The societal value of SARS-CoV-2 booster vaccination in Indonesia

Objectives To estimate the expected socio-economic value of booster vaccination in terms of averted deaths and averted closures of businesses and schools using simulation modelling. Methods The value of booster vaccination in Indonesia is estimated by comparing simulated societal costs under a twelve-month, 187-million–dose Moderna booster vaccination campaign to costs without boosters. The costs of an epidemic and its mitigation consist of lost lives, economic closures and lost education; cost-minimising non-pharmaceutical mitigation is chosen for each scenario. Results The cost-minimising non-pharmaceutical mitigation depends on the availability of vaccines: the differences between the two scenarios are 14 to 19 million years of in-person education and $153 to $204 billion in economic activity. The value of the booster campaign ranges from $2,500 ($1,400-$4,100) to $2,800 ($1,700-$4,600) per dose in the first year, depending on life-year valuations. Conclusions The societal benefits of booster vaccination are substantial. Much of the value of vaccination resides in the reduced need for costly non-pharmaceutical mitigation. We propose cost minimisation as a tool for policy decision-making and valuation of vaccination, taking into account all socio-economic costs, and not averted deaths alone.


Introduction
Vaccines are highly effective in preventing severe disease and deaths from SARS-CoV-2 infection. It is estimated that, between December 2020 and December 2021, COVID-19 vaccines prevented 19.8 million deaths in 185 countries and territories [35]. However, the number of prevented deaths is only one component of the benefits of vaccination [5]. Most countries have implemented nonpharmaceutical interventions (NPIs), with mandated closures of businesses and schools. NPIs limit transmissions, but also economic activity and in-person education, and they are associated with high economic and social costs [4]. Countries may eventually recover from the costs of business closures, but lost education could cause substantial economic costs in years to come [16]. Interrupted schooling has been shown to reduce productivity of the workforce in the long term [24], with adverse consequences on countries' economic growth and welfare.
Vaccination may not only avert deaths but also reduce the need for costly NPIs: as vaccination coverage increases and transmission decreases, countries can gradually scale back NPIs, allow economic activity to resume and schools to re-open for in-person education. This implies that the benefits of vaccination are a complex combination of prevented deaths, averted economic costs, and averted disruption to education. It is challenging to estimate the value of vaccination across wider socio-economic dimensions [5].
The need to estimate benefits was not as pressing for the first round of SARS-CoV-2 vaccines, because there was little doubt that their benefits exceeded their costs [6,26,32,36,39]. However, for subsequent rounds of booster vaccines the case is less clear. Uncertainty about how fast NPIs can be scaled back as vaccination coverage increases -and the associated economic benefitscomplicates the analysis. Moreover, the projections of health impacts are challenged by uncertainty about immunity against disease in the human body after several rounds of vaccination, the role of infection-induced immunity, emergence of SARS-CoV- variants that are partially or fully resistant to vaccine or infectioninduced immunity, strength and duration of waning vaccine-and infection-induced immunity, and many other factors [28].
Repeated booster campaigns will be an integral part of sustainable control strategies against SARS-CoV-2 in the years to come [28,34]. This necessitates careful analyses of their socio-economic value considering many competing priorities for constrained health budgets. However, there is little evidence on the costeffectiveness of booster vaccination [22].
We quantify the socio-economic costs of an epidemic in terms of deaths, gross domestic product (GDP) lost, and education lost, using monetary valuations of lives [27], education [16], and economic activity. An existing integrated economic-epidemiological model DAEDALUS [17] is adapted to project socio-economic outcomes of the epidemic in Indonesia. We evaluate a prospective booster vaccination campaign against a no-booster counterfactual, starting from September 2021 with a one-year projection horizon, parametrising the model using publicly available data. Taking the perspective of a social planner making policy decisions in August 2021, we find the optimal pandemic mitigation via NPIs that minimises expected socio-economic costs for each scenario. This approach allows us to estimate the value of SARS-CoV-2 booster vaccination arising from averted deaths and reduced need for mitigation via business and school closures as the difference in costs between scenarios.

DAEDALUS: Integrated economic-epidemiological model
The basis of the present work is the integrated economicepidemiological model DAEDALUS [17]. DAEDALUS is used to fit and simulate the dynamics of the epidemic. (Details of the data inputs and the model are given in Appendices 1 and 2.) It includes a compartmental epidemic model stratified by age and sector of employment. At the core of DAEDALUS lies the insight that relatively contact-light sectors which employ fewer workers carry less infection back into the community when they are open compared to more contact-intensive sectors with more workers. Partial or full opening and closing of a sector leads to changes in the sector's active workforce, and disease transmission in the workplace, during transport and in the community. It also changes the sector's associated contribution to the economy, in the form of gross value added (GVA). GVA is the value of a sector's output minus the value of products from other sectors that are used in production. GDP is the sum of all sectors' GVA.
The epidemiological model has one compartment for each sector, plus compartments for four age groups (ages 0-4, 5-19, 20-64, and 65 plus). In each month, policy makers can choose an economic configuration that specifies the percentage closure of each sector. We fit the model up to the start of the simulation using observed data. For the simulation period, the configuration is a policy choice. We refer to this choice of economic configuration, which describes the extent of mandated business and school closures over time, as the NPI trajectory. This trajectory determines the economic output, the amount of in-person teaching, and the extent of transmission between different population groups via the numbers of individuals in schools and workplaces.

Defining socio-economic costs
The trade-off between health and economic objectives is captured in the socio-economic costs associated with a specific NPI trajectory. We assign monetary values to health and educational impacts so they can be added to the costs of business closures, yielding a total socio-economic cost (TSC). This enables comparisons between trajectories, allowing us to choose TSC-minimising background NPIs. The TSC includes immediate costs as well as mid-to long-term costs, which are all attributable to the epidemic and the NPI trajectory over the projection period. The TSC is added up over all years for which the loss is expected to persist, using the same discounting rate of 3%. All impacts are considered costs because they are evaluated against a hypothetical no-pandemic scenario. Here, we describe the individual components of the TSC in turn; a summary of the costs we take into account and those we do not is given in Table S7.

Valuing lost lives
The value of a statistical life (VSL) reflects the value that members of the society place on reductions of their own mortality. We use existing estimates of the VSL in Indonesia to generate a plausible range for the value of a life year (VLY). We interpret the VSL as a population-weighted average [3,27], where each age group has a VSL defined by the expected number of life-years remaining [13], and where each discounted year (at rate 3%) has the same value. The total years of life lost (YLL) for a scenario is the sum of the expected YLL per death over all deaths. The socio-economic cost is the total monetised YLL, calculated as the product of the VLY and the total discounted YLL. See Appendix 3.1 for additional details.

Valuing lost economic output
We assume that closures of businesses (except schools) in 2021 and beyond impact short-and mid-term economic output but not long-term growth and GDP of the Indonesian economy, following IMF projections [18]. We measure the short-term economic loss due to mandated closures of non-essential businesses in terms of lost aggregate GVA compared to pre-pandemic GVA, where output is scaled to the active and healthy workforce in each sector. All economic sectors contribute GVA according to the level they are open for production, except for the education sector which contributes its pre-pandemic GVA because the long-term GDP loss of school closures is calculated separately.
In addition, we model the mid-term economic impact of the pandemic and mandated closures. Following projections by the IMF [18,19], we assume in our projections that Indonesia will not suffer additional permanent damage from losses in 2021 and 2022, after the initial severe losses in 2020. Instead, we assume that the economy will eventually return to the IMF-projected GDP path [18,19]. The consequence of business closures is that GDP is lower in the projection period than was forecast. Then, in projecting the scenario GDP forward in time, we assume that it re-joins the IMF-projected GDP path after five years. The midterm economic cost is the cumulative loss arising from departure of the scenario GDP path from the IMF-projected path. Full details are given in Appendix 3.2.

Valuing lost education
Lost education has a long-term negative impact on the productivity of the workforce and on economic growth. We estimate education loss in person-years, which is quantified as the effective number of years of schooling lost, considering the extent of school closure, the number of people aged 5-19, and the reduced educational value of remote compared to in-person teaching. We use an estimate of effectiveness of remote teaching that considers infrastructure coverage, household access and effectiveness of distance learning in Indonesia [38].
We treat the per-person value of a school year (VSY) in Indonesia as an uncertain variable, as we do not have an estimate of its true value but can use two relevant estimates from other settings to generate a plausible distribution. The first value is $34,000, based on a costing of a lost year of education at 202% of current GDP for OECD countries [16]. The second is $15,000, based on a costing of a lost year of education for the Philippines at 89% of GDP [23]. Full details are given in Appendix 3.3.

Scenarios
TSCs are compared for two scenarios over a twelve-month projection horizon starting September 2021: No boosters (counterfactual scenario), and coverage of 80% in three target age groups (187 million Moderna boosters) by the end of the period (see Appendix 5.2 for an alternative scenario with 40% coverage). The target groups for vaccination are school-age individuals, working-age adults, and people aged 65 or over, and we assume a constant administration rate with no particular prioritisation.
The value of the booster campaign is the difference between the TSC of the scenario and the TSC of the counterfactual. The value represents the prospective societal gain for Indonesia in monetary terms posed by the booster campaign. We repeat the analysis using seven different VLYs ranging from $20,000 to $80,000, covering the range of values a policymaker might choose based on estimates for the value of a statistical life in Indonesia (from $592,000 [33] to $1,910,000 [27]).

Decision framework for choosing background non-pharmaceutical interventions
We allow the NPI trajectory to depend on the number of booster vaccines administered and the life-year valuation, finding the socially optimal trajectory in each case. This ensures that scenario-specific background NPIs generate the highest attainable societal welfare, so that when we evaluate the campaign, we are comparing best-case scenarios. We choose the optimal NPI trajectory from among a finite number of options, given by combinations of levels of monthly business closure and school opening. We simplify the trajectories as follows: there is an initial period of four months with a constant configuration, a six-month period where each configuration changes in uniform increments, and a final period of two months where the configurations remain the same. See Appendix 4.2 for details.
We treat the choice of optimal NPI trajectory as a decision made under uncertainty. We compute the TSC 1,000 times per scenario per decision option with randomly varying parameters in order to make decisions while accounting for uncertainty in some model parameters. The uncertain parameters are the ''transmission modifier" (random changes to transmission month by month that are not due to mandated closures), the relative vaccine effect (which reflects ''vaccine escape" in new SARS-CoV-2 variants), the change in hospitalisation rate (reflecting a change in severity of variant), and the VSY. The parameter distributions are described in Appendix 4.3 For each scenario, the optimal decision under uncertainty is the one that minimises the expected TSC [7].
The parameter distributions chosen reflect our uncertainties about the future into which we are projecting. This method of simulation enables value-of-information sensitivity analyses for the uncertain variables, such as the expected value of partial perfect information (EVPPI). This is the expected gain in learning a parameter (or set of parameters) perfectly with reference to a particular objective of reducing an expected loss or estimating a quantity with better precision [20]. This can be used as a means of research prioritisation, suggesting what further information might be required in order to make a better-informed decision.

Optimal background NPI mitigation depends on availability of vaccines and the valuation of a life year
The TSC-minimising NPI trajectories for the two scenarios and two of the seven VLYs are shown in Fig. 1. For the counterfactual (no boosters), the cost-minimising trajectory is to keep businesses closed throughout, while allowing schools to reopen slightly. The booster-campaign scenario begins with a higher degree of opening in both businesses and schools, and reopens with a steeper gradient. The difference in the socially optimal number of person-years of in-person education ranges from 14 to 19 million (Table 1).
For both scenarios, lower VLYs are associated with more open trajectories because lower valuations of a life year imply that the social planner is willing to accept a higher number of deaths in return for lower economic costs and more in-person schooling. The difference in the socially optimal number of in-person school years between low and high VLYs ranges from 4 to 9 million ( Table 1).
The scenarios have similar numbers of expected deaths. For the booster scenario, central estimates range between 97,000 and 380,000 for VLYs of $80,000 and $20,000, compared to the counterfactual with central estimates between 110,000 and 540,000. For some VLYs, the expected number of deaths is higher in the booster scenario, and the uncertainty around the estimates is high, driven by uncertainty in transmission. This implies that a booster campaign can increase societal benefit without reducing the expected number of deaths: when minimising TSC, there is more to be gained in reopening schools and businesses than in averting additional deaths.

Value of the booster vaccination campaign
We estimate the value of the booster campaign to be $470 billion (95% CI $270-770, VLY=$20,000) to $530 billion (95% CI $320-860, VLY=$70,000) ( Table 2). The value per dose ranges from $2,500 (95% CI $1,400-$4,100, VLY=$20,000) to $2,800 (95% CI $1,700-$4,600, VLY=$70,000). The value per dose is higher for a smaller campaign aiming for 40% coverage (see Appendix 5.3). Table 1 shows how the TSC is distributed across the three outcomes. There is a substantial reduction in expected economic losses from business closures with the booster campaign, with central estimates ranging from $153 billion (VLY=$20,000 and $80,000) to $204 billion (VLY=$70,000). The reduction in the expected long-term economic loss from lost schooling exceeds that from business closures, ranging from $254 billion (VLY=$20,000) to $341 billion (VLY=$60,000 and $80,000). Fig. 2 shows the medians, interquartile ranges and extreme values of TSC associated with each scenario and the value of the booster campaign for all seven VLYs, demonstrating that our results have high uncertainty, driven by uncertainty in key parameters.

Discussion
Vaccination strategies are often evaluated in terms of deaths averted [22]. However, most countries adopt some form of mitigation with non-pharmaceutical interventions, most notably business and school closures. This implies that vaccination and closure strategies are complements: as vaccination increases, closures can be reduced. This makes it important to evaluate the societal impact of vaccines comprehensively, and not only in terms of deaths. Our framework takes such evaluations into account. The method we present converts the schooling-economy-health trade-off for mitigation strategies into a single metric: the total socio-economic cost (TSC). The trade-off made by the decision maker is made explicit through the valuations made for a year of life and a year of education, allowing mitigation strategies to be evaluated and compared.
We evaluated the socio-economic costs of a booster campaign in Indonesia (187 million Moderna doses with 80% targetpopulation coverage) and a no-booster counterfactual, both with constant administration of two doses of the Sinovac vaccine. An 80% population coverage of booster doses was attained by a small number of countries by the end of August 2022 [10]. We use this target to represent the best-case scenario. A coverage of 40%, the target we chose for the alternative scenario, is more attainable, with more than one third of countries reaching this coverage by the end of August 2022. Higher coverage is achieved by higherincome countries [37], and for the time period considered, Indonesia was classified as a lower-middle-income country [15]. By the end of the time horizon considered here, Indonesia had administered booster doses to 22% of the population [10].
Each scenario can be considered as a single instance of policy planning: using a decision framework to choose the best NPI trajectory given an uncertain future. By comparing the scenarios' results, we can estimate the value of the booster campaign as the difference between these two best-case scenarios. In effect, we are controlling for policy-makers' choices, and thereby extract a meaningful value: the difference in costs assuming a policy is chosen to minimise the cost. This stands in contrast to a retrospective analysis, for which an assumption on the counterfactual is required. Often, the assumption made is that behaviour and other factors would be exactly the same as in the intervention scenario [35].
We estimate the value of the booster campaign in the first year to be between $470 and $540 billion, depending on life-year valuations. This comes primarily from expected gains of 14-19 million person-years of in-person education, and economic gains from reduced business closures of $153-$204 billion. Both booster and no-booster scenarios have similar numbers of expected deaths with high variance, which is driven by high uncertainty in key parameters.
We estimate each Moderna booster dose to be worth $2,500-$2,800 in the first year. If we assume the cost of acquisition and administration to be no more than $30 per dose [9,14], the returns on investments are at least $83-$93 for every $ spent. This would constitute a substantial societal gain associated with booster vaccination.
We are aware of only one peer-reviewed study on the costeffectiveness of booster vaccination [22]. The benefit-cost ratio of boosters was estimated at 1.95 in 180 days, lower than our estimates corresponding to a 12-month horizon. However, comparison of this study with ours is limited, because it is from a high-income country (USA), is conducted from a health-system perspective, does not quantify the wider socio-economic benefits, and models administration of boosters to only 0.03% of the population.
It was estimated that 27.51 (23.59-31.76) deaths were averted per 10,000 vaccines (first two doses) in Indonesia during the year ending 8 December 2021 [35]. Applying the smallest (or largest) VSL implies the vaccines were worth $1,600 ($1,400-1,900) (or $5,200 ($4,500-6,100)) per dose. These values are the same order of magnitude as our estimates; we note that their range is wider, and the maximum value is larger, as the counterfactual used assumes the same behaviour without vaccination.
The value of a vaccine depends not on its absolute profile of effects (i.e. risk compared to a fully susceptible person) but on its profile relative to existing immunity. Here, the important comparison is between the double dose of Sinovac and the Moderna booster (Table S3), while infection-acquired immunity is also of consequence due to its impact on prevalence (the higher the prevalence across the time horizon, the greater the impact of the vaccine).
Vaccines are most impactful when delivered to populations with no or little immunity. Therefore, all else equal, we would expect the benefit of the first schedule to exceed the benefit of the booster, and, due to their profiles, we would expect a Moderna dose to generate a greater benefit than a Sinovac dose. A vaccination programme will be more impactful in a population with lower historical prevalence: from our epidemic model fit to hospital occupancy data, we estimate that, at the beginning of the time horizon, 11.4% of the population had been infected recently and were not yet susceptible to reinfection. When this value is higher, the impact of the vaccination programme is less. We also expect diminishing marginal returns to the vaccine rollout: additional vaccines administered once substantial population immunity has been achieved yield less additional benefit, as our valuations per dose of 40% coverage vs 80% coverage suggest (Table S9).
In sensitivity analyses, we find that our conclusions are most sensitive to two uncertain parameters: the ''transmission modifier" and the VSY. The contributions from these sources of uncertainty are far greater than those of the other epidemic parameters. Each of these parameters represents the combination of multiple processes. To improve model precision, we would need to first model these mechanisms more explicitly.

Limitations
We do not model the nuanced relationships between education, economic closures and health outcomes in computing the TSC. Such relationships include the impact of school closure on the available workforce [25,8], the impact of economic closure on the availability of funds for tuition, impacts of illness on education, and the impact of economic closures on the ability of populations to access healthcare. We use pre-pandemic valuations of life years and years of education, and do not adjust them for possible pandemic-related changes in GDP, wages, employment, or other factors.
Important socio-economic costs are not considered, such as increased educational inequalities due to school closures disproportionately disadvantaging students who are female [1] or from lower-income households [11,12,21]. We do not consider tertiary education or distinguish between primary and secondary education, as independent estimates of the value of education and the effectiveness of remote teaching are not available at this resolution. We also do not consider COVID-19-driven long-term disability, the cost of lives lost due to deferred medical care for other conditions, or the impacts of increasing inequality in society [29,31].
To estimate the cost of GDP loss, we assume that complete economic recovery will be achieved in five years. This assumption may be less plausible for large losses, and for small, fast-growing emerging economies [4,30].
The epidemic model does not include a population behavioural response to high prevalence. This likely leads to case numbers that are too high for NPI trajectories with little or no mitigation. Uncertainty around future population behaviour is captured in the uncertain transmission modifier, for which we estimate high EVPPI. Together, these suggest that behavioural response is an important area of future research. Another important research priority is estimation of contact rates in work settings, for which informative data are sparse. Last, our findings may be affected by the large informal economy in Indonesia, which is difficult to measure in national accounts [2]. Adherence to mandated closures is potentially less in the informal economy, which means that we may be overestimating transmission reduction and economic loss for NPI trajectories with substantial business closures.

Conclusions
The question of booster vaccination will dominate the discussion on our response to COVID-19 in 2023 and beyond. The framework we present supports countries in making such decisions whenever school or business closures are considered as policy options, by allowing estimation of the expected costs. As countries will hopefully move out of the pandemic and into the endemic phase, the need and also societies' tolerance for business and school closure will likely reduce. There will also likely be political pressure to compare SARS-CoV-2 boosters against other worthwhile public health investments that were neglected during the pandemic years.  A comprehensive evaluation of the benefits and costs of boosters is difficult due to the challenges of assessing its wider societal benefits, the complicated epidemiology of the virus, and the interaction with complex dynamic economic and social impacts. Modelling needs to consider existing vaccination coverage, rates of waning of infection-and vaccine-induced immunity, vaccineeffect profiles, and the emergence of variants. It needs to consider impacts on economic production, and the long-term detrimental effect of school closures on economic growth. Smart mitigation strategies that combine pharmaceutical and non-pharmaceutical interventions will preoccupy public health policy making in the months and years to come. Our approach allows policy makers to adopt an encompassing perspective on a problem that acknowledges that overarching societal objectives, constraints and guidelines have to be combined to take decisions.
Author contributions. RJ, DH, PD, GF, MP, PS and KH conceived and designed the work; RJ undertook the analysis and interpretation of data; BD, DH and PD contributed to the analysis of data; BD, GF, MP, PS, and KH contributed to the interpretation of data; RJ, DH and PD created new software used in the work; RJ and KH wrote the first draft of the paper and DH, PD, GF and PS substantially revised it. All authors have approved the submitted version and have agreed to be personally accountable for the author's own contributions and to ensure that questions related to the accuracy or integrity of any part of the work, even ones in which the author was not personally involved, are appropriately investigated, resolved, and the resolution documented in the literature.
Ethical approval statement Not applicable.

Data availability
All data used are publicly available.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.