An analysis of the potential global impact of dosing regimen and rollout options for the ChAdOx1 nCoV-19 vaccine


 Background The ongoing COVID-19 pandemic has placed an unprecedented health and economic burden on countries at all levels of socioeconomic development, emphasizing the need to evaluate the most effective vaccination strategy in multiple, diverse environments. The high reported efficacy, low cost, and long shelf-life of the ChAdOx1 nCoV-19 vaccine positions it well for evaluation in different settings. Methods Using data from the ongoing ChAdOx1 nCoV-19 clinical trials, an individual-based model was constructed to predict the 6-month population-level impact of vaccine deployment. A detailed probabilistic sensitivity analysis (PSA) was developed to evaluate the importance of epidemiological, demographic, immunological, and logistical factors in determining vaccine effectiveness. Using representative countries, logistical plans for vaccination rollout at various levels of vaccine availability and delivery speed, conditional on vaccine efficacy profiles (efficacy of the booster dose, time interval between doses, and relative efficacy of the first dose) were explored.Findings and Interpretation Our results highlight how expedient vaccine delivery to high-risk groups is critical in mitigating COVID-19 disease and mortality. In scenarios where the number of vaccine doses available is insufficient for high-risk groups (those aged more than 65 years) to receive two vaccine doses, administration of a single dose of vaccine is optimal. This effect is consistent even when vaccine efficacy after one dose is just 75% of the levels achieved after two doses. These findings offer a nuanced perspective of the critical drivers of COVID-19 vaccination effectiveness and can inform optimal allocation strategies. These are relevant to high-income countries with a large high-risk group population as well as to low-income countries with younger populations, where the cost and logistical challenges of procuring and delivering two doses for each citizen represent a significant challenge.


Summary Background
The ongoing COVID-19 pandemic has placed an unprecedented health and economic burden on countries at all levels of socioeconomic development, emphasizing the need to evaluate the most effective vaccination strategy in multiple, diverse environments. The high reported efficacy, low cost, and long shelf-life of the ChAdOx1 nCoV-19 vaccine positions it well for evaluation in different settings.

Methods
Using data from the ongoing ChAdOx1 nCoV-19 clinical trials, an individual-based model was constructed to predict the 6-month population-level impact of vaccine deployment. A detailed probabilistic sensitivity analysis (PSA) was developed to evaluate the importance of epidemiological, demographic, immunological, and logistical factors in determining vaccine effectiveness. Using representative countries, logistical plans for vaccination rollout at various levels of vaccine availability and delivery speed, conditional on vaccine efficacy profiles (efficacy of the booster dose, time interval between doses, and relative efficacy of the first dose) were explored.

Findings and Interpretation
Our results highlight how expedient vaccine delivery to high-risk groups is critical in mitigating COVID-19 disease and mortality. In scenarios where the number of vaccine doses available is insufficient for high-risk groups (those aged more than 65 years) to receive two vaccine doses, administration of a single dose of vaccine is optimal. This effect is consistent even when vaccine efficacy after one dose is just 75% of the levels achieved after two doses.
These findings offer a nuanced perspective of the critical drivers of COVID-19 vaccination effectiveness and can inform optimal allocation strategies. These are relevant to highincome countries with a large high-risk group population as well as to low-income countries with younger populations, where the cost and logistical challenges of procuring and delivering two doses for each citizen represent a significant challenge.

Introduction
As of March 2 nd 2021, almost 115 million people have been diagnosed with  worldwide, and in excess of 2.5 million confirmed deaths have been reported 2,3 . Vaccination is a critical strategy to control the spread of SARS-CoV-2, the virus that causes COVID -19, and to reduce the severity of symptomatic disease. Three vaccines have already received emergency use authorization in the United Kingdom (UK). The developers of two of these vaccines have reported efficacies of 95% for their vaccines in their respective Phase 3 trials (Pfizer/BioNTech and Moderna) 4 . The third vaccine, ChAdOx1 nCoV-19, jointly developed by Oxford University and AstraZeneca, demonstrated an acceptable safety profile and efficacy against symptomatic COVID-19, with no hospital admissions or severe cases of disease reported in the intervention arm during Phase 3 trials conducted in three countries. This vaccine can be stored and distributed at 2-8°C and will be made available at a lower cost than the other vaccines, making it suitable for global access, particularly in low-and middleincome countries (LMICs) [5][6][7] .
While clinical trials have validated the efficacy of the ChAdOx1 nCoV-19 vaccine in reducing symptomatic infection, appropriate national vaccination strategies across the world must consider heterogeneity among populations as well as the diverse demographic and socioeconomic environments of affected countries. In particular, the younger population typically present in LMICs justifies the need to assess the effects of associated behaviours and health profiles on vaccine effectiveness. These countries exhibit competing health, social, and economic challenges owing to inadequate healthcare infrastructure and a high prevalence of immunocompromising and infectious diseases. In these settings, individuals could also suffer complex vaccine responses when compared with responses in individuals in more developed economies 8,9 . At the same time, many LMICs have been unable to secure vaccine doses in advance from potential suppliers and thus are likely to have incomplete coverage of their populations, particularly in the short-term. The global COVID-19 vaccine alliance, COVAX, has pledged to procure and distribute vaccines equitably to LMICs; however, this will cover a maximum of 20% of the total population in each country 10 .
Although the University of Oxford and AstraZeneca have made the largest supply commitment to LMICs at more affordable prices than other vaccine manufacturers, there is a need to evaluate the impact of a range of factors on vaccine effectiveness 11-13 . Due to shortages in supply, the UK government has instituted a policy of administering the booster dose of the vaccine at up to 12 weeks following the initial dose, prompting a debate among scientists, manufacturers, and governments on optimal dosing intervals for COVID-19 vaccines 14,15 . The purpose of this analysis is therefore to evaluate the efficacy of the ChAdOx1 nCoV-19 vaccine in countries with different demographic profiles, as a function of vaccine efficacy, dosage regime (interval between initial and booster doses, or no booster at all), coverage, and immunity wane rate. Given the differences in healthcare infrastructure and vaccine access around the world, decision-makers should consider the effect of these factors on population-level impact to determine the most effective strategy for their context 13 .
Where vaccination programmes have begun, priority has so far been given to older age groups, individuals with co-morbidities, and frontline medical staff. The model developed therefore considers a simplified system where the vaccine is delivered to age groups in descending order while supplies are available. As there is limited evidence of indirect effects, that is, the potential for reductions in transmission, this vaccine effect was assumed to be negligible for the purposes of this analysis.

Methods
The methodology employed was very specifically tailored to the research question and its context. Vaccine production rates are always going to be insufficient to meet the demand generated by a global pandemic. In a context of limited vaccine dose availability, it is imperative to prioritize those individuals who would yield the greatest epidemiological benefit. Assuming the most pressing need is to reduce hospitalization rates and deaths, the initial targeting of those at higher risk for these outcomes seems logical, given that the alternative of immunizing sufficient people at lower risk for the indirect benefits to outweigh the direct benefits of a vaccine targeted at those at higher risk is not feasible with the number of vaccines available in the short-term. Even the UK, where mass production of the AztraZeneca vaccine has enabled 20% of the population to be vaccinated within 3 months, opted to prioritize the high-risk groups (those aged more than 65 years), partially because of the uncertainty around vaccine efficacy against infection. The ChAdOx1 nCoV-19 vaccine 6 clinical trials were the only Phase 3 trials in which infection was evaluated as an outcome. No evidence was found for a transmission reduction effect (VE = 3.8% [−72.4 to 46.3]) 7 , but important questions were raised about how to allocate a limited number of doses to optimize the impact on symptomatic disease, given that a single-dose regimen could offer prolonged protection and thus a delay of the second dose could be warranted.
We began from the premise laid out above and implemented an individual-based, agedependent, static transmission model to predict the number of infections, clinical cases, and deaths expected to occur within 6 months of vaccination programme rollout. Individuals are simulated as autonomous systems, each with a set of attributes, informing their serostatus, vaccination uptake history (number of doses and dosing interval), and age. Box 1 details how the dynamics processes inherent to disease transmission and vaccination campaign logistics are considered in the model.
Thus, vaccines targeting high-contact groups would have to provide indirect effects in the order of 80% (80% reduction in risk in the untargeted population) to prevent an approximate number of deaths similar to that which would be provided by targeting the high-risk group with a direct vaccine effect against death of 85%.

Transmission and clinical cascade
The spread of COVID-19 is sensitive to the underlying network of contacts between infectious and susceptible individuals in their various societal spheres (home, work, public transport, etc). For a given population, we can summarize the number of contacts per day as an agedependent force of infection ( ), i.e. a daily risk of acquiring an infection given age a. The age-dependent risk of infection can then be defined as: Where is the daily number of contacts between age groups i and j for a particular country, is the population age distribution, is the total number of age categories, and is the overall daily risk of infection (which is informed by the number of infectious people in the population We implemented three different logistical implementations of a vaccine campaign rollout: constant effort, frontloaded, and backloaded. The distinction was in the speed at which the target population received vaccine doses during the initial 2 months. As individuals were assigned a vaccine, the number of doses received would be determined by a draw from a uniform distribution according to the desired dose split. Individuals given two doses would be assigned a booster dose interval following a beta distribution with  = 0.15 and  = 0.95. Although vaccine efficacy was explored in the sensitivity analyses presented here, we centred the explored ranges around the point estimates presented in 17 . Vaccine efficacy, ( ), was treated as a direct modulator of the risk of infection, clinical disease, and death; it was then defined for each individual, at each timestep of the simulation, as: , where V is the vaccine efficacy in an individual with baseline status i that received dose number j a t number of days ago, while is the rate of loss of vaccine-induced immunity.
Throughout this paper, we present a sensitivity analysis of the post-dose two maximum efficacy, the relative efficacy of dose one vs dose two, and the booster dose interval. While doing so, we constrain vaccine efficacy against clinical disease to be the same as that against death, while vaccine efficacy against infection is fixed at 5% 7 . We also imposed a stepwise increase in post-dose two vaccine efficacy across an 8-week booster dose interval, as observed in the clinical trial 17 . This means that giving the second vaccine dose less than 8 weeks after the first dose will result in a 25% lower post-dose two efficacy relative to the maximum assumed vaccine efficacy.

Vaccine effectiveness
The vaccination campaign population impact is referred to throughout as vaccine effectiveness and was defined as:

Results
We conducted an extensive initial sensitivity analysis to determine how the impact of rolling having a larger parameter space in which a single-dose option is preferable, as shown in Figure   3.

Discussion
The SARS-CoV-2/COVID-19 pandemic has created an unprecedented public health challenge, spurring a global race to develop and distribute viable vaccines. A vaccine that creates broad immunity against the SARS-CoV-2 virus could be the only effective means to control the pandemic and allow a return to "normalcy". To have a significant impact on the disease, a critical mass of the global population at risk will need to be vaccinated. However, many high-income countries have secured more than half of the available vaccine doses for themselves, leaving LMICs, which make up more than 85% of the global population, to find their own solutions 18  inform optimal allocation programmes, particularly in contexts where the cost and logistics of implementing multiple doses within a short timeframe may be challenging.
Our findings indicate that vaccine effectiveness is dependent upon (i) the country context, which includes the demographic profile, the attack rate of the virus, and the amount of vaccine that is available (which influences the proportion of the population that is for this vaccine 17 . However, this finding may not be applicable to other COVID-19 vaccines. These differences are more profound when considering the demographic characteristics of a population. In high-income countries, which have a larger older population (>65 years), a single-dose regimen will allow the vaccination of more individuals more quickly, with a correspondingly greater impact on cases and deaths. In the UK, the six-month allocation threshold above which a two-dose regimen would be preferred was found to be about 16.5%.
The six-month allocation threshold above which a two-dose regimen would be preferred is much lower in LMICs, mainly due to mortality in the younger population. In these contexts, decision-makers will need to consider the affordability, availability, and logistical constraints and feasibility of implementing a single or a double dose, the dosage intervals, and delivery speed. Most LMICs lack the digital databases necessary to manage patient data, reliably track vaccine inventories, keep track of who has received which vaccine, and inform people where and when they are due for a booster. Governments would also need to ensure that they reserve sufficient stocks to allow the administration of booster doses. In these cases, a robust cost-benefit analysis of each option will need to be considered.
The dosing interval for COVID-19 vaccines has been a subject of debate among scientists, regulators, and governments around the world following the UK government's decision to prioritize administering the first dose of vaccine to as many at-risk people as possible and increasing the interval between the two doses to up to 12 weeks 14,15,20 . A one-dose vaccine regimen or a two-dose regimen with longer time intervals may be sufficient to reduce symptoms of COVID-19 in the most vulnerable individuals and ultimately slow the pandemic, given that the time difference between first and second doses was shown to have a negligible effect on overall vaccine effectiveness (clinical cases, infections, and deaths). Indeed, a recent WHO notification stated that some countries are facing "exceptional circumstances" and may want to delay second doses to "allow for a higher initial coverage". Other exceptional circumstances may involve trade-offs around the relative size of the highest risk population in a country and the currently unknown potential for a vaccine to reduce transmission, which may lead to some countries targeting highcontact groups to benefit from any potential indirect effects.
Nevertheless, these thresholds are likely to differ depending on the country context. For example, smaller countries may be able to rapidly rollout the vaccine to a higher percentage of their population compared with the speed at which larger countries can do this. It should be noted that an implementation strategy is determined at a country level. The assumptions made in this work are based on the association of certain parameters with the health infrastructure and existing population of certain country groups defined by income level.
Published clinical data were used to inform the parameters used in the model described in this paper. These data provide an aggregate efficacy of the ChAdOx1 nCoV-19 vaccine among people of a wide range of ages living in different countries. However, there were limited data available for assessing the effects of certain parameters (such as the effect of the dosing interval on post-dose two efficacy) on vaccine efficacy, which begat the need to conduct the post hoc exploratory analysis presented here.

Conclusion
This analysis demonstrates that in scenarios where the number of vaccine doses available is insufficient for the highest risk groups (>65 years of age) to receive two vaccine doses, allocation of a single vaccine dose to twice the number of individuals or extending the time interval between doses may be more optimal strategies. In contexts without supply constraints, or if the single-dose efficacy is low, providing a booster dose would be preferable. Apart from these specific conditions, the results indicate that providing individuals with two doses of vaccine would have a similar effectiveness to the use of a single dose in twice the number of individuals. In an ideal world, decisions about vaccination strategies would be made within the exact parameters of the trials that have been conducted. However, the limited availability of resources, and specific country contexts, may require decision-makers to consider alternative strategies. Vaccine protection decay post-last dose The boxplots show the median and interquartile ranges of the predicted vaccine effectiveness on each of the outcomes for specific parameters. They were generated by aggregating all model simulations for each of the parameters, with each boxplot summarizing the variance in predicted vaccine efficacy for all possible combinations of the other parameters. The full list of parameters explored and their descriptions can be found in Table 1. The coloured surfaces and respective contour lines indicate the ratio between the predicted vaccine effectiveness for a double-dose regimen vs a single-dose regimen. This ratio is a mean ratio, obtained by averaging out the ratios obtained in all model runs assuming the corresponding x and y parameter values and thus are not expected to be regular. Contour line 1 (thicker black line) indicates the parameter combinations for which there is no expected difference between giving everyone a single dose vs giving everyone two doses. For values greater than 1 (hot colours), a two-dose regimen is preferable, and for values less than 1 (cold colours), a single-dose regimen is preferable.

Fig. 3: Dose allocation thresholds in different countries.
The figure illustrates the parameter combinations that define the allocation threshold above which a two-dose regimen would be preferred over a single-dose regimen. The areas under the curves are 16.5%, 8%, and 3.8% for the UK, Brazil, and Uganda, respectively, which correlates almost perfectly with the proportion of the population above the age of 65 years in those countries. Supplementary Fig. 3: Detailed sensitivity analysis of vaccine effectiveness for the most sensitive parameters, based on UK data.
The boxplots show the median and interquartile ranges of the predicted vaccine effectiveness on each of the outcomes for specific parameters. They were generated by aggregating all model simulations for each of the parameters, with each boxplot summarizing the variance in predicted vaccine efficacy for all possible combinations of the other parameters.