In the first search, our database search returned 2279 studies. After the removal of 459 duplicates, 1820 records were included in the title and abstract screening. After title and abstract screening, 55 studies remained for full-text screening. After assessing the full text of the 55 studies for eligibility, 36 studies were included in the systematic review. After updating the search to look for studies set in LMIC countries having identified this as the predominant gap in the literature, 7 more studies were included (Fig. 2).
Study characteristics are summarised in Table 1. All the reviewed studies used models that captured transmission between individuals, with deterministic compartmental models being most common (28/43). However, agent-based models (6/43), stochastic compartmental models (4/43), a delay differential equation model (1/43), and a linear model (1/43) were also used. Studies most commonly used a SEIR (Susceptible, Exposed, Infected, Recovered) (12/43) or Expanded SEIR (19/43) natural history. Most of them were set in a HIC (26/43); there were few single-country UMIC (3/43) and LMIC (5/43) studies. There were no single-country studies in a LIC setting. Only a few (6/43) looked at more than one country and two did not explicitly state the study setting. Most studies explored multiple policy objectives/outcomes regarding prioritisation: 34/43 investigated strategies to minimise deaths, 27/43 investigated minimisation of cases, 11/43 hospitalisations, 1/43 quality adjusted life years (QALYs), 1/43 disability adjusted life years (DALYs), and 3/43 years of life lost (YLLs). Only 2/43 considered economic outcomes, such as financial or economic costs, in relation to prioritisation.
Prioritisation to minimise deaths
Table S2 (Additional file 1) summarises the study conclusions highlighting the priority group and all the comparators included in each study (see the ‘Methods’ section for how we defined population group categories). Most studies included seniors in the priority group. Nineteen studies recommended that seniors should be prioritised for vaccination to minimise deaths [7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25]. One study recommended prioritising seniors with comorbidities , and six studies recommended vaccinating seniors at the same time as another priority group (middle-aged adults, the highest social contact group, young and middle-aged adults who are in high contact with them, young adults with partial vaccine dose, and health workers) [27,28,29,30,31,32].
Ten studies did not find that prioritising the groups at highest risk of mortality (seniors or people living with comorbidities) minimised deaths (for a summary of these studies see Additional file 1: Table S3) [29, 30, 33,34,35,36,37,38,39,40]. These ‘exceptional’ studies instead found that prioritising groups with a higher risk of infection would lead to fewer deaths; in other words, that the indirect protection from lower transmission outweighs the benefits from direct protection from vaccines for those at the highest risk of mortality. The group at high risk of infection were defined as those with higher contact rates, e.g. a synthetic population with 3–10 times the average number of contacts of the age groups 30–39 , individuals with an expansive social network [37, 38], and individuals with essential worker status [33, 34]. In addition, two studies defined young adults as the group with the highest social interactions and therefore at higher risk of infection [35, 40]. One study examined vaccination of individuals that had high levels of interaction with seniors .
One of the ten exceptional studies concluded that the non-vulnerable group should be prioritised for vaccination compared to the group with comorbidities . In this study, the authors state they assumed that ‘the time required to vaccinate the vulnerable group is identical to that of the much larger non-vulnerable group’. Finally, one study recommended prioritising vaccination through the use of serological testing to achieve the greatest reduction in COVID-19-related deaths .
Prioritisation to minimise COVID-19 cases
Table S4 (Additional file 1) summarises the study conclusions. The largest proportion of the selected studies (N=27) investigated optimal vaccine prioritisation strategies to minimise COVID-19 cases. Of these, seven studies recommended young and middle-aged adults [10, 12, 15, 16, 17, 20, 28], three young adults [23, 25, 39], and two young adults and children [27, 32]. One study recommended young people at the same time as another priority group (seniors at full vaccine dose and young adults at partial dose) .
Seven studies recommended vaccination prioritisation based on social or occupational interactions compared to age group prioritisation [14, 33, 34, 37, 38, 42, 43]. Of these, three studies recommended prioritising essential workers to minimise cases [14, 33, 34], and four studies recommended prioritising high social contact adults compared to other age groups [37, 38, 42, 43].
Two studies recommended prioritising vaccination using serological testing to prioritise antibody-negative individuals compared to not using serological testing [41, 44]. Two studies investigated allocation between geographic areas of disease burden—the first recommended that the geographic area with lower disease burden should be prioritised for vaccination, whereas the second recommended that the geographic area with the highest disease burden should be prioritised [45–46].
There were a few studies concluding differently to the majority recommendations on minimising cases (for a summary of these studies see Additional file 1: Table S5). Three studies found that scenarios targeting seniors [13, 18, 31] led to the highest reduction in cases. However, two of those studies did not have a comparator that modelled those strategies comparted to more socially interactive populations [18, 31]. Chhetri et al. found very small differences between scenarios, and the conclusion was not reported in the “Results” section .
Prioritising other outcomes
Studies investigating strategies to minimise hospitalisations from COVID-19 tended to reach similar conclusions to studies investigating deaths (N=11). Eight studies recommended prioritising seniors [8, 15, 25, 16, 43, 47], senior- and middle-aged adults , or seniors and the high social contact group  for vaccination compared to other age and occupational groups. Four studies concluded differently from the majority of the hospitalisation outcome studies [37, 38, 44, 48]. Two recommended prioritising the high social contact group compared to prioritising senior adults [37, 38]. One study recommended prioritising vaccination by serological testing compared to no serological testing . One study recommended giving equal priority to all age and risk groups compared to a targeted age-based prioritisation .
A few studies investigated the optimal vaccination strategy when maximising QALYs, DALYs, or YLLs. One study modelled a vaccination prioritisation strategy to minimise QALY losses . The authors concluded that the most effective strategy to minimise QALY losses is to prioritise seniors for vaccination compared to other age groups, groups with comorbidities, and no group prioritisation. Three studies investigated within-country vaccine prioritisation strategies for minimising YLLs [12, 28, 34]. Two studies recommended prioritising seniors for vaccination to minimise YLLs [12, 34], and the other recommended prioritising middle aged adults and seniors . One study modelled the impact of COVID-19 vaccination on DALYs . The authors found that the amount of DALYs averted under a base vaccination strategy which prioritised seniors was stable to a scenario where everyone over 15 years old is vaccinated .
One study considered the cost-effectiveness of COVID-19 vaccination . The authors found that the strategy of prioritising seniors for vaccination was similarly cost-effective to vaccinating all individuals over 15 years old .
One study investigated prioritisation strategies for optimising the incremental net monetary benefit (iNMB) of vaccination, i.e. the net economic gain from vaccination including both costs saved and monetised health gains . The authors concluded that giving equal priority to all age and risk groups was most optimal compared to prioritising seniors, high risk individuals, and both seniors and high-risk individuals when vaccine effectiveness was only moderate (40%) and coverage was low (20%). Conversely, when vaccine effectiveness was high (80%) and coverage was moderate (50%), vaccinating high risk individuals resulted in the highest iNMB.
Prioritisation by setting
Five of the included articles were single-country studies modelling LMIC settings [14, 17, 19, 22, 46]. Four studies modelled UMIC settings [21, 23, 25, 33]. These studies reached the same conclusions as the HIC studies i.e. studies minimising deaths recommended prioritising seniors, while those minimising cases recommended prioritising high transmission groups. The exception was one study from Thailand on minimising cases which recommended prioritising high transmission groups to minimise deaths .
There were also five multi-country studies which modelled LMIC settings [8, 9, 12, 20, 24] and two modelling UMIC settings [20, 29]. The conclusions for these studies were in line with the majority conclusions for deaths and cases (except for one study which recommended prioritising both the high social contact group and seniors to minimise deaths) . See Additional file 1: Table S6 for a summary of the studies modelling UMIC and LMIC settings.
One study also considered prioritisation between countries, in addition to within countries. This study made recommendations on global vaccine allocation strategies to optimise different health objectives . The authors concluded that the optimal strategy to minimise deaths was to allocate doses equitably across all income settings relative to population size and then to prioritise vaccination of seniors within countries. This performed better than allocating vaccines to countries based on their respective senior population sizes, giving preferential allocation to HICs, giving preferential allocation to LICs and LMICs, or allocating doses in proportion to population plus providing a set number of extra doses to HIC and UMICs . When YLLs were used as an optimisation outcome measure, LMIC settings received the most doses.
Factors that influence prioritisation strategy
40 out of 43 (93%) studies included a sensitivity analysis (see Additional file 1: Table S7 for a summary of these studies). Of these, 17 studies reported a sensitivity analysis that led to a potential change in the recommended prioritisation strategy. While there were a wide range of parameters tested in the uncertainty analysis, there were only a few that consistently drove a change in prioritisation. The most common parameters that influenced prioritisation all related to vaccine coverage, i.e. level of vaccine supply, coverage, and speed of rollout (see Additional file 1: Table S8 for a summary of coverage level assumptions made by the exceptional studies to the majority of the study conclusions). Transmission rates and vaccine efficacy were also considered.
Eight studies reported that the trade-off between direct and indirect protection is sensitive to the proportion of people vaccinated [9, 12, 15, 22, 27, 28, 38, 25]. These papers stated that when vaccine supply is very low, vaccination has a minimal impact on interrupting transmission, so more deaths can be prevented by vaccinating groups at risk of severe disease (e.g. essential workers, seniors, and clinical risk groups). However, as supply increases, this opens up the possibility of interrupting transmission, which shifts the optimal policy for preventing deaths to prioritising the young or those with many contacts. If there is very high vaccine supply, seniors are again favoured for prioritisation if aiming to reduce deaths, as there is sufficient coverage to achieve both direct protection of the most vulnerable and indirect protection of key transmitters. One study stated that direct effects of immunisation take precedence in deciding prioritisation when the vaccine supply is sufficient to cover the priority groups in the study which make up 18% of the population (key workers, individuals with comorbidities, and the over-60s) .
The influence of COVID-19 transmission rates was reported often in LMICs, but with varied results. One study in India suggested that when the transmission rate is low, those with comorbidities should be prioritised over those aged over 60 years old . In Brazil, modelling results suggested that the impact on deaths of vaccinating the young increases with earlier vaccination dates, lower vaccine efficacy, and higher transmission rates . In Columbia, the presence of the Delta variant reduced the magnitude of difference in the impact of vaccinating different groups. However, in all papers, the base-case result was to prioritise seniors .
Several studies tested different values of vaccine efficacy, with most reporting before full results of vaccine trials became available starting in late 2020. Generally, variations in vaccine efficacy did not appear to change prioritisation unless efficacy was significantly lower in older rather than younger populations. However, a number of studies assumed that vaccines had similar levels of efficacy against severe disease, infection, or transmission. Where vaccines were more efficacious against severe disease strategies, the priority was to vaccinate highest transmitters.