impact.
The variable 
is the value of the Government Response Index, a composite index of lockdown stringency, non-pharmaceutical and economic policies on a scale of 0 (no response) to 100 (maximal response).8 We include country fixed effects (
and quarter fixed effects (
) to control for country-specific time-invariant factors and global time-varying factors. The terms 
are error terms. To allow for correlated error terms across equations (1) and (2), and other violations of classical error variance assumptions, we estimate these with seemingly unrelated regression and robust standard errors.
We lag all independent variables to avoid reverse causality from dependent variables.
Gross domestic product
For countries with quarterly GDP data, we estimate a panel regression with an identical specification as (1) and (2) except for the dependent variable:
(3)
The dependent variable 
is the GDP shortfall in country-quarter 
, where the shortfall is the ratio of actual GDP to its prepandemic projected value. The variables 
and 
are the country and quarter fixed effects, respectively, and 
is the error term.
For the 66 countries without quarterly GDP data but with annual GDP data and vaccination starting in either 2020Q4 or 2021Q1, we estimate a cross-sectional regression:
(4)
and 
are the GDP shortfall in 2021 and 2020, respectively; 
and 
are cumulative Pfizer-BioNTech and non-Pfizer-BioNTech doses per capita at the end of 2021Q4, 
is cumulative infections per capita at the end of 2020Q4 and 
is the error term. We omit variables reflecting infections in 2021 from the right-hand side of 
because of potential reverse causality from economic activity during 2021 to infections during 2021 (eg, resulting from workplace exposure).
We allow for a lagged GDP shortfall on the right hand side to allow for dynamic convergence, whereby larger initial shortfalls may imply larger ‘rebounds’. We include lagged cumulative per capita infections to represent natural immunity and pandemic trajectory.
Comments on regression specifications
Regression equations (1)–(3) are parsimonious in that they have relatively few explanatory variables. We chose parsimonious specifications to reduce the risk of overfitting chance correlations, preserve the precision of estimates, preserve sample size (ie, avoid dropping countries without the requisite data) and avoid introducing endogenous and collinear variables. Some endogenous variables such as household and firm behaviours, for example, are likely driven by vaccination (eg, households and firms will likely re-engage in workplace-centred or public-facing economic activity or reduce social distancing when vaccination rates are high), so behavioural changes are part of the mechanism for vaccine impact and should not be controlled for when estimating that impact. While we include lagged infections as explanatory variables, we do not also include lagged deaths: deaths are relatively stable proportions of infections, so lagged infections can proxy for, and will be highly collinear with, lagged deaths. Parsimony also reflects scarcity of relevant variables measured on a quarterly basis and available for many countries. For example, measures of health system capacity (eg, hospital beds per capita) are not available on a quarterly basis during the study time horizon. We note that while our regression specifications are linear in variables, they are non-linear in infections and vaccinations, since we allow two lagged terms for each, thus allowing their timing to matter (we tested more complex lag structures, but these did not improve fit).
A technical issue: in contrast to equations (2) and (3) which are static panel equations, equation (1) is a dynamic panel equation since it has a lagged dependent variable 
as an explanatory variable. Dynamic panels with only a small number of time periods face the well-known problem of Nickell bias: data transformations (either demeaning or first differencing) required to control for the fixed effects induce a correlation between the transformed lagged dependent variable and the transformed error term.9 The standard solution is to use instrumental variables for the transformed lagged dependent variable, where the instruments are even earlier lags of levels and differences of the dependent variable.10 11 However, under certain circumstances, such instrumentation can yield estimates that are inconsistent, inefficient or have poor finite sample properties.12 Having to use earlier lags as instruments requires dropping earlier observations from the regression since these would not have the necessary lagged instruments, impairing precision. Importantly, this standard solution is not available to us because these earlier values already play an explanatory role in (1) through the 
term, which represents natural immunity from earlier infections, rendering them ineligible as instruments. We are aware of no widely accepted solutions to Nickell bias that use instruments distinct from these earlier values. Seeking and justifying such distinct instruments would bring us to the frontier of econometrics research, which is far outside our study scope. For simplicity, we therefore do not attempt to correct for Nickell bias.
Vaccination’s impacts
We separately measure the impact of all CVs relative to a ‘no vaccination’ counterfactual and the impact of ‘Pfizer-BioNTech’ vaccines relative to ‘no Pfizer-BioNTech’ counterfactual. These counterfactuals are simulated worlds where no CV is one scenario and no Pfizer-BioNTech is another scenario. We simulate these counterfactuals using the results of our regression analyses as we describe in online supplemental appendix S3.
Quality-adjusted life years
We compute country-specific QALY losses per fatal and non-fatal infection. QALY losses per fatal infection equal the weighted average of age-specific QALY losses from death,13 14 with weights reflecting the age distribution of COVID-19 deaths.15 QALY losses from death are discounted at 3% following WHO recommendations.16 QALY losses per non-fatal infection are age-invariant and equal the weighted average of QALY losses from infections of different severity levels,17 with weights reflecting the relative probabilities of those levels (‘severity-weighted average’). The severity levels are asymptomatic, mild (not requiring hospitalisation), severe (requiring hospitalisation without ICU admission) and critical (requiring ICU admission). We allow long COVID to affect QALY losses following severe and critical infections.18 We compute QALY gains from vaccination from the product of averted fatal infections from vaccination and the QALY loss per fatal infection, as well as the product of averted non-fatal infections from vaccination and the QALY loss per non-fatal infection.
Direct and indirect costs
We estimate country-specific age-invariant direct costs per infection (which we apply equally to fatal and non-fatal infections) as a severity-weighted average of the product of inpatient or outpatient unit costs19 and severity-specific durations of utilisation (eg, hospital or ICU length of stay, or one outpatient visit per mild infection).20
Country-specific age-invariant indirect costs per non-fatal infection consist of lost unpaid work related to infection, which is a severity-weighted average of the product of severity-specific workdays lost (our proxy for the days of unpaid work lost), daily hours spent on unpaid work and the hourly wage. We exclude paid work from indirect costs to eliminate double counting given our consideration of GDP impacts. We exclude all indirect costs of fatal cases to eliminate double counting given our monetisation of QALYs.
Full income
We monetise QALYs at country-specific age-invariant full income, equal to the sum of per capita annual income21 and the per capita value of annual non-market time. Annual non-market time consists of unpaid work and leisure time,22 23 where each hour is valued at the hourly wage.24 As discussed below and shown in the online supplemental appendix, full income is a conservative approximation to individuals’ willingness-to-pay (WTP) per QALY.
The broad value formula
We take the broad VoCV to be the sum of monetised QALY gains, GDP gains and averted direct and indirect costs. We theoretically derive and justify this broad value measure more fully in online supplemental appendix S1. Put simply, when utility is a multiplicative function of health and full income, then the impact of a shock on utility can be approximated by the sum of two terms. The first term is the impact on utility of the shock-induced impact on health holding full income fixed, and the second is the impact on utility of the shock-induced impact on full income holding health fixed. We convert these utility impacts into WTP to avoid such impacts by dividing them by the expected marginal utility of full income. Under certain functional form and simplifying assumptions, we find that the first term can be conservatively approximated by the product of full income (serving as a conservative estimate of WTP per QALY) and the QALY impacts of the policy, while the second term equals the policy-induced impact on full income. (See Murphy and Topel25 (their equation 13) for an example of full income as we define it being a component of, and a conservative approximation to, WTP for health.) Since the COVID-19 pandemic is a shock and CV is a reduction in the magnitude of the shock, the broad VoCV can be approximated by the resulting form.
Since our broad value measure reflects WTP for the health and economic impacts of policy, our analysis is a form of cost-benefit analysis (CBA). Since we derive our broad value measure from a utility framework in which QALYs represent health-related utilities, our analysis is also a form of cost-utility analysis (CUA). However, given that we value QALYs at individuals’ WTP per QALY, ours is a societal perspective CUA, as opposed to a health payer perspective CUA where QALYs are valued using the shadow price of some exogenously given health payer budget. Our broad value measure can also be understood within a societal-perspective cost of illness (CoI) framework in which costs of illness are defined broadly to encompass health and productivity losses, health losses are valued in terms of individuals’ WTP for health, health is measured by QALYs and the broad value of vaccination equals its averted CoI.
Marginal versus non-marginal risks
The approximations we use to derive the broad value formula are valid for marginal risks, but the risks imposed by COVID-19 are potentially non-marginal. Standard models of WTP for mortality risk reductions suggest that the rate of substitution of wealth for risk reduction declines when the magnitude of the risk reduction grows,26 raising the prospect that formulas like ours derived for marginal risks will overestimate WTP for non-marginal risks. Despite this concern, we persist with our approximations for three reasons. First, a published calibration exercise suggests such overestimation is modest in the range of risk reductions we consider. Table 1 shows median per capita deaths per calendar quarter of 14/100 000, implying an annualised mortality risk shock of 56/100 000=0.56/1000, which is within the 1/1000 to 1/10 000 range in which overestimation appears modest.26 Second, the same standard models ignore several salient features of COVID-19—dread, uncertainty, ambiguity and catastrophe—that may double the WTP.26 This would considerably offset the modest overestimation, suggesting that our approach is likely on balance conservative. Third, WTP for non-marginal risks is significantly understudied, and there are no broadly accepted or well-established empirical estimates of such WTP.26
Scenario and sensitivity analyses
We perform one-way scenario analyses, using lower and upper bound estimates of infections in place of mean estimates, using confirmed deaths in place of under-reporting adjusted deaths and using 0% and 6% discount rates.
We perform a probabilistic sensitivity analysis (PSA), taking 1000 draws from a multivariate normal distribution constructed using the model coefficients and variance-covariance matrix. For each draw, we calculate the VoCV. We estimate a 95% CI from these 1000 VoCVs.
Results
As shown in table 1, on average, fatal infections incur vastly larger QALY losses (2.74) than non-fatal infections (0.0074), and indirect costs per non-fatal infection (US$489.33) are an order of magnitude larger than the direct cost per fatal or non-fatal infection (US$46.13).
We present the results of the base-case regression models in table 2.
Base-case regression results
Vaccination coefficients generally have the correct sign (negative in the infections and deaths regressions, and positive in the GDP regressions) and are statistically significant. We provide our interpretation of the coefficients as infection and mortality risk reductions and percentage-point GDP gains in online supplemental appendix S3.
We summarise our central findings in tables 3 and 4. Table 3 presents pandemic outcomes (columns A, B), outcomes in the no vaccination counterfactuals (columns C, D) and attributes the differences to CV (columns E, F). Table 4 converts the aggregated values in table 3 (colums E, F) to per capita and per dose values.
Global VoCV by value component and focal brand (2019 US$)
Global value of COVID-19 vaccination (2019 US$) per capita and per dose by value component
Base-case population-level results
The similarity across (A) and (B) in table 3 indicates that our regressions provide reasonable predictions of infections, deaths and GDP. Table 3 also shows that the pandemic caused US$5.6 trillion in GDP losses, that the absence of all CV and Pfizer-BioNTech vaccines would have raised this loss to US$8.2 and US$6.5 trillion, respectively. Therefore, we estimate that CV as a whole and Pfizer-BioNTech vaccines specifically produced GDP benefits of US$2.6 trillion and US$877.8 billion, respectively.
Our results suggest that the pandemic caused 5.2 billion infections (symptomatic and asymptomatic), that the absence of all CV and Pfizer-BioNTech vaccines would have raised this to 6.9 and 5.5 billion infections, respectively and that therefore CV as a whole and Pfizer-BioNTech vaccines in particular, averted 1.7 billion and 270.7 million infections, respectively. The pandemic caused 15.4 million deaths, the absence of all CV and Pfizer-BioNTech vaccines would have raised this to 19.4 and 16.5 million deaths, respectively, and therefore CV as a whole and Pfizer-BioNTech vaccines in particular averted 4.1 and 1.1 million deaths, respectively.
Attributing differences in predicted and counterfactual outcomes to vaccines in this manner, we find that CV (Pfizer-BioNTech alone) averted 54.7 (11.8) million QALYs whose monetary value is US$2.1 trillion (US$873.3 billion), US$62.6 (US$27.8) billion in direct costs and US$340.9 (US$128.9) billion in indirect costs.
Combining GDP losses, monetised QALY losses and direct and indirect costs, the pandemic caused a total loss of US$10.2 trillion, the absence of all CV and Pfizer-BioNTech vaccines would have raised this to US$15.4 and US$12.2 trillion, respectively, and therefore CV as a whole and Pfizer-BioNTech vaccine specifically produced US$5.2 and US$1.9 trillion in value, respectively.
Considering the postvaccination period (2021Q1–2021Q4), we estimate the pandemic burden to be US$4.98 trillion, which would have been US$10.13 trillion without all CV or US$6.89 trillion without Pfizer-BioNTech specifically. Thus, once the vaccine rollout began, all CV reduced the burden of the pandemic by 50.9% ((US$10.13 trillion−US$4.98 trillion)/US$10.13 trillion) and Pfizer-BioNTech specifically reduced it by 27.7% ((US$6.89 trillion–US$4.98 trillion)/US$6.89 trillion).
Recall that we value QALYs conservatively using full income, implying that our estimated share of health in total burdens and benefits is likewise conservative. Subject to that caveat, we find that GDP losses constitute at least half (54%) the total pandemic burden (table 3), suggesting pandemics are as much a threat to the global economy as they are to global health, and that addressing pandemics is as much an economic imperative as a health imperative. Furthermore, the GDP-related value of vaccination is at least as large as the value of its impact on global population health as measured by monetised QALY gains (51% vs 41% for all CV and 46% vs 46% for Pfizer-BioNTech vaccination). We find that the broader economic value of vaccination (reflecting its impact on global GDP and indirect costs) exceeds its health system-related values (reflecting its impact on global population health and direct costs): economic values comprise 51.2%+6.6%=57.8% of the VoCV while health-related values comprise 40.9%+1.2%=42.1% of its value; economic values comprise 46.0%+6.8%=52.8% of the value of Pfizer-BioNTech vaccination while health-relative values comprise 45.8%+1.5%=47.3%.
Per capita and per dose results
On average, each person in our sample receives US$710 in value from CV and US$269 in value from Pfizer-BioNTech. On average, each dose of CV and Pfizer-BioNTech produces US$741 and US$1640 in value, respectively. However, these values can vary widely by country. For example, the 25th percentile VoCV per dose for all vaccines (Pfizer-BioNTech) is US$93.00 (US$252.16) and the 75th percentile is US$965.92 (US$2482.13) with a few major outliers (figure 1).
Country-specific broad value per dose for all COVID-19 vaccination (top) and Pfizer-BioNTech specifically.
Ratios of our per capita and per dose values yield a pooled estimate of doses per capita: US$710.06/US$741.16=0.96 and US$269.23/US$1640.28=0.16 doses per capita for CV and Pfizer-BioNTech, respectively. Fewer doses per capita explains why Pfizer-BioNTech has greater value per dose but lesser value per capita.
Benefit-cost ratio
In the USA, the average cost per original monovalent booster dose was US$20.69, and administration costs were between US$25 and US$40 per dose,27 yielding a per dose cost estimate inclusive of administration costs of US$20.69+(US$25+US$40)/2=US$53.19. We treat this as a simple upper-bound estimate of the cost per dose. Given the per dose benefits, the benefit-cost ratio for CV in general is US$741.16/US$53.19=13.93 and for Pfizer-BioNTech vaccination is US$1640.28/US$53.19=30.84. Thus, vaccination benefits are an order of magnitude larger than costs. Our calculations suggest that macroeconomic benefits alone rationalise CV’s costs multiple times over4: macroeconomic benefits per dose are US$347.45/US$53.19=6.5 and US$611.94/US$53.19=11.5 times per dose costs for CV and Pfizer-BioNTech, respectively.
Sensitivity analyses
Our PSA suggests that 95% confidence intervals are given by VoCVs 21% above and below our main estimates (table 5). Using IHME’s lower bound of infections reduces the VoCV by 20.6% for all vaccines and 20.1% for Pfizer-BioNTech. Using the upper bound increases the VoCV by 21.8% and 22.3%, respectively. Using IHME’s unscaled deaths variable reduces the VoCV by 18.2% and 20.7%, respectively. Reducing the annual discount rate to 0% increases the VoCV by 13.3% for all vaccines and 14.4% for Pfizer-BioNTech. Increasing the annual discount rate to 6% decreases it 7.2% and 7.9%, respectively. In summary, varying values for key parameters result in estimates that are about 10%–20% above or below the base-case estimates.
Total Value of vaccination sensitivity analysis results (2019 US$)
Vaccine effectiveness
We estimate vaccine effectiveness against infections and deaths to be 36.5% and 30.6%, respectively, for CV, and 49.3% and 63.2%, respectively, for Pfizer-BioNTech. Thus, Pfizer-BioNTech effectiveness against infections and deaths is at least 50% and 100% higher, respectively, than that of CVs on average (see online supplemental appendix S5.1 for details).
Discussion
This study uses a regression-based approach to estimate the global health and macroeconomic impact of CV over the first 2 years of the pandemic. Combining GDP losses, monetised QALY losses and direct and indirect costs, we find that the pandemic caused an economic loss of US$10.2 trillion. The absence of all CV and Pfizer-BioNTech vaccines would have raised this loss to US$15.4 and US$12.2 trillion, respectively. Therefore, CV as a whole, and Pfizer-BioNTech vaccines in particular, generated values of US$5.2 and US$1.9 trillion, respectively.
Table 6 compares our results with those of other studies.
Comparisons of our results with those from other selected studies
Our estimates of averted infections are higher than those found in the literature (Di Fusco et al20; Yang et al5); our averted deaths from vaccination are similar to those of other studies (Di Fusco et al20; Steele et al28; Yang et al5) and fall considerably below the least conservative estimates (Watson et al3); our estimates of the GDP loss from the pandemic are similar to those of The Economist29 and Cutler and Summers,30 but smaller than less conservative estimates of Walmsley et al31; our monetised QALY losses during the pandemic is smaller than the most comparable estimates we found for the globe32 and for the USA30; our VoCV is similar to that of Kirson et al33 for the USA, but much higher than that of Sandmann et al34 for the UK.
Substantial differences between our study and other studies are likely due to methods (regression vs models, see section S2.1) and data input differences. For example, IHME data on infections (symptomatic and asymptomatic)35 used in this study far exceed Our World in Data estimates of confirmed cases.36 While estimates in table 6 vary widely across studies, the literature reveals that the COVID-19 pandemic has imposed profound health and economic burdens. Our study contributes to this literature by estimating the broad loss of the pandemic and the broad benefit of CV.
The following are limitations of our study. Like all regression analyses, our predicted values may underestimate or overestimate observed values for individual country-quarter observations even if such deviations wash out in aggregate, thereby making country-specific projections less reliable than global projections. We also have data limitations: some countries only have annual as opposed to quarterly GDP data; infections, deaths and other important quantities are not age-disaggregated; COVID-19 health utility impacts are inferred from proxy conditions; some variables are missing for some countries, which require extrapolation from other countries. We do not measure potentially important value elements, such as fiscal effects, mental health, education, community transmission reduction and caregiver burdens, which makes our VoCV estimates conservative. We do not address the relationship between COVID-19 and comorbidities. We assume long COVID only affects QALY losses in severe and critical infections, while there is evidence that it could also affect patients with mild infections.18 We ignore the equity issues raised by cross-country differences in WTP resulting from global inequality in per capita gross domestic product. We are unable to address concerns regarding Nickell bias. And we use formulas derived for marginal risks and do not more fully address the impact on WTP of non-marginal risks, dread, uncertainty, anxiety and catastrophe.
There are many possible extensions of our study. First, a more rigorous treatment of VfM calculations could include country-specific and manufacturer-specific prices. Second, we adopt a CBA approach where every dollar’s worth of benefit has equal value. This implies that when we value QALYs using full income, a rich country’s QALY counts more than a poor country’s QALY (because of the former’s higher full income), and that US$1 of GDP gain counts equally in rich and poor countries, even if such GDP gain may produce larger well-being gains in poor countries given diminishing marginal utility of income. Therefore, future extensions could use more equity-sensitive value frameworks like social welfare functions. Other extensions include the VoCV in controlling variant emergence, the global health and economic costs of CV-related vaccine hesitancy and optimal spending on pandemic preparedness.
We find CV impacts to be empirically large, and though we do not fully investigate CV costs, simple approximations suggest significant VfM. Importantly, we find that the broader economic costs of the pandemic exceed its narrower health-related costs, that the broader economic benefits of CV exceed its narrower health-related benefits and that the macroeconomic benefits alone justify the costs of CV many times over.
These results have implications for perhaps the most important policy debate in vaccine evaluation: whether to value vaccines narrowly (ie, focusing only on health-centric outcomes) or broadly (ie, incorporating broader socio-economic and other outcomes). Our results suggest narrow approaches ignore perhaps half the harm of pandemics, and perhaps half the benefits of pandemic vaccines. This suggests that achieving optimal societal and global investment in CV requires valuing it broadly.
Conclusions
We find that the COVID-19 pandemic had a profound impact on global health as measured by QALY losses, and on the global economy as measured by GDP losses. Our study finds that GDP losses constitute at least half the total health and economic burden of the COVID-19 pandemic, suggesting that addressing pandemics is as much an economic imperative as a health imperative. In per capita terms, the broad VoCV confers considerable value to each person in the world on average. Our benefit-cost ratios suggest that vaccination benefits are an order of magnitude larger than vaccination costs, indicating that CV offers significant value for money. The GDP-related VoCV is at least as large as its QALY-related value, suggesting that the broader economic VoCV exceeds its narrow health-related value, and macroeconomic benefits alone rationalise CV’s costs multiple times over. Perhaps the most important policy debate in vaccine evaluation is whether to value vaccines narrowly (focusing only on health impacts) or broadly (including broader socio-economic impacts). Our findings suggest that narrow approaches neglect at least half the harm of pandemics and at least half the benefits of pandemic vaccines, and that broad valuation is required to achieve optimal societal investment in CV. Our results highlight the profound health and economic impacts of the COVID-19 pandemic and of CV, and the importance of valuing CV from a societal perspective that takes into account economic impacts.
Data availability statement
Data not subject to license restrictions are available in a public open access repository. The IHME data are not available due to license restrictions. All public data, Stata and Python code, and supplementary results used and generated by this study are freely and publicly available in the following GitHub repository: https://github.com/DataforDecisionsLLC/The-global-health-and-economic-value-of-COVID-19-vaccination.
Ethics statements
Patient consent for publication
Ethics approval
Not applicable.
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Handling editor Lei Si
Contributors JPS, MDF, JY and DEB conceptualised the study. MDF and DEB acquired the study’s funding. JPS, DB, JSK and DEB decided on the analytical methodology. MDF provided resources for the study, and JPS, DB, JSK, MDF, MHK and JY contributed to the investigative process. DB and JSK collected, managed and analysed data using their own code. JPS, DB and JSK validated this analysis. JPS, DB, JSK, MDF and DEB participated in project administration roles, and JPS, MDF, MHK, JLN, JY and DEB supervised the study. JPS, DB and JSK wrote the original manuscript, with DB and JSK providing visualisations. All authors reviewed, edited and approved the final draft. All authors have approved the decision to submit for publication and meet the ICMJE criteria for authorship. JPS is the guarantor.
Funding This study was funded by Pfizer Inc. and editorial support was provided by Data for Decisions.
Map disclaimer The inclusion of any map (including the depiction of any boundaries therein), or of any geographic or locational reference, does not imply the expression of any opinion whatsoever on the part of BMJ concerning the legal status of any country, territory, jurisdiction or area or of its authorities. Any such expression remains solely that of the relevant source and is not endorsed by BMJ. Maps are provided without any warranty of any kind, either express or implied.
Competing interests JPS, DB and JSK are employees of Data for Decisions (DfD) and worked on this study in that capacity. JPS and DB have worked on other studies funded by grants from Pfizer Inc. to DfD. DEB is an external consultant to DfD and in that capacity has worked on this and other studies funded by grants from Pfizer Inc. to DfD. JPS and DEB in their personal capacities have received compensation from Pfizer Inc. for providing consulting services and for speaking and participating in meetings and advisory boards. MDF, MK, JLN and JY are employees of Pfizer Inc. and each held Pfizer stock or stock options at the time of the study. Pfizer Inc. employs MDF, MK, JLN and JY, but otherwise played no role in study design, data collection and analysis, decision to publish or preparation of the article.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.