key: cord-1017968-6ydbsy7c
authors: Siedner, M. J.; Alba, C.; Fitzmaurice, K. P.; Gilbert, R. F.; Scott, J. A.; Shebl, F. M.; Ciaranello, A.; Reddy, K. P.; Freedberg, K. A.
title: Cost-effectiveness of COVID-19 vaccination in low- and middle-income countries
date: 2021-05-02
journal: nan
DOI: 10.1101/2021.04.28.21256237
sha: 647196d1d74bdc7359933704ef4cf9c3770333bd
doc_id: 1017968
cord_uid: 6ydbsy7c

Despite the advent of safe and highly effective COVID-19 vaccines, pervasive inequities in global distribution persist. In response, multinational partners have proposed programs to allocate vaccines to low- and middle-income countries (LMICs). Yet, there remains a substantial funding gap for such programs. Further, the optimal vaccine supply is unknown and the cost-effectiveness of investments into global vaccination programs has not been described. We used a validated COVID-19 simulation model8 to project the health benefits and costs of reaching 20%-70% vaccine coverage in 91 LMICs. We show that funding 20% vaccine coverage over one year among 91 LMICs would prevent 294 million infections and 2 million deaths, with 26 million years of life saved at a cost of US$6.4 billion, for an incremental cost effectiveness ratio (ICER) of US$250/year of life saved (YLS). Increasing vaccine coverage up to 50% would prevent millions more infections and save hundreds of thousands of additional lives, with ICERs below US$8,000/YLS. Results were robust to variations in vaccine efficacy and hesitancy, but were more sensitive to assumptions about epidemic pace and vaccination costs. These results support efforts to fund vaccination programs in LMICs and complement arguments about health equity, economic benefits, and pandemic control11.

By late April 2021, over 147 million cases and over 3 million deaths were attributed to the COVID-19 pandemic globally 12 . Approximately 53% of COVID-19 deaths have been reported in low-and middle-income countries (LMICs) 13 , although such estimates may underestimate the epidemic's health impacts 14,15 and do not account for disruptions in economic productivity, healthcare delivery, and social wellbeing [16] [17] [18] [19] [20] . Nonetheless, there has been major progress toward containing the pandemic with the advent and licensure of multiple highly effective vaccines for prevention of COVID-19 [1] [2] [3] [4] .

Although safe and effective vaccines provide a strategic path out of the pandemic, their benefits have thus far largely been confined to high and upper-middle income countries, which have secured contracts for purchase and distribution of nearly 90% of the global vaccine supply 5 . LMICs have had less success in procuring vaccines 5 , but global initiatives to ameliorate this gap are underway. The COVAX Advance Market Commitment (AMC) program aims to ensure access to low-cost SARS-CoV-2 vaccines to 92 LMICs 7 . By February 2021, higher-income governments and donors had committed over US$6.3 billion to this initiative 21 . However, there remains a funding shortfall for the COVAX AMC initiative to achieve initial goals of vaccine delivery for 20% of the population in LMICs 7 . Moreover, the 20% vaccine target set out by COVAX was an initial target is below the projected levels needed to achieve epidemic control 22 .

Whereas the US government and other high-income countries have made longstanding commitments to development assistance for health in LMICs, with an estimated US$39 billion contributed worldwide in 2018 alone 23 , the costs, benefits, and cost-effectiveness of donations into the global COVID-19 vaccine supply have not been estimated.

Our objective was to assess the clinical benefits (COVID-19 infections, hospitalizations, deaths, life years saved), program costs, and value (incremental cost-effectiveness ratios [ICERs]) of donor outlays into the global vaccine supply for 91 LMICs (we excluded India, a COVAXeligible country, because of its plan to produce vaccines domestically 24 ).

We used the Clinical and Economic Analysis of COVID-19 interventions (CEACOV) model, a validated, dynamic microsimulation of the natural history of COVID-19 8,25-27 (Supplementary Methods and Supplementary Fig. 1 ). We modeled discrete epidemics over 360 days in each country using country-specific data on age distribution, population size, and hospital and ICU bed capacities (Supplementary Table 1) . We compared a situation in which participating LMICs have no vaccine access to one in which vaccination supplies reach 20%-70% population coverage, with vaccines prioritized to individuals aged 60 years or older, and increasing in 10% increments. In our base case, we assumed a modest epidemic growth rate (effective reproductive number [R e ] of 1.2), and used data from the Johnson & Johnson (J&J)/Janssen Ad26.COV2.S vaccine trial to inform vaccine efficacy 1 (Extended Data Table 1 ). Costs were from the donor perspective, and included fixed costs for the vaccination program (planning, training, social mobilization, cold chain equipment, and pharmacovigilance 28 ) and variable costs per person vaccinated (vaccines, logistics and delivery and technical assistance) 28,29 (Extended Data Table   1 ). We conducted sensitivity analyses among a subset of nine countries, chosen based on a cluster analysis, to assess robustness of our estimates to assumptions about epidemic severity, as well as vaccination hesistancy, efficacy, rollout pace and costs.

In the base case analysis, achieving 20% vaccine coverage (508 million people vaccinated) in 91 LMICs would decrease infections by over 50% in the following year, from 547 million to 253 million, and decrease projected COVID-19 deaths nearly 80%, from 2.4 million to 512,000, saving 26 million years of life compared to no vaccine coverage (Table 1 ; individual country and regional estimates are in Supplementary Table 3 ). Total vaccination program costs to achieve 20% coverage would be US$6.4 billion, resulting in an ICER of US$20/infection prevented and US$250/year of life saved (YLS) compared with no vaccination (Table 1 and Fig. 1 ). Compared to 20% population vaccine coverage, increasing coverage to 30% would prevent an additional 74 million infections and 208,000 deaths, with an ICER of US$40/infection prevented and US$870/YLS. Increasing coverage from 30% to 40% and from 40% to 50% would prevent an additional 70,000 and 31,000 deaths, and result in ICERs of US$80/infection prevented and US$2,820/YLS and US$150/infection prevented and US$7,240/YLS, respectively. Beyond 50% coverage, reductions in infections and deaths continue, although with diminishing efficiency; the ICER for expanding coverage from 60% to 70% is US$760/infection prevented and US$41,900/YLS.

In one-way sensitivity analyses, the cost-effectiveness of providing a global vaccination supply to 20% of the population was most affected by the prevalence of prior protective immunity, the infection fatality ratio (IFR) of COVID-19, the epidemic effective reproductive number (R e ), and, to a lesser extent, program costs (Fig. 2) . Aside from scenarios in which the population . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint prevalence of prior protective immunity against SARS-CoV-2 infection was 25% or greater, the ICER would remain below US$60/infection prevented and US$1,700/YLS for achieving 20% coverage across a wide range of assumptions ( Fig. 2 and Extended Data Tables 2-10). Vaccine efficacy against infection, efficacy against symptomatic disease, and efficacy against severe/critical disease; as well as the pace of vaccination rollout, vaccine uptake, and vaccine cost, had negligible effects on the cost-effectiveness of a 20% vaccine supply to LMICs (Fig. 2) .

For example, the ICER for a vaccine supply of 20% would be US$500/YLS even with a doubling of program costs ( Table 2 ).

The projected ranges of ICERs as vaccine supply increased from 20% to 70% while R e , IFR, and program costs were independently varied are presented in Table 2 . For lower levels of R e and IFR, ICERs would remain below US$5,000/YLS up to 30% coverage ( Table 2 ). Our findings were also robust to assumptions about vaccination costs: if program costs were doubled, ICERs would remain below US$2,500/YLS up to 30% coverage, and below US$8,000/YLS up to 40% coverage ( Table 2 ). The full set of cost and clinical outcomes for all sensitivity analyses conducted (including analyses in which we varied prior protective immunity, IFR, R e , program costs, vaccine uptake, pace of vaccination rollout, and vaccine efficacy) is shown in Extended Data Tables 2-10.

In two-way sensitivity analyses, we simultaneously varied IFR and program costs and found that the ICER to achieve 20% vaccine coverage would remain at or below US$3,350/YLS, including in a scenario in which the overall IFR was 83% lower than in the base case and program costs doubled (Fig. 3A) . In the lowest IFR scenario tested, the ICER for extending the supply from 20% to 30% would be US$7,210/YLS when program costs are doubled, but would otherwise remain at or below US$3,610/YLS at all tested program costs (Fig. 3B ). ICERs would range from US$1,790/YLS to US$43,460/YLS when considering expanding vaccine supplies from 30% to 40% (Fig. 3C) . We also simultaneously varied R e and program costs and found that the ICER remained below US$9,140/YLS to achieve up to 30% vaccine supply even with a lower R e (1.1) and doubling of program costs (Supplementary Figure 2 ).

Investments into COVID-19 vaccine supply and distribution to LMICs sufficient to achieve 20% population coverage would prevent 294 million infections and 2 million deaths over a one-year period and be highly cost-effective with an ICER of approximately US$20/infection prevented and US$250/YLS compared to no vaccine coverage. Financing an expanded vaccine supply for up to 50% of these populations would prevent millions of infections and save hundreds of thousands of additional lives, with ICERs below US$200/infection prevented and US$8,000/YLS. These results demonstrate the substantial health benefits and value of global efforts to promptly support LMICs with the infrastructure and supply needed to vaccinate large proportions of their populations, and complement arguments focused on health equity 9 , economic benefits 10 , and pandemic control efforts 11 .

While there is no universally accepted ICER threshold to determine value among donor countries investing on behalf of lower-income countries, the ICERs we estimate for funding of vaccination programs in 91 LMICs are $250/YLS at 20% coverage and below $8,000/YLS at 50% coverage.

These ICERs are substantially lower than or comparable to similar and important donor-financed . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Our findings were most sensitive to the prevalence of protective immunity at the time of vaccine rollout. As expected, higher prevalence of prior protective immunity would reduce the vaccination program's value, due to expenditures on individuals who do not derive benefit in the model. Most LMICs that have conducted population-based estimates have found rates of prior exposure of 4-10% 15,33,34 . Further, our model assumes that prior immunity offers complete protection from re-infection and is durable for the 360-day time horizon, which likely overestimates the protective effects of prior infection and may underestimate the value of vaccination 35-37 .

We also found that effective reproductive numbers lower than our base case assumption (R e =1.2) would reduce total infections, hospitalizations, and deaths, thereby reducing the value of a vaccination program. In the absence of effective vaccine rollout, however, both the COVID-19 pandemic itself and non-pharmaceutical interventions to limit COVID-19 transmission are expected to continue to have deleterious impacts on both health and the economy in each of these countries, costs that are not reflected in our model 10 . Moreover, the characteristics of COVID-19 surges have been highly variable and unpredictable, and it is likely that, in the absence of vaccines, countries may continue to experience epidemic waves with higher R e , particularly if and when non-pharmaceutical interventions are relaxed 38 .

Finally, our results were sensitive to assumptions about COVID-19 disease severity. Our base case IFR for the 91 LMICs in the absence of vaccines (0.45%) is derived from published data on the natural history of COVID-19 and calibrated to reflect IFRs in published meta-analyses that include paired population seroprevalence and death reporting data 39 . These estimates are supported by data from South Africa, which includes monitoring of excess natural deaths 40 , and by an autopsy study in Zambia, which found that approximately 16% of deceased individuals were unrecognized as having COVID-19 at death but tested positive for COVID-19 postmortem 14 . However, IFR estimates in some LMICs based on official death reporting, such as Kenya, have been 20 times lower than our base case value 41 and there have been sizeable variations in estimated IFR between global regions (e.g. 0.37% in western sub-Sahran Africa versus 1.45% in eastern Europe) 42 . To account for this, we varied the probability of developing severe or critical infection to as low as 25% of the estimates in our base case, resulting in an IFR of 0.07%, or 83% lower than in the base case, and nearly 70% lower than pooled estimates in the sub-Saharan African region 43 . Even in this low IFR scenario, funding 20% vaccination would have an ICER of US$20/infection prevented and US$1,680/YLS. Consequently, although estimates of COVID-19-specific mortality rates remain a matter of debate in LMICs 44,45 , the value of vaccination programs would remain high in most scenarios.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint This analysis has several limitations. First, natural history inputs were originally derived and validated as part of an analysis based in South Africa 8 . We used inputs calibrated to data from South Africa for three reasons: 1) accurate IFR data from other countries, particularly many LMICs, is limited; 2) age is well-established as the greatest risk factor for COVID-19 mortality and, after accounting for age, additional co-morbidities appear to have little additional effect on expected and reported mortality in LMICs 46,47 ; and 3) use of data from South Africa is likely to more closely reflect SARS-CoV-2 natural history estimates in LMICs than data from highincome countries. Second, our model assumes homogeneous mixing, such that all individuals within a country are equally likely to become infected and transmit to others, and we do not include transmission between countries. The homogenous mixing assumption may underestimate transmissions in high contact and densely populated settings while overestimating transmissions in low contact and rural settings. Not including transmission between countries might also underestimate the value of increased global vaccine distribution. Third, our model includes data on vaccine efficacy, hesitancy, and costs, which are all from published studies but subject to uncertainty. Despite this, our findings were robust to plausible ranges in these parameters.

Fourth, we did not account for the potential secondary health benefits of COVID-19 vaccination.

The pandemic has been predicted to indirectly increase morbidity and mortality in LMICs through the overwhelming of health systems, worsening of food insecurity, disruption of supply chains, infections of health care workers, and repurposing of healthcare sector budgets 20,48,49 . We also did not account for potential longer term secondary benefits of vaccination programs, such as prevention of the emergence of viral variants and strengthening public health infrastructure in LMICs. Since this analysis is from the donor perspective, we did not account for averted . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 2, 2021. domestic healthcare costs due to fewer COVID-19 hospitalizations that would be borne by recipient countries in the absence of vaccination. Finally, we do not model the potential economic losses from failure to accomplish global vaccination, losses estimated up to US$9 trillion, as much as half of which are expected to be borne by high-income countries 10 .

In summary, we found that donor investments in the COVAX initiative to fully subsidize 20% vaccine coverage in LMICs, would prevent nearly 300 million infections and 2 million deaths over one year, and would be cost-effective compared to current widely-supported public health and medical interventions in the US 30,31 . Attaining coverage levels up to 50% would provide major additional benefits and remain cost-effective at thresholds below those of other donor aid programs for health. These findings, in conjunction with ethical, social, and economic benefits of global vaccine equity, support urgent global efforts to promote vaccine distribution and implementation of vaccination programs in LMICs.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint Abbreviations: AMC, Advance Market Commitment; R e , effective reproductive number.

One-way sensitivity analyses were conducted in a subset of 9 representative countries:

Afghanistan, Cambodia, Lesotho, Moldova, Mongolia, Morocco, Nicaragua, Sri Lanka, and

Zambia. These countries were chosen to reflect the variation in global region, age structure, hospital bed capacity, and ICU bed capacity that exists among the 91 COVAX Advance Market Commitment (AMC) countries (Methods). Incremental cost-effectiveness ratios are presented as US$/infection prevented and US$/year of life saved (YLS) and rounded to the nearest ten.

Dominated strategies are ones that provide fewer health benefits than a less costly strategy (strong dominance) or have a higher ICER than that of a strategy providing greater health benefits (extended dominance). In incremental scenarios resulting in relatively few additional infections or deaths, strategies with increased vaccine supply may appear to be dominated by . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint those with lower vaccine supply due to stochastic variation. The population-wide infection fatality ratio for the 9 included countries (0.44%) differed from that of all 91 countries (0.45%).

Reductions in the infection fatality ratio were modeled by decreasing the risk of developing severe/critical disease to 50% and 25% of the base case risk, resulting in an IFR of 0.18% and 0.07%, respectively. Total cost of the vaccination program included both fixed and variable costs.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint a vaccine for the 9 included countries. Decreasing the risk of developing severe/critical disease to 50% and 25% of the base case risk resulted in an IFR of 0.18% and 0.07%, respectively. For each scenario, the IFR in the absence of a vaccine is displayed on the horizontal axis, the incremental cost-effectiveness ratio (ICER) compared to the next-highest supply level is displayed on the vertical axis, and the program cost relative to the base case is denoted by color.

The effect of concurrently varying IFR and program cost on the ICER compared to the nexthighest supply level is displayed for 20% supply (A), 30% supply (B), and 40% supply (C). The scales of the vertical axes differ in the three plots.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint

The Clinical and Economic Analysis of COVID-19 interventions (CEACOV) model is a validated, dynamic microsimulation of the natural history of COVID-19 and the impact of public health interventions 8, [25] [26] [27] . We used the CEACOV model to project the clinical impact, cost, and cost-effectiveness of donor outlays to purchase, distribute, and deploy SARS-CoV-2 vaccines to 91 low-and middle-income countries (LMICs). We excluded India from our base case analysis, despite India being named in the original COVAX plan, because of its plan to produce vaccines domestically 24 . We modeled discrete epidemics in each country using country-specific age distribution, population size, and hospital and ICU bed capacity (Supplementary Table 1 ); we present both country-level and aggregate outcomes.

The CEACOV model is based on an SEIR framework, and includes susceptible, exposed, infectious, recovered, and dead states 50 (Supplementary Methods and Supplementary Fig. 1 ).

Susceptible individuals face a daily probability of infection with SARS-CoV-2, while infected individuals face daily probabilities of disease progression through six COVID-19 states: preinfectious latency, asymptomatic, mild/moderate, severe, critical, and recuperation (Supplementary Table 2 ). With mild/moderate disease, individuals have symptoms, such as cough or fever, but do not require inpatient management. With severe disease, symptoms warrant inpatient management, and with critical disease, patients require ICU care to survive. Recovered individuals are assumed immune from repeat infection for the duration of the modeled time . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. Table 2 ).

Individuals with SARS-CoV-2 infection transmit to susceptible individuals at health statestratified rates (Extended Data Table 1 

We quantified the clinical benefit of a vaccination program in terms of years of life saved from prevented COVID-19 deaths. Life-years occurring in future periods were discounted at a rate of 3%/year-as commonly done in cost-effectiveness analyses of healthcare interventions 52 . For each country included in our analysis, we calculated the average number of years a person would have lived subsequent to his or her current age had he or she not died of COVID-19 using sexstratified life tables published by the United Nations World Population Prospects for the period . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. Because we were unable to identify age-and sex-stratified data on the reported number of infections in most of the countries included in our analysis, we assumed that the age-sex distribution of infections would mirror the overall population structure of a given country (i.e., homogeneous mixing).

Vaccine supply . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint

We modeled the impact of a COVAX program providing vaccines at various coverage levelsfrom 20% to 70% of each country's population. The vaccine would first be given to those aged ≥ 60 years, regardless of history of COVID-19 55 . If additional doses were available, they were given to those aged 20-59 years, and then to those aged <20 years (trials ongoing) 56-58 . 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint

Equations (4) 

The model tallies resource utilization, including hospital and ICU admissions for those with severe or critical disease, accounting for country-specific capacity constraints. Hospitalization is provided for those with severe illness, and ICU care, if available, is provided for those with critical illness. Costs are from a COVAX payer perspective, and thus include only vaccination . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. prevented) compared to the next least-expensive non-dominated strategy 59 . A strategy is considered "dominated" if it is more expensive and less efficacious (resulting in more years of life lost) than an alternative strategy, or if it is less economically efficient (resulting in a higher ICER) than a more efficacious strategy. In general, dominated strategies are considered unfavorable from a decision-making perspective, and thus are not included when calculating ICERs for the remaining non-dominated strategies.

For each of the 91 countries, we grouped age-stratified population data from the United Nations 2019 World Population Prospects 53 to calculate the proportion of the population in each of the three modeled age strata: 0-19, 20-59, and ≥ 60 years (Supplementary Table 1 ). Hospital and ICU bed capacity for each country were derived from data published by the World Health Organization, the World Bank, and country-level health agencies, as well as from peer-reviewed literature (Supplementary Table 1) .

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint Table 2 ). The probability of developing severe or critical disease, as well as mortality, increases with age 60, 61 . Transmission rates are highest for individuals in asymptomatic and mild/moderate states, whereas individuals in severe and critical states have fewer infectious contacts due to hospitalization or being homebound (Extended Data Table 1) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint SARS-CoV-2 infection (VE 1 ). We believe this is a conservative estimate given that data from the J&J/Janssen phase III clinical trial suggest an efficacy of 74.2% against asymptomatic infection after day 29 post-vaccination in a limited sample of study participants 68 .

We derived the cost of the vaccine program using the COVAX Working Group's February 2021 updated delivery cost estimates 28 and a review of negotiated prices for COVID-19 vaccines 29 .

The COVAX Working Group estimated the costs of delivering sufficient vaccine for 9% of India's population and 20% of the remaining 91 AMC countries' population: a total of 546.3 million people. The estimated upfront cost of vaccine delivery was US$576.4 million and included costs attributed to planning and coordination, training, social mobilization, cold chain equipment, pharmacovigilance, and hand hygiene. To this, we added the estimated US$98.5 million in global and regional level costs (for innovations, post-introduction evaluations, and additional pharmacovigilance) anticipated over a 3-year period. Given that our analysis focuses on an investment from the donor countries' perspective, these global and regional level costs were treated as an upfront investment despite these costs potentially being utilized over a 3-year period. Our estimated total fixed costs to vaccinate 546.3 million people was US$674.9 million, or US$1.24 per person vaccinated. Since our analysis excluded India, we re-scaled the fixed costs to a population of 507.7 million (20% of the population in the 91 included AMC countries) under the assumption that fixed costs are evenly distributed on a per capita basis. For the 91 included AMC countries, this resulted in a total fixed cost of US$630 million for strategies in which 20%-70% of the population is vaccinated (Extended Data Table 1 ).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. Finally, in our base case, we included a per-dose vaccine cost of US$9, based on the negotiated prices of the J&J/Janssen vaccine (US$8.50 by the European Union, US$10 by the United States, and US$10 by the African Union) 29 . UNICEF and Gavi also announced a negotiated price for the Novavax NVX-CoV2373 and AstraZeneca ChAdOx1 nCoV-19 vaccines of US$6 per vaccine course (US$3 per dose) for COVAX-eligible countries 69 . In our base analysis, we chose the US$9 per vaccine course estimate because it is more conservative. However, we adjusted the total program costs to be 50% and 200% of the base case value in sensitivity analyses to account for the uncertainty in delivery costs and variability in per-unit vaccine costs.

To evaluate the influence of key parameters on measures of cost-effectiveness, we performed sensitivity analyses on a subset of representative countries. First, we partitioned our dataset of 91 COVAX AMC-eligible economies (excluding India) into five groups based on age structure, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint hospital beds per capita, and ICU beds per capita using k-means clustering. Next, we selected 1-2 countries from each group to create a subset of 9 countries representing the range of "country domains" observed in our dataset: Afghanistan, Cambodia, Lesotho, Moldova, Mongolia, Morocco, Nicaragua, Sri Lanka, and Zambia (combined population: 148 million).

Under the base case scenario, we were able to reproduce the ICERs of a global vaccination program targeting 91 AMC countries with reasonable accuracy while only considering the costs and benefits (years of life saved) accrued within the chosen subset of 9 countries. The costs of program startup, vaccine purchase, and delivery were scaled to reflect the combined population of these 9 countries. For the 20-40% coverage levels, the mean absolute percentage error (MAPE) between the ICERs calculated using a 9-country approximation and the ICERs calculated using the full dataset of 91 countries was 16% and rose to 30% for the 50-60% coverage levels. The 70% coverage level was found to be dominated when a 9-country approximation was used to calculate ICERs; in contrast, this coverage level was not found to be dominated when the full dataset of 91 countries was used. This suggests that program costeffectiveness may be reasonably approximated from a subset of countries for coverage levels less than 70%.

In one-way sensitivity analyses, we varied multiple parameters: speed of vaccine rollout; vaccine uptake; vaccine efficacy in preventing SARS-CoV-2 infection, any symptomatic COVID-19, and severe or critical COVID-19 disease requiring hospitalization; baseline probability of developing severe/critical disease in the absence of vaccines among those infected (and resulting IFR); prevalence of prior protective immunity; mean epidemic R e ; and total vaccination program costs.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint Variations in vaccine efficacy and costs were ranged to include estimates that correspond to the J&J/Janssen Ad26.COV2.S, Pfizer-BioNTech BNT162b2, and AstraZeneca ChAdOx1 nCoV-19 vaccines 1, 3, 70 , vaccines that are currently being distributed as part of the COVAX-AMC program or are included in future distribution forecasts 71,72 . In a two-way sensitivity analysis, we simultaneously varied the effective IFR in the absence of vaccine and vaccination program costs.

Data and code availability statement: There are no original data generated as part of this study.

All data are derived from published or publically available data and referenced as such. Code data will be made available to any requesting party by contacting the corresponding author.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. 

All authors have no competing interests to report.

Supplementary Information is available for this paper. Correspondence and reqeuests for materials should be addressed to Mark Siedner at msiedner@mgh.harvard.edu. Reprints and permissions information is available at www.nature.com/reprints. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint Extended Data Table 2 . One-way sensitivity analyses: influence of prior immunity to COVID-19 on clinical and economic outcomes across 9 representative countries. Abbreviations: YLS, years of life saved; US$, US Dollar; ICER, incremental cost-effectiveness ratio.

Total population vaccinated, COVID-19 infections, hospital and ICU admissions, deaths, total YLS, and total cost of vaccination are rounded to the nearest thousand. Dollars per incremental infection prevented and ICERs are calculated using unrounded values and then rounded to the nearest ten.

* Dominated strategies are ones that are more costly and less effective than another strategy or ones that have a higher ICER than a more effective strategy, and are thus not economically efficient. In incremental scenarios resulting in small changes in the total number of infections or deaths (e.g., increasing vaccine supply from 60% to 70% or in a scenario with high prevalence of pre-existing protective immunity), strategies may appear to be dominated due to random variation.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint Extended Data Table 3 . One-way sensitivity analyses: influence of risk of infection fatality ratio on clinical and economic outcomes across 9 representative countries. Total population vaccinated, COVID-19 infections, hospital and ICU admissions, deaths, total YLS, and total cost of vaccination are rounded to the nearest thousand. Dollars per incremental infection prevented and ICERs are calculated using unrounded values and then rounded to the nearest ten.

* Dominated strategies are ones that are more costly and less effective than another strategy or ones that have a higher ICER than a more effective strategy, and are thus not economically efficient. In incremental scenarios resulting in small changes in the total number of infections or deaths, strategies may appear to be dominated due to random variation. Table 4 . One-way sensitivity analyses: influence of R e on clinical and economic outcomes across 9 representative countries. Total population vaccinated, COVID-19 infections, hospital and ICU admissions, deaths, total YLS, and total cost of vaccination are rounded to the nearest thousand. Dollars per incremental infection prevented and ICERs are calculated using unrounded values and then rounded to the nearest ten.

* Dominated strategies are ones that are more costly and less effective than another strategy or ones that have a higher ICER than a more effective strategy, and are thus not economically efficient. In incremental scenarios resulting in small changes in the total number of infections or deaths, strategies may appear to be dominated due to random variation.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint Extended Data Table 5 . One-way sensitivity analyses: influence of total vaccination program cost on clinical and economic outcomes across 9 representative countries. Total population vaccinated, COVID-19 infections, hospital and ICU admissions, deaths, total YLS, and total cost of vaccination are rounded to the nearest thousand. Dollars per incremental infection prevented and ICERs are calculated using unrounded values and then rounded to the nearest ten.

* Dominated strategies are ones that are more costly and less effective than another strategy or ones that have a higher ICER than a more effective strategy, and are thus not economically efficient. In incremental scenarios resulting in small changes in the total number of infections or deaths, strategies may appear to be dominated due to random variation.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint Extended Data Table 6 . One-way sensitivity analyses: influence of vaccine uptake on clinical and economic outcomes across 9 representative countries. * Dominated strategies are ones that are more costly and less effective than another strategy or ones that have a higher ICER than a more effective strategy, and are thus not economically efficient. In incremental scenarios resulting in small changes in the total number of infections or deaths, strategies may appear to be dominated due to random variation. † A given level of vaccine supply will be distributed differently across age groups depending on uptake. In scenarios with increased uptake, high-priority age groups (i.e., older individuals) will . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint receive a greater share of doses since more are willing to be vaccinated. In scenarios with decreased uptake, there will be more doses left over for younger individuals since fewer people in older age groups are willing to be vaccinated. ‡ Scenarios in which vaccine supply exceed vaccine uptake have the same clinical outcomes compared to the strategy in which vaccine supply equals vaccine uptake. Greater vaccine supply results in additional cost regardless of vaccine uptake.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) * Dominated strategies are ones that are more costly and less effective than another strategy or ones that have a higher ICER than a more effective strategy, and are thus not economically efficient. In incremental scenarios resulting in small changes in the total number of infections or deaths, strategies may appear to be dominated due to random variation.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Total population vaccinated, COVID-19 infections, hospital and ICU admissions, deaths, total YLS, and total cost of vaccination are rounded to the nearest thousand. Dollars per incremental infection prevented and ICERs are calculated using unrounded values and then rounded to the nearest ten.

* Dominated strategies are ones that are more costly and less effective than another strategy or ones that have a higher ICER than a more effective strategy, and are thus not economically efficient. In incremental scenarios resulting in small changes in the total number of infections or deaths, strategies may appear to be dominated due to random variation.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint symptomatic disease on clinical and economic outcomes across 9 representative countries. Total population vaccinated, COVID-19 infections, hospital and ICU admissions, deaths, total YLS, and total cost of vaccination are rounded to the nearest thousand. Dollars per incremental infection prevented and ICERs are calculated using unrounded values and then rounded to the nearest ten.

* Dominated strategies are ones that are more costly and less effective than another strategy or ones that have a higher ICER than a more effective strategy, and are thus not economically efficient. In incremental scenarios resulting in small changes in the total number of infections or deaths, strategies may appear to be dominated due to random variation.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 2, 2021. ; https://doi.org/10.1101/2021.04.28.21256237 doi: medRxiv preprint Extended Data Table 10 . One-way sensitivity analyses: influence of vaccine efficacy against infection on clinical and economic outcomes across 9 representative countries. Abbreviations: YLS, years of life saved; US$, US Dollar; ICER, incremental cost-effectiveness ratio.

Total population vaccinated, COVID-19 infections, hospital and ICU admissions, deaths, total YLS, and total cost of vaccination are rounded to the nearest thousand. Dollars per incremental infection prevented and ICERs are calculated using unrounded values and then rounded to the nearest ten.

* Dominated strategies are ones that are more costly and less effective than another strategy or ones that have a higher ICER than a more effective strategy, and are thus not economically efficient. In incremental scenarios resulting in small changes in the total number of infections or deaths, strategies may appear to be dominated due to random variation.

Safety and efficacy of single-dose Ad26.COV2.S vaccine against Covid-19

Safety and efficacy of the BNT162b2 mRNA Covid-19 vaccine

Safety and efficacy of the ChAdOx1 nCoV-19 vaccine (AZD1222) against SARS-CoV-2: an interim analysis of four randomised controlled trials in Brazil, South Africa, and the UK

Efficacy and safety of the mRNA-1273 SARS-CoV-2 vaccine

directorgeneral/speeches/detail/director-general-s-opening-remarks-at-the-media-briefing-on-covid

COVID-19 vaccines: how to ensure Africa has access

The Gavi COVAX AMC Investment Opportunity

Cost-effectiveness of public health strategies for COVID-19 epidemic control in South Africa: a microsimulation modelling study

An ethical framework for global vaccine allocation

Pfizer-BioNTech announce positive topline results of pivotal COVID-19 vaccine study in adolescents

Calculating and interpreting ICERs and net benefit

Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus-infected pneumonia in Wuhan, China

Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study

Report of the WHO-China Joint Mission on Coronavirus Disease

Clinical characteristics of 24 asymptomatic infections with COVID-19 screened among close contacts in Nanjing

The reproductive number of COVID-19 is higher compared to SARS coronavirus

SeroTracker: a global SARS-CoV-2 seroprevalence dashboard

High SARS-CoV-2 seroprevalence in health care workers but relatively low numbers of deaths in urban Malawi

CC-BY-NC-ND 4.0 International license It is made available under a