## Introduction

The development and licensure of coronavirus disease 2019 (COVID-19) vaccines offer a critically important opportunity to curtail the global COVID-19 pandemic1,2,3,4. Even before the efficacy and safety of the leading vaccine candidates were established, many high-income countries (HICs) pre-emptively procured stocks of doses in excess of population need5. By contrast, most low- and middle-income countries (LMICs) do not have access to sufficient quantities of vaccine due to cost, limitations in available doses, and logistical challenges of production, distribution, and storage6. Meanwhile, the Africa Centres for Disease Control and Prevention have announced a goal of vaccinating 60% of Africans by the end of 20227.

There has been much discussion about reported efficacies and costs of different vaccines. However, factors specific to implementation, including vaccine supply, vaccination pace, and acceptance among communities, are increasingly recognized to be crucial to the effectiveness of a vaccination program in promoting epidemic control in HICs—in some cases, even more so than vaccine efficacy8,9,10,11. How these program implementation factors will affect the clinical and health economic consequences of COVID-19 in LMICs has not been well-defined. This is a particularly urgent question given the emergence of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) variants, such as B.1.351 in South Africa, which appear to partially reduce efficacy of some vaccines4,12,13,14,15.

In this work, we use a microsimulation model to estimate the clinical and economic outcomes of COVID-19 vaccination programs in South Africa, examining different implementation strategies that policymakers could directly influence. We simulate COVID-19-specific outcomes over 360 days, including daily and cumulative infections (detected and undetected), deaths, years-of-life lost (YLL) attributable to COVID-19 mortality, resource utilization (hospital and intensive care unit [ICU] bed use), and health care costs from the all-payer (public and private) health sector perspective. We examine different strategies of vaccination program implementation under multiple scenarios of vaccine effectiveness and epidemic growth, thereby projecting which factors have the greatest impact on clinical and economic outcomes and cost-effectiveness. Our goal was to inform vaccination program priorities in South Africa and other LMICs.

## Results

### Clinical and economic benefits of vaccination strategies

To understand the trade-offs inherent to policy decisions regarding the total vaccine supply to purchase and the speed with which to administer vaccinations, we compared the clinical and economic outcomes of different strategies of population coverage (vaccine supply) and vaccination pace. We determined the incremental cost-effectiveness ratio (ICER) of each strategy as the difference in health care costs (2020 USD) divided by the difference in years-of-life saved (YLS) compared with other strategies of supply and pace. We considered multiple scenarios of epidemic growth, including a scenario in which the effective reproduction number (Re) varies over time to produce two waves of SARS-CoV-2 infections.

In both the Re = 1.4 scenario and the two-wave epidemic scenario, the absence of a vaccination program resulted in the most infections (~19–21 million) and deaths (70,400–89,300), and highest costs (~$1.69–1.77 billion) over the 360-day simulation period (Table 1). Vaccinating 40% of the population decreased deaths (82–85% reduction) and resulted in the lowest total health care costs (33–45% reduction) in both scenarios. Increasing the vaccinated population to 67%, the government’s target for 2021, decreased deaths and raised costs in both scenarios. Increasing the vaccine supply to 80%, while simultaneously increasing vaccine acceptance to 80%, reduced deaths and raised costs even further in both scenarios. In the Re = 1.4 scenario, the 67% supply strategy was less efficient (had a higher ICER) than the 80% supply strategy and the latter had an ICER of$4270/YLS compared with the 40% supply strategy. In the two-wave epidemic scenario, the 67% and 80% supply strategies had ICERs of $1,990/YLS and$2,600/YLS, respectively. A vaccine supply of 20%, while less efficient than higher vaccine supply levels, still reduced deaths by 72–76% and reduced costs by 15–32% compared with no vaccination. The highest vaccination pace, 300,000 vaccinations daily, resulted in the most favorable clinical outcomes and lowest costs compared with lower paces in both the Re = 1.4 and the two-wave epidemic scenarios (Table 1).

Supplementary Table 1 details the differences between a reference vaccination program (supply 67%, pace 150,000 vaccinations/day) and no vaccination program in age-stratified cumulative infections and deaths, hospital and ICU bed use, and health care costs. The reference vaccination program reduced hospital bed days by 67% and ICU bed days by 54% compared with no vaccination program.

When varying both vaccine supply and vaccination pace across different scenarios of epidemic growth (Re), a faster vaccination pace decreased both COVID-19 deaths and total health care costs, whereas the impact of a higher vaccine supply on deaths and costs varied (Table 1 and Supplementary Table 2). In all four Re scenarios, a vaccination strategy with supply 40% and pace 300,000/day resulted in fewer deaths and lower costs than a strategy with higher supply (67%) and slower pace (150,000/day). At a vaccination pace of 300,000/day, increasing the vaccine supply from 40% to 67% was cost-saving in the two-wave epidemic scenario, whereas it resulted in ICERs of $520/YLS when Re = 1.4,$1160/YLS when Re = 1.8, and $85290/YLS when Re = 1.1. ### Sensitivity analysis: vaccine characteristics and alternative scenarios To understand the influence of extrinsic factors (i.e., those outside the direct control of vaccination program decision-makers, such as vaccine effectiveness and costs and epidemic growth), we performed sensitivity analyses in which we varied each of these factors. In each alternative scenario, we projected clinical and economic outcomes and determined the ICER of a reference vaccination program (67% vaccine supply, 150,000 vaccinations/day, similar to stated goals in South Africa) compared with no vaccination program16,17,18. In one-way sensitivity analysis, the reference vaccination program remained cost-saving compared with a scenario without vaccines across different values of effectiveness against infection, effectiveness against mild/moderate disease, effectiveness against severe/critical disease, and vaccine acceptance (Table 2). When increasing the cost per person vaccinated up to$25, the vaccination program remained cost-saving. At cost per person vaccinated between $26 and$75, the vaccination program increased health care costs compared with a scenario without vaccines, but the ICERs increased only to $1500/YLS (Table 2). The reference vaccination program had an ICER <$100/YLS or was cost-saving compared with a scenario without vaccines across different values of prior immunity (up to 40%), initial prevalence of active COVID-19, reduction in transmission rate among vaccinated but infected individuals, and costs of hospital and ICU care (Table 2 and Supplementary Table 3). When there was 50% prior immunity, the vaccination program still reduced deaths but it increased costs, with an ICER of $22,460/YLS compared with a scenario without vaccines. Notably, when excluding costs of hospital care and ICU care, and only considering costs of the vaccination program, the program increased costs, but its ICER compared with no vaccination program was only$450/YLS (Supplementary Table 3). When several of the main analyses were repeated with lower costs of hospital and ICU care, some ICERs increased, but vaccine supplies of 40% or 80% remained non-dominated (with the latter providing greater clinical benefit), whereas a faster vaccination pace still resulted in greater clinical benefit and lower costs (Supplementary Table 4).

The influence of different scenarios into which the vaccination program would be introduced on cumulative infections, deaths, and health care costs is depicted in Fig. 1. Varying the prevalence of prior immunity and Re had the greatest influence on both infections and deaths, whereas varying the cost per person vaccinated had the greatest influence on health care costs. Vaccine effectiveness against infection and effectiveness against severe disease requiring hospitalization were more influential than effectiveness against mild/moderate disease in terms of reductions in deaths and costs.

### Multi-way sensitivity analyses

In a multi-way sensitivity analysis in which we simultaneously varied vaccine effectiveness against infection and cost per person vaccinated, the reference vaccination program was cost-saving compared with a scenario without vaccines when cost per person vaccinated was $14.81, even when effectiveness against infection was as low as 20% (Fig. 2). When cost per person vaccinated was$25, the program was cost-saving when effectiveness against infection was at least 40%. Even at the highest examined cost per person vaccinated ($75) and the lowest examined effectiveness against infection (20%), the vaccination program had an ICER <$2000/YLS compared with no vaccination program (Fig. 2).

We performed several additional multi-way sensitivity analyses in which we simultaneously varied combinations of vaccine supply, vaccination pace, vaccine effectiveness against infection, cost per person vaccinated, Re, and prevalence of prior immunity (Table 3 and Supplementary Figs. 48). Of note, to optimize efficiency, increasing vaccination pace was more important than increasing vaccine supply. At a cost of $45 or$75 per person vaccinated, increasing vaccination pace led to similar or lower ICER (greater economic efficiency), while increasing vaccine supply led to a similar or higher ICER (less economic efficiency) (Supplementary Fig. 4). At a cost up to $25 per person vaccinated, the vaccination program was cost-saving under nearly all strategies and scenarios (Supplementary Figs. 46). Even when the vaccination program increased costs, the ICERs were <$2000/YLS compared with a scenario without vaccines (Supplementary Figs. 46).

## Discussion

Using a dynamic COVID-19 microsimulation model, we found that vaccinating 67% of South Africa’s population, meeting the government’s goal for 202116, would both decrease COVID-19 deaths and reduce overall health care costs compared with a scenario without vaccines or with a 20% vaccine supply, by reducing the number of infections, hospitalizations, and ICU admissions. Further increasing the vaccine supply to 80%, while simultaneously increasing vaccine acceptance, would save even more lives while modestly increasing costs. Vaccination pace—the number of vaccine doses administered daily--rather than supply itself may be most influential to maximizing public health benefits and economic efficiency. Increasing the pace would reduce both deaths and overall health care costs. The program remained cost-saving even with conservative estimates of vaccine effectiveness and with higher per-person vaccination costs, highlighting that the characteristics of vaccination program implementation are likely to be more influential than the characteristics of the vaccine itself. Furthermore, the vaccination program remained economically efficient (either cost-saving or with a relatively low ICER representing good clinical value for additional money spent) across most epidemic scenarios, including various rates of epidemic growth and a broad range of prevalence of prior population immunity. Although there is no consensus on an ICER threshold for cost-effectiveness in South Africa, for context, the country’s gross domestic product per capita in 2019 was ~$6000 and a published South Africa cost-effectiveness threshold from an opportunity cost approach was ~$2950 (2020 USDs) per disability-adjusted life-year averted19,20.

Much has been made about differences in the leading vaccine candidates and the impact of variants, such as the B.1.351 (β) variant, which eventually accounted for over 90% of SARS-CoV-2 infections in South Africa and the B.1.617.2 (δ) variant, on vaccine effectiveness4,15. However, we found that, even with substantially lower vaccine efficacy than reported in clinical trials, vaccination programs would prevent the majority of COVID-19 deaths compared to scenarios without vaccines. For example, decreasing vaccine effectiveness against mild/moderate disease and severe/critical disease requiring hospitalization to 40% still reduced COVID-19 deaths by 65,800 (74%) compared with a scenario without vaccines. Although efficacy against symptomatic and severe disease have been the focus of vaccine trials, these parameters were less influential on population-wide health and cost outcomes than efficacy against infection, which is less commonly reported in trials1,2,3,4. Nonetheless, the effectiveness ranges we examined in sensitivity analysis include the point estimates of efficacy against symptomatic and severe disease reported in clinical trials of the AstraZeneca ChAdOx1, Moderna mRNA-1273, and Pfizer-BioNTech mRNA BNT162b2 vaccines1,2,3. This suggests that all of these vaccines are likely to have both health and economic benefits. Furthermore, our sensitivity analysis examining different Re scenarios likely captures the potential influence of more contagious SARS-CoV-2 variants such as δ.

### Validation

We previously validated our natural history assumptions by comparing model-projected COVID-19 deaths with those reported in South Africa24. We updated our validation by comparing the model-projected number of COVID-19 infections and deaths with the number of cases and deaths reported in South Africa through 10 April 2021, accounting for underreporting (Supplementary Methods and Supplementary Fig. 3)40,51.

### Sensitivity analysis

We used sensitivity analysis to examine the relative influence on clinical and cost projections of various parameters around vaccine characteristics and epidemic growth (Table 3). Specifically, we varied the following: vaccine acceptance (50–90% among eligible individuals); vaccine effectiveness in preventing infection (20–75%), mild/moderate disease (29–79%), and severe/critical disease requiring hospitalization (40–98%); cost (\$9–75/person); initial prevalence of COVID-19 disease (0.05–0.5%); initial Re (1.1–1.8); prior immunity (10–50% of population); reduction in transmission rate among vaccinated but infected individuals (0–50%); and hospital and ICU daily costs (0.5×–2.0× base case costs). The ranges of vaccine effectiveness against mild/moderate disease and severe/critical disease requiring hospitalization were based on efficacies and 95% confidence intervals reported in the Johnson & Johnson/Janssen vaccine trial (Supplementary Methods)4. We also examined ICERs when the relatively high costs of ICU care were excluded and when all hospital care costs (non-ICU and ICU) were excluded. We performed multi-way sensitivity analyses in which we simultaneously varied parameters influential in one-way sensitivity analyses.

### Reporting summary

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