Equitable access to COVID-19 vaccines makes a life-saving difference to all countries

The study addresses how global COVID-19 vaccine allocation—particularly inequities between high-income countries and low- and middle-income countries—shapes pandemic trajectories, including emergence and spread of new variants. Motivated by the insufficiency of non-pharmaceutical interventions alone and the globalized risk of reintroduction of cases, the authors note stark disparities in vaccination coverage (over 70% fully vaccinated in HICs vs 4% in low-income countries as of 31 December 2021). They highlight concerns raised by prior work about the public health and economic costs of inequitable allocation and the risk that prioritizing domestic booster programs in HICs over donations to LMICs may accelerate emergence of variants (for example, Delta and Omicron). The objective is to provide a data-driven, global analysis of vaccine equity versus inequity, incorporating viral evolution and human mobility, and to identify practical donation strategies that protect both HICs and LMICs, reframing vaccine distribution as a cooperative rather than zero-sum game.

The paper situates itself within warnings from public health and policy literature that pandemic control requires equitable access to vaccines and that the costs of inequity are ultimately shared globally. Evidence from influenza suggests cross-border vaccination subsidies can protect both donors and recipients. Game-theoretic models have indicated that coordinated, optimal vaccine or drug allocation across countries can minimize epidemic size and economic costs, though such analyses often rely on hypothetical networks. Despite mechanisms like the WHO-backed COVAX Facility, global inequity persists due to HICs’ bilateral purchasing and underestimation of variant threats, with many HICs expanding booster programs before donating doses. The authors aim to fill a gap in data-driven modeling of global vaccine coordination using real-world mobility networks in the COVID-19 context with viral mutation.

The authors develop an integrated, data-driven multistrain SVEIRD metapopulation model on a global mobility network. Model structure and viral evolution: - Multistrain model with M strains in a linear strain space and stepwise one-direction mutation from strain m to m+1 with mutation probability μ_m per infection; most mutations are neutral, but some increase transmissibility and severity. - Country-specific transmissibility (T_i) and severity (F_i) diagonal matrices capture heterogeneity in transmission and infection fatality rate (e.g., due to age structure and healthcare burden). Transmissibility increases with strain index (T_{m+1} = (1+θ) T_m); severity correlates positively with transmissibility. - Probability of emergence of new, more dangerous strains per infection declines as the virus evolves due to finite genome constraints, parameterized via a decrease rate λ. - Cross-immunity from vaccines declines with antigenic distance between vaccine strain (assumed strain 1) and mutant strains; vaccine efficacy against infection and death for strain n decays with distance from strain 1. SVEIRD compartments (per country i and strain m): - Susceptible (S_i), Vaccinated (V_i), Exposed without vaccinal immunity (E^S_{i,m}), Exposed with vaccinal immunity (E^V_{i,m}), Infectious without vaccinal immunity (I^S_{i,m}), Infectious with vaccinal immunity (I^V_{i,m}), Recovered (R_i), and Deceased (D_i). - Vaccination moves S_i to V_i at rate φ_i(t) determined by global allocation; vaccinated individuals lose vaccinal immunity at rate ε and return to S_i. - Exposed become infectious at rate σ (incubation period 1/σ); infectious individuals recover or die at rate α (infectious period 1/α) with transitions modulated by severity F_i,m and vaccine protection. - Co-infection is disallowed; recovery confers immunity to all strains in the model. Non-pharmaceutical interventions (NPIs): - Contacts are modulated by an adaptive, reproduction-number-based policy: more stringent NPIs activate when the local effective reproduction number exceeds a threshold and relax when below it, independently per country. Mobility network: - Global metapopulation built from real-world air traffic (Official Aviation Guide) defining an undirected, dynamic mobility network. The number of travelers G_{ij}(t) depends on average mobility rate, available travelers in origin, and fixed passenger fractions derived from 2020 seat data. Infectious and deceased individuals do not travel. Global vaccine supply and allocation: - Cumulative supply q(t) grows exponentially until time τ (maximum daily production capacity), then increases linearly at that capacity: q(t) piecewise by a ramp parameter ν. Daily supply is q(t+1)−q(t). - Vaccines are assumed two-dose, co-administered simultaneously, with immediate full vaccinal immunity; upper bounds on daily vaccination rates per country group (HIC vs LMIC) are set to the maximum observed rates from 1 Dec 2020 to 15 Jun 2021. - Two allocation regimes: • Equitable: available doses at each time allocated across all countries based on dynamic prioritization criteria. • Inequitable: a minimum fraction γ of doses is purchased by HICs; the remainder goes to LMICs; within each group, allocation follows prioritization. - Prioritization criteria (updated dynamically): population size, prevalence (active cases per capita), mortality rate (new deaths per capita over past two weeks), or incidence (new cases per capita over past two weeks). - Country classification: HICs include World Bank high-income economies plus China and Russia (due to vaccine production capacity); remaining countries are LMICs. Donation strategies (allow-donation): - An HIC donates a portion δ of its vaccine supply to an international facility (e.g., COVAX) when its local prevalence is below a threshold I_thre; donated doses are equitably allocated to LMICs per prioritization criteria. - The study also examines donating to k-hop neighbors (k = 1,2,3,4) in the air travel network versus donating to all LMICs. Initialization and data: - t = 0 is 15 June 2021; all active cases initially from strain 1; E, I, R, D for strains m ≠ 1 are 0. V_i(0) from fully vaccinated counts; I_i(0) from active cases; R_i(0) and D_i(0) from cumulative recovered and deaths; S_i(0) computed by population balance. - Probabilistic bias analysis corrects under-reporting of infections, recoveries, deaths, and active cases. - Data sources: Epidemiology (JHU CSSE), vaccination (Our World in Data), income (World Bank), population (UN WPP), mobility (OAG, licensed). Code and processed data available on Zenodo. Scenarios and parameters: - Example parameters: M = 5, μ_1 = 5.6×10^-3, θ = 0.2, λ = 5×10^-2; scenarios also include highly transmissible strains (e.g., M = 6, θ = 0.26). Supply assumptions follow Airfinity predictions: production reaches capacity in ~6 months and suffices to fully vaccinate half the world in six months.

- Vaccine inequity provides only limited, short-term benefits to HICs: it accelerates initial incidence decline in HICs during the first year but prolongs the pandemic globally and increases infections and deaths in LMICs. Subsequent waves arise in HICs due to variant emergence in LMICs under inequitable allocation. - Sharper disparities (larger HIC share γ) yield earlier and larger future waves, little additional mortality reduction in HICs, and noticeably higher mortality in LMICs. If an extremely transmissible strain emerges, vaccine inequity provides no benefit to HICs. - Prioritization criteria: Under equitable allocation, prioritizing by population slightly increases prevalence and mortality versus prioritizing by prevalence, incidence, or mortality; under inequity, population-priority leads to earlier new waves. In rare scenarios with extremely transmissible variants, prioritizing densely populated countries can reduce long-run prevalence; otherwise, prioritize countries with higher incidence/prevalence/mortality to contain spread. - Variant emergence: Inequitable allocation accelerates the rise and peak timing of successive, more transmissible strains; LMICs have a higher fraction of cases from newer strains than HICs. Equitable allocation substantially curbs the spread of new strains globally. - Donation strategies (allow-donation): Almost all LMICs benefit from HIC donations irrespective of timing; reductions in LMIC cumulative mortality are more sensitive to the quantity donated than the start time. Benefits become pronounced when donated portion δ reaches about 46%. For HICs, donating more yields higher benefits up to around 80% donation share. A strategy donating a moderate share (e.g., δ = 0.5) when local prevalence is relatively low (I_thre = 5×10^-5) performs best among tested donation policies and ends the pandemic earlier (second only to fully equitable allocation). Donating a small portion even when local prevalence is still high can mitigate future waves more than waiting for full domestic control. - Donating to neighbors only: Limiting donations to immediate (1-hop) neighbors leads to higher cumulative mortality in LMICs and does not materially improve HIC outcomes compared with broader donation. Donating to more LMICs (2–4 hops, effectively all LMICs) leads to earlier pandemic end with little difference among 2–4 hop scenarios. - Contextual data: As of 31 Dec 2021, >9 billion doses administered worldwide (~116 doses/100 people), >70% fully vaccinated in HICs vs 4% in low-income countries; 31 vaccines approved by at least one country and 137 in clinical development.

The findings demonstrate that vaccine hoarding by HICs yields only transient domestic advantages while elevating global and domestic risks through accelerated variant emergence in under-vaccinated LMICs. Incorporating global mobility and mutation dynamics reveals that inequity extends pandemic duration and heightens the probability and size of future waves in all regions. Equitable allocation and timely donations suppress the emergence and spread of new strains, reduce global morbidity and mortality, and can better protect HICs than exclusive domestic vaccination, especially when donations reach meaningful shares and are not restricted to immediate neighbors. Prioritizing vaccines to countries with higher incidence, prevalence, or mortality generally performs better than population-based allocation, except in rare circumstances with extremely transmissible variants where pre-emptive protection of large populations may be advantageous. Overall, strategic, cooperative vaccine sharing aligns with rational self-interest for HICs and advances global control.

This work provides a data-driven global analysis integrating viral evolution and human mobility to quantify the impacts of vaccine equity versus inequity. It shows that inequitable allocation affords only short-lived benefits to HICs and increases risks and deaths in LMICs, while equitable allocation and donation strategies substantially curb variant emergence and accelerate pandemic resolution. Practical policy guidance emerges: HICs should donate vaccines generously and early (even before full domestic coverage), target broader LMIC coverage rather than only neighbors, and prioritize countries with high incidence/prevalence/mortality in most scenarios. Future research should refine the model by incorporating age-stratified contact and susceptibility structures, more realistic and strain-specific natural immunity durations, and richer evolutionary dynamics in a multi-dimensional antigenic space to capture selection pressures that may favor reduced severity.

- No within-country age stratification: although country-specific severity matrices reflect heterogeneous age structures and healthcare burdens, the model does not explicitly stratify transmission and susceptibility by age due to limited data on age-specific susceptibility and mixing patterns. - Natural immunity assumptions: analyses consider lifelong and short-lived natural immunity settings; reinfections were uncommon at the time. Sensitivity analyses suggest that short-lived immunity further reduces any benefits of inequity to HICs. More realistic, strain-specific immunity durations should be incorporated as data become available. - Simplified evolution: the multistrain model assumes a linear strain space with monotonically increasing severity alongside transmissibility; in reality, selection may act against severe strains. Extending to a multi-dimensional antigenic space with selection dynamics is an important future direction.