Cost-effective boosting allocations in the post-Omicron era of COVID-19 management

1. Introduction

Since the emergence of SARS-CoV-2 in late 2019, the world has experienced multiple epidemic waves of COVID-19 disease and diverse evolutionary variants of SARS-CoV-2.¹ In parallel to the evolution of the virus, a range of COVID-19 vaccines have been developed,² and there have been multiple rounds of vaccination across the world.

Both prior infection and vaccination can reduce the chance of future infection and severity of outcomes, combining to form “hybrid immunity”³ against COVID-19. In the post-Omicron era, most populations have high levels of past infection across multiple epidemic waves, creating exposure-derived ‘natural’ immunity. Vaccine coverage has been variable due to inequities of access, eligibility and uptake, with consequences for hybrid immunity landscapes. ^4;5;6 Unfortunately, all forms of immunity wane over time, enabling possible reinfection within a matter of months, though protection against severe outcomes is longer lived. ^6;7

How can COVID-19 vaccines incorporated into routine immunisation schedules help minimise the impact of recurring epidemic waves and promote resilience against future variants? Heterogeneous population experiences of infection and vaccination, and the irregular emergence of immune escape variants, makes anticipating the timing, magnitude and clinical burden of future epidemics challenging. The World Health Organization (WHO) recently reviewed evidence for the impact and cost-effectiveness of COVID-19 booster vaccine strategies. Key questions to inform strategic guidelines included: the incremental benefits of boosters, identification of optimal vaccine target groups in high seroprevalence settings, ongoing vaccine-preventable disease burden, optimum boosting strategy including frequency for priority populations, and the cost effectiveness of those vaccination strategies. We need flexible frameworks to investigate this multi-dimensional problem space.

As one of several groups commissioned by WHO to support decision making, we adapted an existing model representing diverse population and hybrid immunity landscapes⁵ to address questions relevant to future COVID-19 vaccine prioritisation. ⁸ We mimicked realistic epidemic exposure histories and specified timings for the emergence of immune escape variants. We also proposed plausible vaccine rollout and coverage assumptions. We developed exemplar demographies and costings based on high and low-middle income countries in the Western Pacific Region. Our findings played a role in informing the WHO updated COVID-19 vaccination guidance for March 2023. ⁹

2. Methods

Our modelling pipeline is depicted in Figure 1. First, the immunological model informs an infection transmission/dynamics model within a mechanical agent-based model. The outputs of the agent-based model are input to a clinical pathways model to obtain clinical outcomes. These clinical outcomes then link to a cost-effectiveness model evaluating alternative vaccination strategies. Based on the problem space, we configure our model using numerous parameters, including population distribution, vaccine program and health systems costs for exemplar country contexts.

• Download figure
• Open in new tab

Figure 1:

Simulation and analysis model framework, from immunological model, to cost-effectiveness analysis. The immunological model converts vaccinations and infections into neutralising antibody parameters, which influence outbreak dynamics and clinical outcomes. The simulation model has two parts: first an agent-based model of population level transmission, with multiple primary inputs including vaccination and demography, followed by a clinical pathways model, which generates outcomes using the timeseries of symptomatic infections, neutralising antibodies and age-dependent clinical progression probabilities. The resultant clinical outcomes are used in the cost-effectiveness analysis, along with other input parameters such as vaccine and disease management costs.

Our methods extend previous work⁵ with adjusted scenarios to answer policy-relevant questions, and include cost-effective analyses. Full model details can be found in the appendix and references.^5;10

3. Results

3.1. High vaccination coverage scenarios

3.1.1. Comparing target use groups

Figure 2 compares the impacts of alternative vaccine allocations in older and younger populations depending on the time of immune escape emergence (1.5 vs 2.5 years), given past (1.5 years) high seroprevalence. We find limited impact of different strategies on infection dynamics. In the late (2.5 years) immune escape scenarios (panels (c) and (d)), the epidemic peaks are shifted to the right, but overall the infection curves maintain the same qualitative shape.

• Download figure
• Open in new tab

Figure 2:

Outbreaks in high transmission settings with high vaccination coverage, for older and younger demographics with early and late seeding of an immune evading variant (dark grey shading). (a) older population with early immune escape (1.5 years); (b) younger population with early immune escape (1.5 years); (c) older population with late immune escape (2.5 years); (d) younger population with late immune escape (2.5 years). Scenarios (a)-(d) are run with strategies of no further boosting, pediatric boosting (ages 5-15), high-risk boosting (65+ in the older population, 55+ in the younger population) and random boosting at 2 years. Scenarios are presented with lines representing pointwise medians from 1000 simulations and shaded regions representing the minimum and maximum from the simulations. The medium grey and dark grey background define the currently circulating variant, Omicron BA.1-like and BA.4/5-like respectively. The impact of boosting on infections is limited.

We find that vaccination has greater impact on severe outcomes. Across the scenarios, high-risk boosting averts the most severe disease (Table 2).

View this table:

• View inline
• View popup
• Download powerpoint

Table 2

Median deaths between 1.5 - 3 years in high transmission settings with high vaccination coverage from 5000 simulations (5 clinical pathway simulations are produced per infection transmission simulation), with 0.025 and 0.975 quantiles. Early immune escape occurs at 1.5 years; late immune escape occurs at 2.5 years. Each scenario is run with four different boosting strategies at 2 years.

In Figure 3(a), we find that in older populations, boosting is more cost-effective when it occurs prior to immune escape. High-risk boosting is likely to be highly cost-effective or cost-saving, while pediatric boosting does not appear to be cost-effective. Random boosting does worse than high-risk boosting. Cost-effectiveness of high-risk boosting is driven primarily by vaccine program (delivery and dose) costs, followed by disease management costs in general ward (see Fig. E5 in the appendix).

• Download figure
• Open in new tab

Figure 3:

Cost-effectiveness results for different vaccination strategies in high transmission high vaccination coverage settings, for older (Group A) and younger (Group B) demographics with early (1.50 year) and late (2.50 year) seeding of an immune evading variant. Top figures represent cost-effectiveness planes for (a) older population and (b) younger population. Bottom figures show cost-effectiveness acceptability curves considering stochastic uncertainty and economic parameter uncertainty for the most cost-effective high-risk strategies for (c) older population and (d) younger population. The boosting strategies considered here are: pediatric boosting (ages 5-15), high-risk boosting (65+ in the older population, 55+ in the younger population) and random boosting at 2 years. High risk boosting is the most cost-effective strategy and dominates pediatric and random boosting strategies.

High risk boosting may be cost-effective or even cost-saving in younger MIC populations, depending on country-level willingness to pay (WTP) per DALY averted threshold, or CET, and other key model inputs, which we see in Figure 3(right side). These results are driven primarily by home-based care cost inputs, and vaccine delivery costs, which remain highly uncertain in these settings (Fig. E6 in the appendix). In sensitivity analyses investigating no home-based care costs in these settings, we find elder boosting strategies may remain cost-effective when vaccine is donated and for countries with higher CETs (Figure E7 in the appendix).

The relative benefits of boosting, especially high-risk boosting, hold across all the scenarios considered in Figures 2, 3 and Table 2. However, the absolute benefits are greater in the older population.

Note that we assume high transmission potential, which leads to high seroprevalence. The relative impact of vaccination in the low seroprevalence scenario is similar (appendix Section E.1).

3.1.2. Boosting frequency

The timing and frequency of boosting, relative to emergence of the immune escape variant, influenced the impact of vaccination on the number of infections (Figure 4). Half-yearly boosting consistently achieves fewer severe outcomes when compared to boosting once (see appendix Table E3).

• Download figure
• Open in new tab

Figure 4:

Outbreaks in high transmission settings with high vaccination coverage, for older and younger demographics, comparing boosting once at a range of times and frequent boosting. (a) older population with early immune escape (1.5 years); (b) younger population with early immune escape (1.5 years); (c) older population with late immune escape (2.5 years); (d) younger population with late immune escape (2.5 years). Scenarios (a)-(d) are run with high-risk boosting (65+ in the older population, 55+ in the younger population) rolled out at either 1.75 years, 2.0 years, 2.25 years, 2.5 years, or half-yearly starting from 1.75 years. Scenarios are presented with lines representing pointwise medians from 1000 simulations and shaded regions representing the minimum and maximum from the simulations. The timing and frequency of boosting, relative to emergence of the immune escape variant, influences the impact of vaccination on the subsequent outbreaks.

The cost-effectiveness of high-risk boosting varies according to immune escape timing, but generally is very cost-effective or cost-saving in the older (HIC) population and in younger (MIC) populations. Half-yearly boosting remains highly cost-effective in older populations, but more expensive than boosting only once (see appendix Figure E8).

However, half-yearly boosting is unlikely to be cost-effective for younger (MIC) countries with high vaccine coverage (∼ 80%) unless vaccines are donated. These results were driven by home-based care costs (lower costs indicate half-yearly boosting is unlikely to be cost-effective) and vaccine program costs (lower costs, for example, donated vaccine, would mean half-yearly boosting may be very cost-effective or cost-saving) (see appendix Figure E9).

3.1.3. Age cut-off for cost-effective boosting

We systematically expanded booster eligibility to younger age groups (Figure 5). The difference in health outcomes between boosting 45+ and younger age groups is minimal (see appendix Table E4), but increasing the age cohorts in the program leads to higher costs and thus lower cost effectiveness. Boosting 65+ and 55+ is likely highly cost-effective or cost-saving. Boosting 45+ is also likely highly cost-effective.

• Download figure
• Open in new tab

Figure 5:

Outbreaks and cost-effectiveness analyses in the high transmission high vaccination coverage setting, for older and younger demographics, comparing the impact of lowering the age cut-off for high risk boosting. (a) epidemic waves in the older population; (b) epidemic waves in the younger population; (c) cost-effectiveness analysis in the older population; d cost-effectiveness analysis in the younger population. All scenarios here had an immune escape variant seeded at 2 years, with boosting at 2 years. The solid lines in (a) and (b) represent the pointwise median infections from 1000 simulations and the shaded regions represent the pointwise maximum and minimum infections. Boosting 65+ and 55+ is likely to be cost-effective or cost-saving.

3.2. Low-medium vaccination coverage scenarios

3.2.1. Comparing primary and booster strategies

We consider the trade-off between new primary vaccination and high risk boosting strategies. We find that high risk boosting strategies perform better, reducing the size of epidemic peaks (Figure 6 (a), (b)).

• Download figure
• Open in new tab

Figure 6:

Outbreaks and cost-effectiveness analyses in the high transmission setting, with low and medium vaccination coverage for younger demographics. (a) epidemic curves for younger population with low vaccination coverage (initial coverage 20%); (b) for medium coverage (initial coverage 50%); (c) cost-effective analyses for low and medium vaccination coverage scenarios; (d) cost-effectiveness acceptability curves considering stochastic uncertainty and economic parameter uncertainty. All scenarios had an immune escape variant seeded at 2 years. These scenarios are run with vaccination strategies of pediatric vaccination (ages 5-15), high risk boosting (65+ first), and new primary vaccinations (allocated randomly among ages 5+) at 2 years. The solid lines in (a) and (b) represent the pointwise median infections from 1000 simulations and the shaded regions represent the pointwise maximum and minimum infections. Pediatric and new primary vaccinations are unlikely to be cost-effective.

New primary pediatric vaccination and new primary vaccination strategies are unlikely to be cost-effective (Figure 6(c)). High-risk boosting strategies may be cost-effective for younger (MIC) countries with high WTP thresholds or CETs (Figure 6(d)), but this depends on unit cost inputs, which are driven primarily by home-based care and vaccine program costs.

3.2.2. Impact of bivalent boosting

We found a modest benefit in bivalent boosters over monovalent boosters on the dynamics of infections, with very minor differences in infection peak height (Figure E15 in the appendix) and also minor differences in the number of deaths between vaccine types(Table E6 in the appendix).

Given that the modelled benefits of bivalent boosters is minor compared with monovalent formulations, we anticipate that bivalent vaccines in this context will not be cost-effective unless willingness to pay is high. As such, we have not subjected these scenarios to formal cost-effective analyses.

4. Discussion

Our modelling approach allowed the exploration of COVID-19 booster dose impacts in diverse population settings. We assumed differing levels of prior vaccination delivery and infection experience, linked to income group level characteristics and vaccine program and healthcare costs. Given the age-dependency of severe disease risk, we concluded that elder-targeted strategies are most likely to be cost-effective (or even cost-saving) across a broad range of uncertainties. Notably, we consistently found pediatric programs (primary series or boosting) are not cost-effective. Absolute harms averted by vaccination are influenced by: age and risk profile of the population, prior immune landscape (infection exposure history, vaccination rollout), and timing of emergence of an immune escape variant in relation to booster delivery. Half-yearly ‘high risk’ booster programs are more expensive but may be cost effective in older, high income populations. However, this finding is much more uncertain in populations with younger demographics (representing upper- and lower-MIC), depending on the costs of community-based care and vaccine implementation.

A strength of our study was the modelling flexibility, enabling distinct configuration of the various interweaving elements. We used the immunity model to implement assumptions relevant to two exemplar Omicron variants, against which the effectiveness of ancestral and bivalent vaccines were explored. By introducing variants at different time points, we systematically evaluated the importance of epidemic timing in relation to plausible immune escape scenarios. Separate representation of clinical pathways and cost-effectiveness analysis further enabled adaptation to different settings. Health sector costs were exemplified by ‘averages’ based on regional data, with sensitivity analyses highlighting local drivers that will be influential for decision making depending on context.

As with all models, multiple simplifying assumptions were made for the purpose of tractability that do not necessarily reflect reality. We assumed circulation of a single dominant SARS-CoV-2 variant, to interrogate the ‘worst case’ scenario of a step change in vaccine effectiveness at a single point in time. The present state of multiple lineage co-circulation with variable immunological cross-reactivity and breadth likely lessens such impacts.³⁰ We could have explored more optimistic values for bivalent vaccine effectiveness, which may lead to greater estimated benefits.³¹ Neither social restrictions nor antiviral agents were included, given that neither are presently being widely applied at the population level globally.

Long COVID was not included as a potential outcome of infection, as there remains limited quantitative data of this clinical burden. The anticipated costs of therapeutic pathways are yet to be determined pending identification of those which are most likely to be effective for various syndromic presentations.

A more significant limitation of our work in relation to health system costs is the assumption that the modelled adult-targeted vaccination coverage is achievable and can be costed across all country settings considered. In reality, most LMICs do not have adult immunisation programs in place, as highlighted by the COVID-19 pandemic.This deficiency is a global public health priority. We also do not consider indirect costs to society, such as productivity losses, associated with illness or death. A societal perspective would increase the cost-effectiveness of elder-targeted vaccination.

Our study considers a diversity of demographics and hybrid immunity histories, with lessons offered for countries that may fall between the “older” and ‘younger” exemplar groupings. We note that all our considered populations had some level of background vaccination (22%) by the end of the first Omicron wave; our work complements other studies^32;33 that have considered the zero past vaccination setting with prior natural immunity in low and middle income countries.

As the world transitions towards COVID-19 endemicity, our study demonstrates the ongoing value of COVID-19 booster doses targeted towards older age groups at risk of severe outcomes in a range of demographics with different hybrid immunity histories. This approach is cost-effective or even cost-saving across multiple country settings. However, this assumes that adult immunisation programs are in place and do not impact on delivery of other services. Pediatric vaccination may be more readily implementable within existing health systems but was not cost-effective in any of the scenarios explored. These results were presented to the Advisory Committee on Immunization and Vaccines-related Implementation Research (IVIR-AC),⁸ with our work being subsequently cited as part of the WHO updated COVID-19 vaccination guidance for March 2023.⁹

SARS-CoV-2 reporting is being subsumed into broader respiratory pathogen surveillance systems. The WHO continues to encourage vigilance for identification of new variants with heightened transmissibility or pathogenic potential. A downward age shift in disease severity would require revision of current strategies. Our framework is sufficiently flexible to incorporate emerging evidence of virus and vaccine characteristics and can be configured to specific country settings. More broadly, it provides a template for the use of modelling to evaluate strategies for the control of any emerging or epidemic infectious disease.

5. Data sharing

Code is available in the following GitHub repositories: https://github.com/goldingn/neuts2efficacy/ for the immunological model; https://github.com/spectrum-spark/covid_singlestrain_scenarios/tree/singlestrain-paper for the population transmission and clinical pathway models; and https://github.com/spectrum-spark/covid-CEA/ for the cost-effectiveness analysis.

CRediT authorship contribution statement

Thao P. Le: Data curation, Formal Analysis, Investigation, Visualization, Writing – original draft, Writing – review & editing. Eamon Conway: Data curation, Methodology, Software, Formal Analysis, Writing – original draft, Writing – review & editing. Edifofon Akpan: Data curation, Formal Analysis, Investigation, Visualization, Writing – review & editing. Isobel Abell: Methodology, Writing – original draft, Writing – review & editing. Patrick Abraham: Data curation, Formal Analysis, Investigation, Visualization, Writing – original draft, Writing – review & editing. Christopher M. Baker: Conceptualization, Methodology, Supervision, Writing – original draft, Writing – review & editing. Patricia T. Campbell: Formal Analysis, Writing – original draft, Writing – review & editing. Deborah Cromer: Data curation, Methodology, Writing – original draft, Writing – review & editing. Michael J. Lydeamore: Methodology, Resources, Writing – original draft, Writing – review & editing. Yasmine McDonough: Validation, Writing – original draft, Writing – review & editing. Ivo Mueller: Conceptualization, Supervision, Writing – review & editing. Gerard Ryan: Data curation, Methodology, Writing – original draft, Writing – review & editing. Camelia Walker: Data curation, Methodology, Software, Writing – original draft, Writing – review & editing. Yingying Wang: Data curation, Formal Analysis, Investigation, Visualization, Writing – review & editing. Natalie Carvalho: Formal Analysis, Methodology, Supervision, Writing – review & editing. Jodie McVernon: Conceptualization, Funding acquisition, Supervision, Writing – original draft, Writing – review & editing.