Multiple models for outbreak decision support in the face of uncertainty.

Policymakers must make management decisions despite incomplete knowledge and conflicting model projections. Little guidance exists for the rapid, representative, and unbiased collection of policy-relevant scientific input from independent modeling teams. Integrating approaches from decision analysis, expert judgment, and model aggregation, we convened multiple modeling teams to evaluate COVID-19 reopening strategies for a mid-sized United States county early in the pandemic. Projections from seventeen distinct models were inconsistent in magnitude but highly consistent in ranking interventions. The 6-mo-ahead aggregate projections were well in line with observed outbreaks in mid-sized US counties. The aggregate results showed that up to half the population could be infected with full workplace reopening, while workplace restrictions reduced median cumulative infections by 82%. Rankings of interventions were consistent across public health objectives, but there was a strong trade-off between public health outcomes and duration of workplace closures, and no win-win intermediate reopening strategies were identified. Between-model variation was high; the aggregate results thus provide valuable risk quantification for decision making. This approach can be applied to the evaluation of management interventions in any setting where models are used to inform decision making. This case study demonstrated the utility of our approach and was one of several multimodel efforts that laid the groundwork for the COVID-19 Scenario Modeling Hub, which has provided multiple rounds of real-time scenario projections for situational awareness and decision making to the Centers for Disease Control and Prevention since December 2020.

Heterologous vaccination interventions to reduce pandemic morbidity and mortality: Modeling the US winter 2020 COVID-19 wave.

COVID-19 remains a stark health threat worldwide, in part because of minimal levels of targeted vaccination outside high-income countries and highly transmissible variants causing infection in vaccinated individuals. Decades of theoretical and experimental data suggest that nonspecific effects of non-COVID-19 vaccines may help bolster population immunological resilience to new pathogens. These routine vaccinations can stimulate heterologous cross-protective effects, which modulate nontargeted infections. For example, immunization with Bacillus Calmette-Guérin, inactivated influenza vaccine, oral polio vaccine, and other vaccines have been associated with some protection from SARS-CoV-2 infection and amelioration of COVID-19 disease. If heterologous vaccine interventions (HVIs) are to be seriously considered by policy makers as bridging or boosting interventions in pandemic settings to augment nonpharmaceutical interventions and specific vaccination efforts, evidence is needed to determine their optimal implementation. Using the COVID-19 International Modeling Consortium mathematical model, we show that logistically realistic HVIs with low (5 to 15%) effectiveness could have reduced COVID-19 cases, hospitalization, and mortality in the United States fall/winter 2020 wave. Similar to other mass drug administration campaigns (e.g., for malaria), HVI impact is highly dependent on both age targeting and intervention timing in relation to incidence, with maximal benefit accruing from implementation across the widest age cohort when the pandemic reproduction number is >1.0. Optimal HVI logistics therefore differ from optimal rollout parameters for specific COVID-19 immunizations. These results may be generalizable beyond COVID-19 and the US to indicate how even minimally effective heterologous immunization campaigns could reduce the burden of future viral pandemics.

Participatory Mathematical Modeling Approach for Policymaking during the First Year of the COVID-19 Crisis, Jordan.

We engaged in a participatory modeling approach with health sector stakeholders in Jordan to support government decision-making regarding implementing public health measures to mitigate COVID-19 disease burden. We considered the effect of 4 physical distancing strategies on reducing COVID-19 transmission and mortality in Jordan during March 2020-January 2021: no physical distancing; intermittent physical distancing where all but essential services are closed once a week; intermittent physical distancing where all but essential services are closed twice a week; and a permanent physical distancing intervention. Modeling showed that the fourth strategy would be most effective in reducing cases and deaths; however, this approach was only marginally beneficial to reducing COVID-19 disease compared with an intermittently enforced physical distancing intervention. Scenario-based model influenced policy-making and the evolution of the pandemic in Jordan confirmed the forecasting provided by the modeling exercise and helped confirm the effectiveness of the policy adopted by the government of Jordan.

Individual variation in susceptibility or exposure to SARS-CoV-2 lowers the herd immunity threshold.

Individual variation in susceptibility and exposure is subject to selection by natural infection, accelerating the acquisition of immunity, and reducing herd immunity thresholds and epidemic final sizes. This is a manifestation of a wider population phenomenon known as "frailty variation". Despite theoretical understanding, public health policies continue to be guided by mathematical models that leave out considerable variation and as a result inflate projected disease burdens and overestimate the impact of interventions. Here we focus on trajectories of the coronavirus disease (COVID-19) pandemic in England and Scotland until November 2021. We fit models to series of daily deaths and infer relevant epidemiological parameters, including coefficients of variation and effects of non-pharmaceutical interventions which we find in agreement with independent empirical estimates based on contact surveys. Our estimates are robust to whether the analysed data series encompass one or two pandemic waves and enable projections compatible with subsequent dynamics. We conclude that vaccination programmes may have contributed modestly to the acquisition of herd immunity in populations with high levels of pre-existing naturally acquired immunity, while being crucial to protect vulnerable individuals from severe outcomes as the virus becomes endemic.

Levels of SARS-CoV-2 population exposure are considerably higher than suggested by seroprevalence surveys.

Accurate knowledge of prior population exposure has critical ramifications for preparedness plans for future SARS-CoV-2 epidemic waves and vaccine prioritization strategies. Serological studies can be used to estimate levels of past exposure and thus position populations in their epidemic timeline. To circumvent biases introduced by the decay in antibody titers over time, methods for estimating population exposure should account for seroreversion, to reflect that changes in seroprevalence measures over time are the net effect of increases due to recent transmission and decreases due to antibody waning. Here, we present a new method that combines multiple datasets (serology, mortality, and virus positivity ratios) to estimate seroreversion time and infection fatality ratios (IFR) and simultaneously infer population exposure levels. The results indicate that the average time to seroreversion is around six months, IFR is 0.54% to 1.3%, and true exposure may be more than double the current seroprevalence levels reported for several regions of England.

Severe Acute Respiratory Syndrome Coronavirus 2 Infection Among Healthcare Workers in South Africa: A Longitudinal Cohort Study.

From April to September 2020, we investigated severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infections in a cohort of 396 healthcare workers (HCWs) from 5 departments at Chris Hani Baragwanath Hospital, South Africa. Overall, 34.6% of HCWs had polymerase chain reaction-confirmed SARS-CoV-2 infection (132.1 [95% confidence interval, 111.8-156.2] infections per 1000 person-months); an additional 27 infections were identified by serology. HCWs in the internal medicine department had the highest rate of infection (61.7%). Among polymerase chain reaction-confirmed cases, 10.4% remained asymptomatic, 30.4% were presymptomatic, and 59.3% were symptomatic.

The assembly effect: the connectedness between populations is a double-edged sword for public health interventions.

BACKGROUND: Many public health interventions lead to disruption or decrease of transmission, providing a beneficial effect for people in the population regardless of whether or not they individually participate in the intervention. This protective benefit has been referred to as a herd or community effect and is dependent on sufficient population participation. In practice, public health interventions are implemented at different spatial scales (i.e., at the village, district, or provincial level). Populations, however defined (i.e., neighbourhoods, villages, districts) are frequently connected to other populations through human movement or travel, and this connectedness can influence potential herd effects. METHODS: The impact of a public health intervention (mass drug administration for malaria) was modelled, for different levels of connectedness between populations that have similar disease epidemiology (e.g., two nearby villages which have similar baseline malaria incidences and similar malaria intervention measures), or between populations of varying disease epidemiology (e.g., two nearby villages which have different baseline malaria incidences and/or malaria intervention measures). RESULTS: The overall impact of the interventions deployed could be influenced either positively (adding value to the intervention) or negatively (reducing the impact of the intervention) by how much the intervention units are connected with each other (e.g., how frequent people go to the other village or town) and how different the disease intensity between them are. This phenomenon is termed the "assembly effect", and it is a meta-population version of the more commonly understood "herd effect". CONCLUSIONS: The connectedness of intervention units or populations is an important factor to be considered to achieve success in public health interventions that could provide herd effects. Appreciating the assembly effect can improve the cost-effective strategies for global disease elimination projects.

Assessing the impact of next-generation rapid diagnostic tests on Plasmodium falciparum malaria elimination strategies.

Mass-screen-and-treat and targeted mass-drug-administration strategies are being considered as a means to interrupt transmission of Plasmodium falciparum malaria. However, the effectiveness of such strategies will depend on the extent to which current and future diagnostics are able to detect those individuals who are infectious to mosquitoes. We estimate the relationship between parasite density and onward infectivity using sensitive quantitative parasite diagnostics and mosquito feeding assays from Burkina Faso. We find that a diagnostic with a lower detection limit of 200 parasites per microlitre would detect 55% of the infectious reservoir (the combined infectivity to mosquitoes of the whole population weighted by how often each individual is bitten) whereas a test with a limit of 20 parasites per microlitre would detect 83% and 2 parasites per microlitre would detect 95% of the infectious reservoir. Using mathematical models, we show that increasing the diagnostic sensitivity from 200 parasites per microlitre (equivalent to microscopy or current rapid diagnostic tests) to 2 parasites per microlitre would increase the number of regions where transmission could be interrupted with a mass-screen-and-treat programme from an entomological inoculation rate below 1 to one of up to 4. The higher sensitivity diagnostic could reduce the number of treatment rounds required to interrupt transmission in areas of lower prevalence. We predict that mass-screen-and-treat with a highly sensitive diagnostic is less effective than mass drug administration owing to the prophylactic protection provided to uninfected individuals by the latter approach. In low-transmission settings such as those in Southeast Asia, we find that a diagnostic tool with a sensitivity of 20 parasites per microlitre may be sufficient for targeted mass drug administration because this diagnostic is predicted to identify a similar village population prevalence compared with that currently detected using polymerase chain reaction if treatment levels are high and screening is conducted during the dry season. Along with other factors, such as coverage, choice of drug, timing of the intervention, importation of infections, and seasonality, the sensitivity of the diagnostic can play a part in increasing the chance of interrupting transmission.

Estimating the programmatic cost of targeted mass drug administration for malaria in Myanmar.

BACKGROUND: Mass drug administration (MDA) has received growing interest to accelerate the elimination of multi-drug resistant malaria in the Greater Mekong Subregion. Targeted MDA, sometimes referred to as focal MDA, is the practice of delivering MDA to high incidence subpopulations only, rather than the entire population. The potential effectiveness of delivering targeted MDA was demonstrated in a recent intervention in Kayin State, Myanmar. Policymakers and funders need to know what resources are required if MDA, targeted or otherwise, is to be included in elimination packages beyond existing malaria interventions. This study aims to estimate the programmatic cost and the unit cost of targeted MDA in Kayin State, Myanmar. METHODS: We used financial data from a malaria elimination initiative, conducted in Kayin State, to estimate the programmatic costs of the targeted MDA component using a micro-costing approach. Three activities (community engagement, identification of villages for targeted MDA, and conducting mass treatment in target villages) were evaluated. We then estimated the programmatic costs of implementing targeted MDA to support P. falciparum malaria elimination in Kayin State. A costing tool was developed to aid future analyses. RESULTS: The cost of delivering targeted MDA within an integrated malaria elimination initiative in eastern Kayin State was approximately US$ 910,000. The cost per person reached, distributed among those in targeted and non-targeted villages, for the MDA component was US$ 2.5. CONCLUSION: This cost analysis can assist policymakers in determining the resources required to clear malaria parasite reservoirs. The analysis demonstrated the value of using financial data from research activities to predict programmatic implementation costs of targeting MDA to different numbers of target villages.

Algorithm in the Diagnosis of Febrile Illness Using Pathogen-specific Rapid Diagnostic Tests.

BACKGROUND: In the absence of proper guidelines and algorithms, available rapid diagnostic tests (RDTs) for common acute undifferentiated febrile illnesses are often used inappropriately. METHODS: Using prevalence data of 5 common febrile illnesses from India and Cambodia, and performance characteristics (sensitivity and specificity) of relevant pathogen-specific RDTs, we used a mathematical model to predict the probability of correct identification of each disease when diagnostic testing occurs either simultaneously or sequentially in various algorithms. We developed a web-based application of the model so as to visualize and compare output diagnostic algorithms when different disease prevalence and test performance characteristics are introduced. RESULTS: Diagnostic algorithms with appropriate sequential testing predicted correct identification of etiology in 74% and 89% of patients in India and Cambodia, respectively, compared with 46% and 49% with simultaneous testing. The optimally performing sequential diagnostic algorithms differed in India and Cambodia due to varying disease prevalence. CONCLUSIONS: Simultaneous testing is not appropriate for the diagnosis of acute undifferentiated febrile illnesses with presently available tests, which should deter the unsupervised use of multiplex diagnostic tests. The implementation of adaptive algorithms can predict better diagnosis and add value to the available RDTs. The web application of the model can serve as a tool to identify the optimal diagnostic algorithm in different epidemiological settings, while taking into account the local epidemiological variables and accuracy of available tests.

Costing malaria interventions from pilots to elimination programmes.

BACKGROUND: Malaria programmes in countries with low transmission levels require evidence to optimize deployment of current and new tools to reach elimination with limited resources. Recent pilots of elimination strategies in Ethiopia, Senegal, and Zambia produced evidence of their epidemiological impacts and costs. There is a need to generalize these findings to different epidemiological and health systems contexts. METHODS: Drawing on experience of implementing partners, operational documents and costing studies from these pilots, reference scenarios were defined for rapid reporting (RR), reactive case detection (RACD), mass drug administration (MDA), and in-door residual spraying (IRS). These generalized interventions from their trial implementation to one typical of programmatic delivery. In doing so, resource use due to interventions was isolated from research activities and was related to the pilot setting. Costing models developed around this reference implementation, standardized the scope of resources costed, the valuation of resource use, and the setting in which interventions were evaluated. Sensitivity analyses were used to inform generalizability of the estimates and model assumptions. RESULTS: Populated with local prices and resource use from the pilots, the models yielded an average annual economic cost per capita of $0.18 for RR, $0.75 for RACD, $4.28 for MDA (two rounds), and $1.79 for IRS (one round, 50% households). Intervention design and resource use at service delivery were key drivers of variation in costs of RR, MDA, and RACD. Scale was the most important parameter for IRS. Overall price level was a minor contributor, except for MDA where drugs accounted for 70% of the cost. The analyses showed that at implementation scales comparable to health facility catchment area, systematic correlations between model inputs characterizing implementation and setting produce large gradients in costs. CONCLUSIONS: Prospective costing models are powerful tools to explore resource and cost implications of policy alternatives. By formalizing translation of operational data into an estimate of intervention cost, these models provide the methodological infrastructure to strengthen capacity gap for economic evaluation in endemic countries. The value of this approach for decision-making is enhanced when primary cost data collection is designed to enable analysis of the efficiency of operational inputs in relation to features of the trial or the setting, thus facilitating transferability.

The impact of targeted malaria elimination with mass drug administrations on falciparum malaria in Southeast Asia: A cluster randomised trial.

BACKGROUND: The emergence and spread of multidrug-resistant Plasmodium falciparum in the Greater Mekong Subregion (GMS) threatens global malaria elimination efforts. Mass drug administration (MDA), the presumptive antimalarial treatment of an entire population to clear the subclinical parasite reservoir, is a strategy to accelerate malaria elimination. We report a cluster randomised trial to assess the effectiveness of dihydroartemisinin-piperaquine (DP) MDA in reducing falciparum malaria incidence and prevalence in 16 remote village populations in Myanmar, Vietnam, Cambodia, and the Lao People's Democratic Republic, where artemisinin resistance is prevalent. METHODS AND FINDINGS: After establishing vector control and community-based case management and following intensive community engagement, we used restricted randomisation within village pairs to select 8 villages to receive early DP MDA and 8 villages as controls for 12 months, after which the control villages received deferred DP MDA. The MDA comprised 3 monthly rounds of 3 daily doses of DP and, except in Cambodia, a single low dose of primaquine. We conducted exhaustive cross-sectional surveys of the entire population of each village at quarterly intervals using ultrasensitive quantitative PCR to detect Plasmodium infections. The study was conducted between May 2013 and July 2017. The investigators randomised 16 villages that had a total of 8,445 residents at the start of the study. Of these 8,445 residents, 4,135 (49%) residents living in 8 villages, plus an additional 288 newcomers to the villages, were randomised to receive early MDA; 3,790 out of the 4,423 (86%) participated in at least 1 MDA round, and 2,520 out of the 4,423 (57%) participated in all 3 rounds. The primary outcome, P. falciparum prevalence by month 3 (M3), fell by 92% (from 5.1% [171/3,340] to 0.4% [12/2,828]) in early MDA villages and by 29% (from 7.2% [246/3,405] to 5.1% [155/3,057]) in control villages. Over the following 9 months, the P. falciparum prevalence increased to 3.3% (96/2,881) in early MDA villages and to 6.1% (128/2,101) in control villages (adjusted incidence rate ratio 0.41 [95% CI 0.20 to 0.84]; p = 0.015). Individual protection was proportional to the number of completed MDA rounds. Of 221 participants with subclinical P. falciparum infections who participated in MDA and could be followed up, 207 (94%) cleared their infections, including 9 of 10 with artemisinin- and piperaquine-resistant infections. The DP MDAs were well tolerated; 6 severe adverse events were detected during the follow-up period, but none was attributable to the intervention. CONCLUSIONS: Added to community-based basic malaria control measures, 3 monthly rounds of DP MDA reduced the incidence and prevalence of falciparum malaria over a 1-year period in areas affected by artemisinin resistance. P. falciparum infections returned during the follow-up period as the remaining infections spread and malaria was reintroduced from surrounding areas. Limitations of this study include a relatively small sample of villages, heterogeneity between villages, and mobility of villagers that may have limited the impact of the intervention. These results suggest that, if used as part of a comprehensive, well-organised, and well-resourced elimination programme, DP MDA can be a useful additional tool to accelerate malaria elimination. TRIAL REGISTRATION: ClinicalTrials.gov NCT01872702.

Unraveling the drivers of MERS-CoV transmission.

With more than 1,700 laboratory-confirmed infections, Middle East respiratory syndrome coronavirus (MERS-CoV) remains a significant threat for public health. However, the lack of detailed data on modes of transmission from the animal reservoir and between humans means that the drivers of MERS-CoV epidemics remain poorly characterized. Here, we develop a statistical framework to provide a comprehensive analysis of the transmission patterns underlying the 681 MERS-CoV cases detected in the Kingdom of Saudi Arabia (KSA) between January 2013 and July 2014. We assess how infections from the animal reservoir, the different levels of mixing, and heterogeneities in transmission have contributed to the buildup of MERS-CoV epidemics in KSA. We estimate that 12% [95% credible interval (CI): 9%, 15%] of cases were infected from the reservoir, the rest via human-to-human transmission in clusters (60%; CI: 57%, 63%), within (23%; CI: 20%, 27%), or between (5%; CI: 2%, 8%) regions. The reproduction number at the start of a cluster was 0.45 (CI: 0.33, 0.58) on average, but with large SD (0.53; CI: 0.35, 0.78). It was >1 in 12% (CI: 6%, 18%) of clusters but fell by approximately one-half (47% CI: 34%, 63%) its original value after 10 cases on average. The ongoing exposure of humans to MERS-CoV from the reservoir is of major concern, given the continued risk of substantial outbreaks in health care systems. The approach we present allows the study of infectious disease transmission when data linking cases to each other remain limited and uncertain.

Infectivity of Chronic Malaria Infections and Its Consequences for Control and Elimination.

Assessing the importance of targeting the chronic Plasmodium falciparum malaria reservoir is pivotal as the world moves toward malaria eradication. Through the lens of a mathematical model, we show how, for a given malaria prevalence, the relative infectivity of chronic individuals determines what intervention tools are predicted be the most effective. Crucially, in a large part of the parameter space where elimination is theoretically possible, it can be achieved solely through improved case management. However, there are a significant number of settings where malaria elimination requires not only good vector control but also a mass drug administration campaign. Quantifying the relative infectiousness of chronic malaria across a range of epidemiological settings would provide essential information for the design of effective malaria elimination strategies. Given the difficulties obtaining this information, we also provide a set of epidemiological metrics that can be used to guide policy in the absence of such data.

International risk of yellow fever spread from the ongoing outbreak in Brazil, December 2016 to May 2017.

States in south-eastern Brazil were recently affected by the largest Yellow Fever (YF) outbreak seen in a decade in Latin America. Here we provide a quantitative assessment of the risk of travel-related international spread of YF indicating that the United States, Argentina, Uruguay, Spain, Italy and Germany may have received at least one travel-related YF case capable of seeding local transmission. Mitigating the risk of imported YF cases seeding local transmission requires heightened surveillance globally.

Estimating the Severity and Subclinical Burden of Middle East Respiratory Syndrome Coronavirus Infection in the Kingdom of Saudi Arabia.

Not all persons infected with Middle East respiratory syndrome coronavirus (MERS-CoV) develop severe symptoms, which likely leads to an underestimation of the number of people infected and an overestimation of the severity. To estimate the number of MERS-CoV infections that have occurred in the Kingdom of Saudi Arabia, we applied a statistical model to a line list describing 721 MERS-CoV infections detected between June 7, 2012, and July 25, 2014. We estimated that 1,528 (95% confidence interval (CI): 1,327, 1,883) MERS-CoV infections occurred in this interval, which is 2.1 (95% CI: 1.8, 2.6) times the number reported. The probability of developing symptoms ranged from 11% (95% CI: 4, 25) in persons under 10 years of age to 88% (95% CI: 72, 97) in those 70 years of age or older. An estimated 22% (95% CI: 18, 25) of those infected with MERS-CoV died. MERS-CoV is deadly, but this work shows that its clinical severity differs markedly between groups and that many cases likely go undiagnosed.

Feature selection methods for identifying genetic determinants of host species in RNA viruses.

Despite environmental, social and ecological dependencies, emergence of zoonotic viruses in human populations is clearly also affected by genetic factors which determine cross-species transmission potential. RNA viruses pose an interesting case study given their mutation rates are orders of magnitude higher than any other pathogen--as reflected by the recent emergence of SARS and Influenza for example. Here, we show how feature selection techniques can be used to reliably classify viral sequences by host species, and to identify the crucial minority of host-specific sites in pathogen genomic data. The variability in alleles at those sites can be translated into prediction probabilities that a particular pathogen isolate is adapted to a given host. We illustrate the power of these methods by: 1) identifying the sites explaining SARS coronavirus differences between human, bat and palm civet samples; 2) showing how cross species jumps of rabies virus among bat populations can be readily identified; and 3) de novo identification of likely functional influenza host discriminant markers.

Preventing antimalarial drug resistance with triple artemisinin-based combination therapies.

Increasing levels of artemisinin and partner drug resistance threaten malaria control and elimination globally. Triple artemisinin-based combination therapies (TACTs) which combine artemisinin derivatives with two partner drugs are efficacious and well tolerated in clinical trials, including in areas of multidrug-resistant malaria. Whether early TACT adoption could delay the emergence and spread of antimalarial drug resistance is a question of vital importance. Using two independent individual-based models of Plasmodium falciparum epidemiology and evolution, we evaluated whether introduction of either artesunate-mefloquine-piperaquine or artemether-lumefantrine-amodiaquine resulted in lower long-term artemisinin-resistance levels and treatment failure rates compared with continued ACT use. We show that introduction of TACTs could significantly delay the emergence and spread of artemisinin resistance and treatment failure, extending the useful therapeutic life of current antimalarial drugs, and improving the chances of malaria elimination. We conclude that immediate introduction of TACTs should be considered by policy makers in areas of emerging artemisinin resistance.

Epidemiological waves - Types, drivers and modulators in the COVID-19 pandemic.

Introduction: A discussion of 'waves' of the COVID-19 epidemic in different countries is a part of the national conversation for many, but there is no hard and fast means of delineating these waves in the available data and their connection to waves in the sense of mathematical epidemiology is only tenuous. Methods: We present an algorithm which processes a general time series to identify substantial, significant and sustained periods of increase in the value of the time series, which could reasonably be described as 'observed waves'. This provides an objective means of describing observed waves in time series. We use this method to synthesize evidence across different countries to study types, drivers and modulators of waves. Results: The output of the algorithm as applied to epidemiological time series related to COVID-19 corresponds to visual intuition and expert opinion. Inspecting the results of individual countries shows how consecutive observed waves can differ greatly with respect to the case fatality ratio. Furthermore, in large countries, a more detailed analysis shows that consecutive observed waves have different geographical ranges. We also show how waves can be modulated by government interventions and find that early implementation of NPIs correlates with a reduced number of observed waves and reduced mortality burden in those waves. Conclusion: It is possible to identify observed waves of disease by algorithmic methods and the results can be fruitfully used to analyse the progression of the epidemic.

How to model the impact of vaccines for policymaking when the characteristics are uncertain: A case study in Thailand prior to the vaccine rollout during the COVID-19 pandemic.

Thailand faced a dilemma of which groups to prioritise with a limited first tranche of COVID-19 vaccinations in early 2021, at a time when there was low incidence and low mortality in the country. A mathematical modelling analysis was performed to compare the potential short-term impact of allocating the available doses to either the high severity group (over 65-year-olds) or the high transmission group (aged 20-39). At the time of the analysis, there was uncertainty about the precise characteristics of the vaccines available, in terms of their potential impact on transmission and reductions to the severity of infection. As such, a range of vaccine characteristic scenarios, with differing levels of severity and transmission reductions were explored. Using the evidence available at the time regarding severity reduction of infection due to the vaccines, the model suggested that vaccinating high severity group should be the priority if reductions in deaths is the priority. Vaccinating this group was found to have a direct impact on reducing the number of deaths, while the incidence and hospitalisations remained unchanged. However, the model found that vaccinating the high transmission group with a vaccine with sufficiently high protection against infection (more than 70%) could provide enough herd effects to delay the expected epidemic peak, resulting in both case and death reductions in both target groups. The model explored a 12-month time horizon. These analyses helped to inform the vaccination strategy in Thailand throughout 2021 and can inform future modelling studies for policymaking when the characteristics of vaccines are uncertain.