Transmission dynamics of the 2016-18 outbreak of hepatitis A among men who have sex with men in England and cost-effectiveness analysis of vaccination strategies to prevent future outbreaks

Summary Background Despite being vaccine-preventable, hepatitis A virus (HAV) outbreaks occur among men who have sex with men (MSM). We modelled the cost-effectiveness of vaccination strategies to prevent future outbreaks. Methods A HAV transmission model was calibrated to HAV outbreak data for MSM in England over 2016-2018, producing estimates for the basic reproduction number (R0) and immunity levels (seroprevalence) post-outbreak. For a hypothetical outbreak in 2023 (same R0 and evolving immunity), the cost-effectiveness of pre-emptive (vaccination between outbreaks among MSM attending sexual health services (SHS)) and reactive (vaccination during outbreak among MSM attending SHS and primary care) vaccination strategies were modelled. Effectiveness in quality-adjusted life-years (QALYs) and costs were estimated (2017 UK pounds) from a societal perspective (10-year time horizon; 3% discount rate). The incremental cost-effectiveness ratio (ICER) was estimated. Findings R0 for the 2016-2018 outbreak was 3·19 (95% credibility interval (95%CrI) 2·87-3·46); seroprevalence among MSM increased to 70·4% (95%CrI 67·3-72·8%) post-outbreak. For our hypothetical HAV outbreak in 2023, pre-emptively vaccinating MSM over the preceding five-years was cost-saving (compared to no vaccination) if the yearly vaccine coverage rate among MSM attending SHS was <9·1%. Reactive vaccination was also cost-saving compared to no vaccination, but was dominated by pre-emptive vaccination if the yearly vaccination rate was >8%. If the pre-emptive yearly vaccination rate fell below this threshold, it became cost-saving to add reactive vaccination to pre-emptive vaccination. Interpretation Although highly transmissible, existing immunity limited the recent HAV outbreak among MSM in England. Pre-emptive vaccination between outbreaks, with reactive vaccination if indicated, is the best strategy for limiting future HAV outbreaks. Funding NIHR.


Summary
Background Despite being vaccine-preventable, hepatitis A virus (HAV) outbreaks occur among men who have sex with men (MSM). We modelled the cost-effectiveness of vaccination strategies to prevent future outbreaks.
Methods A HAV transmission model was calibrated to HAV outbreak data for MSM in England over 2016-2018, producing estimates for the basic reproduction number (R 0 ) and immunity levels (seroprevalence) post-outbreak. For a hypothetical outbreak in 2023 (same R 0 and evolving immunity), the cost-effectiveness of pre-emptive (vaccination between outbreaks among MSM attending sexual health services (SHS)) and reactive (vaccination during outbreak among MSM attending SHS and primary care) vaccination strategies were modelled. Effectiveness in qualityadjusted life-years (QALYs) and costs were estimated (2017 UK pounds) from a societal perspective (10-year time horizon; 3% discount rate). The incremental cost-effectiveness ratio (ICER) was estimated.
Findings R 0 for the 2016-2018 outbreak was 3¢19 (95% credibility interval (95%CrI) 2¢87-3¢46); seroprevalence among MSM increased to 70¢4% (95%CrI 67¢3-72¢8%) post-outbreak. For our hypothetical HAV outbreak in 2023, pre-emptively vaccinating MSM over the preceding five-years was cost-saving (compared to no vaccination) if the yearly vaccine coverage rate among MSM attending SHS was <9¢1%. Reactive vaccination was also cost-saving compared to no vaccination, but was dominated by pre-emptive vaccination if the yearly vaccination rate was >8%. If the pre-emptive yearly vaccination rate fell below this threshold, it became cost-saving to add reactive vaccination to preemptive vaccination.

Introduction
Hepatitis A virus (HAV) is transmitted through the faecal-oral route, with most adults experiencing symptomatic disease. Although most individuals clear the infection with resultant lifelong immunity, many require hospitalisation with death from fulminant hepatitis being a rare outcome. 1,2 Effective vaccines can prevent HAV infection.
In high-income countries (HIC) where the risk of HAV infection from contaminated food or water is low, periodic outbreaks of HAV occur among men who have sex with men (MSM). 3 These outbreaks can be prolonged and cause substantial morbidity. The most effective way to control these outbreaks is through immunisation.
However, many HIC do not have targeted vaccination programmes because of concerns that it may not be cost-effective, particularly if they already provide universal childhood vaccination for HAV. 4 In the UK there is no universal childhood vaccination, but selective vaccination is recommended for those at risk, such as travellers to endemic countries and MSM. 5 Following recent outbreaks of HAV among MSM, 6,7 HAV vaccination has been recommended for MSM attending sexual health services (SHS) across England. However, implementation of this recommendation has been variable 8 and uptake has not been routinely monitored. As a result, there was uncertainty around the susceptibility of MSM to HAV prior to the outbreak that mainly affected MSM across England in 2016-2018. 9 This outbreak was initiated by importations of HAV among MSM returning from European countries. 9 The main outbreak response was to increase HAV vaccination among MSM. Public Health England (PHE) -now UK Health Security Agency (UKHSA) -recommended that all MSM attending SHS should have at least one dose of HAV vaccine. However, immunisation was initially limited by vaccine supply constraints and local commissioning challenges. PHE also advised that contacts of cases should be vaccinated in primary care. 9 The direct healthcare costs of the outbreak were estimated to be £1¢5 million, primarily driven by hospitalisations. 9 This estimation did not incorporate broader costs of the public health response, including health protection, awareness-raising, stakeholder engagement and incident response support.
Given this outbreak's health burden and costs, it is important to evaluate the cost-effectiveness of different HAV vaccination strategies for preventing future outbreaks. For England, we use an HAV transmission model to compare the cost-effectiveness of either vaccinating MSM before an outbreak occurs (pre-emptive vaccination) or only vaccinating MSM once an outbreak has started (reactive vaccination).

Methods
In this paper, we firstly develop and use a HAV outbreak model with data from the 2016/18 outbreak (and other sources) to estimate the transmission potential and existing levels of immunity for HAV among MSM in England. This model is then adapted to simulate a hypothetical outbreak in 2023 to assess the impact and cost-effectiveness of different vaccine strategies to prevent it, either reactive, based on levels of vaccination achieved in the 2016/18 outbreak, or a hypothetical pre-emptive vaccination scenario with different yearly rates of vaccination.
Outbreak model description

Research in context
Evidence before this study Hepatitis A outbreaks in high-income countries occur periodically, particularly among men who have sex with men (MSM) subpopulations. These outbreaks can be prolonged and cause substantial morbidity. The most effective way to control these outbreaks is through immunisation. We did a search on PubMed on 25 th Jan 2022 for economic evaluations of hepatitis A vaccinations using the terms "Hepatitis A" AND "economic evaluation" OR "Cost-effectiveness" OR "Cost-benefit". We found several economic evaluations for childhood vaccination and for travellers demonstrating mixed results regarding its cost-effectiveness. Two studies evaluated Hepatitis A vaccination in the US. One study reported that Hepatitis A and B vaccination among individuals attending STI clinics could be cost-effective, although they did not distinguish MSM from non-MSM, and another study reported that Hepatitis A vaccination would be cost-effective among MSM in USA. Currently there is no consensus on the best Hepatitis A vaccination strategy to prevent future outbreaks among MSM.

Added value of this study
Using data from the most recent Hepatitis A outbreak in the United Kingdom (2016-2018), we created a transmission dynamic model to examine the cost-effectiveness of various Hepatitis A vaccination strategies. We show that reactive vaccination of MSM during future outbreaks can be a cost-saving strategy for substantially reducing its magnitude. However, pre-emptively vaccinating MSM between outbreaks in sexual health services can save more money and have more impact if the pre-emptive vaccination rate is sufficiently high (9% vaccination rate per year among MSM attending Sexual Health Services if done for the 5 years preceding an outbreak), suggesting that this should be the best first choice.

Implications of all the available evidence
Maintaining sufficient background immunity (at least 70%) can prevent large Hepatitis A outbreaks. This immunity is best maintained through a pre-emptive vaccination strategy occurring between outbreaks, and then reactive vaccination strategies if the pre-emptive vaccination coverage is insufficient to prevent propagation of an outbreak. force of infection λ, they are exposed to HAV. When exposed they become latently infected but not yet infectious, after a latent period of L, they then become occultly infectious, with a proportion (p s ) of individuals after an infectious period d 1 then becoming symptomatic and the rest remaining asymptomatic but infectious. They then recover after d D days for symptomatic and d 2 days for asymptomatic individuals and become immune to HAV. For the model assessing vaccination strategies (panel B), among occultly infectious individuals, there is a proportion p F developing fulminant hepatitis and a proportion p L requiring liver transplant in addition to those becoming symptomatic and asymptomatic individuals. Individuals start as susceptible to infection. When exposed, they become latently infected but not yet infectious. They then become occultly infectious, with individuals either becoming symptomatic or remaining asymptomatic but infectious. A proportion of symptomatic cases are admitted to hospital. Individuals recover and become immune to HAV. The rare possibility of developing fulminant hepatitis is not included in this model for the 2016/18 outbreak.
The rate of infection is proportional to the number currently infectious and the rate of sexual contacts among MSM. However, we assume that symptomatic men are too sick to have sexual contacts 10 and so are not infectious, but relax this assumption in the sensitivity analysis. Occult and asymptomatic infections are assumed to have the same transmissibility.
The model stratifies MSM into low and high sexual risk groups (defined by the frequency of new sexual partners among MSM in the last year) and by whether they attend SHS or not. Rates of sexual contact across these four groups are defined by a contact matrix that determines their infection rate.
The outbreak responses of immunising (i) all MSM attending SHS and (ii) close contacts of HAV cases in primary care were defined as two vaccination rates in the model. However, because HAV immune status is generally unknown, these vaccinations only protect MSM susceptible to HAV, with individuals assumed to be protected immediately after vaccination. 5,11 As the 2016/18 outbreak was about two years, we ignored inflows and outflows of MSM when modelling the outbreak.

Parameterisation for the 2016/18 outbreak
Previous studies were used to obtain fixed values for HAV natural history parameters and uncertainty distributions for other model parameters (Table 1). This included the duration of different disease stages. We used the occurrence of jaundice to categorise symptomatic infection, with data 14 suggesting 84¢1% of infected adults experience jaundice. European data suggests a wide range for susceptibility to HAV, 17 so we assumed 20-70% for all MSM sub-groups.
Natural HAV infection confers lifelong immunity. However, the duration of immunity after vaccination is 5-7 years after one dose and about 25 years after two doses. 15,16 We assumed that immunisation provides 100% efficacy against infection, 5 with two doses effectively giving life-long protection. We used monthly data on vaccinations via SHS clinics, which started in July 2017 (Table S2), to parameterise the rate of reactive vaccination of MSM attending SHS, while the rate of vaccination of contacts in primary care used data on the number of vaccines done per case (SI.1.2). The estimated MSM population size for England was 531,559. 12 Data from the European MSM Internet Survey (EMIS), which was undertaken in 22 European countries in 2010, including 18,000 MSM in the UK was used to define the sexual behaviour of MSM and the proportion attending SHS. 18 The HAV transmission coefficient was estimated during model calibration. Additionally, because outbreak data suggests a downturn in the number of new cases after April 2017, we allowed a reduction in the contact rate after a specific date. Both parameters were estimated through model calibration (SI.1.3, SI.1.4).

Calibration to 2016/18 outbreak
We used a Bayesian Markov Chain Monte Carlo (MCMC) framework to calibrate the model to data on the weekly number of cases during the outbreak and the seroprevalence post-outbreak. This included data from 725 male outbreak cases between 17 July 2016 to 30 May 2018, which were all assumed to be MSM despite fewer reported as such (n=507). For seroprevalence, pooled data from seven surveys estimated 65¢8% (SD=15¢0%) of MSM being antibody positive in England in 2018 (Table S3)

Vaccination strategies for future outbreaks
We considered the impact and cost-effectiveness of two vaccination strategies on a hypothetical modelled HAV outbreak in 2023: pre-emptive vaccination before the outbreak and reactive vaccination during the outbreak.
Pre-emptive vaccination (see SI.2.1). We simulated different hypothetical levels of pre-emptive vaccination for SHS attendees during a 5-year non-outbreak period of 2018-2023. Although initial testing for immunity was considered, the cheapest strategy involved vaccinating without testing and so is considered here. We assumed a 50% return rate for second vaccination visits based on variable return rates (30-95%) achieved for other vaccines. 21,22 We estimated the vaccinations required to achieve different increases in immunity/seroprevalence amongst SHS attendees over 2018-2023, which were then used in our model to simulate their effect on the modelled HAV outbreak.
Reactive vaccination (see SI.2.2). We assumed that during the HAV outbreak, pre-emptive vaccination would be replaced with reactive vaccination initiatives in SHS and primary care. Reactive vaccination is assumed to be responsive to the ongoing number of cases, based on the responsive vaccination rate during Articles the 2016/18 outbreak and starting when the outbreak was signalled. We assumed the daily number of vaccinations depended on the previous days' number of cases, with the vaccines per case being 34¢41 via SHS clinics and 3¢31 via primary care.
For each vaccination scenario, we estimated its impact on the hypothetical HAV outbreak initiated in 2023 with 39 imported cases over 8 months (similar to 2016-18 outbreak). The model assumed the median basic reproductive rate (R 0 ) for the 2016-18 outbreak, no reduction in contact rate during the outbreak, and the pooled seroprevalence/immunity estimate (65¢8% ;  Table S3) following the end of the 2016/18 outbreak. The adapted model also included inflow and outflow of individuals, and so the immunity level decreased over 2018-2023 without pre-emptive vaccination. Following discussions with PHE, it was decided that an outbreak would be signalled if the number of cases exceeded 30 in a 3-month window, with the start date being the end of this period. Outcomes for the outbreak were then calculated up to when the case rate decreased to ≤1 per month. No uncertainty was assumed in the hypothetical outbreak or the parameters used to model the vaccination scenarios.

Estimating health benefits and costs
The outputs for each modelled outbreak in 2023 incorporating different vaccination scenarios were used to estimate the costs from a societal perspective for that vaccination strategy, public health response (PHR), health care costs, productivity losses, and qualityadjusted life years (QALY) gained. To make these costs and QALYs comparable across different vaccination scenarios, they were always estimated over 10 years, with costs reported in 2017 UK pounds (GB£) and a 3% annual discount rate applied to all costs and QALYs. A model adaptation was included ( Figure 1B) to incorporate the possibility of fulminant hepatitis or liver transplant. 15 No uncertainty was incorporated into these model projections.

Health benefits
To estimate QALYs, we incorporated utility weights for asymptomatic (0¢83), symptomatic (0¢64), fulminant (0¢26) and post-liver transplant (0¢73) cases, 15,23 with other health states assumed to have a utility weight of 0¢90 (averaged over three age ranges from 15 ). See the supplementary materials for more details. The total QALYs were calculated for the outbreak by applying these utility weights to individuals in each model category over time.

Productivity losses
We estimated the productivity losses due to HAV. Estimates of the weekly wage for adult males (£573) and their employment rate (80¢4%) 24 for England in 2020 were applied to the estimated days of absenteeism 15

Costs of pre-emptive vaccination
The pre-emptive vaccination scenarios involve costs for staff time and vaccines used, with these calculations assuming that 25% of the cost for the first SHS clinic visit was allotted to vaccination (SI.2.1).

Costs of outbreak and reactive vaccination
Outbreaks of HAV incur costs for the PHR and healthcare management of cases. These costs were estimated from the 2016/18 outbreak, with the PHR including costs for reactive vaccination initiatives, staff time for incidents of different magnitude, preparing and coordinating meetings and developing health promotional materials. Cost estimates are in Table S6, with more details in SI.3.

Cost-effectiveness of vaccination strategies
We estimated the incremental cost-effectiveness ratio (ICER) 23 for different pre-emptive and reactive vaccination initiatives, and both strategies combined. We used the NICE recommended willingness-to-pay threshold of £20,000 per QALY gained to determine if an intervention was cost-effective. 25

Sensitivity analyses
We

Role of the funding source
The funders had no role in the study design, data collection, data analysis, interpretation, writing of the report.
Vaccination required for controlling outbreaks.
For our modelled outbreak in 2023, if no pre-emptive vaccination occurred over 2018-2023, the seroprevalence among MSM would reduce from 65¢8% to 64¢2% by 2023. Additionally, if there was no reactive vaccination or reduction in contact rate during the outbreak, then our hypothetical outbreak from 2023 becomes endemic and oscillates, resulting in 52,951 cases over 10 years (Figure 3, Figure S3, Table 2).

Pre-emptive vaccination strategies
For this modelled outbreak initiating in 2023, our results in Table 2 show that incorporating pre-emptive vaccination would be cost-saving if it increased the seroprevalence among MSM attending SHS by ≤9¢0% in absolute terms over 5 years. This occurs with a yearly pre-outbreak vaccination coverage rate among SHS attendees of ≤9¢1% over 5 years. For instance, an annual vaccination coverage rate of 9% shortens the modelled outbreak to 1054 days with 571 cases (Table 2), which     CS denotes cost saving where the option is cheaper compared to the comparator and QALYs are gained. # The outbreak is periodic (see Figure 3) for all these vaccine scenarios − they either undertake pre-emptive vaccination and increase the seroprevalence of MSM attending SHS by 1 to 8% (over 5 years) or they undertake reactive vaccination with pre-emptive vaccination and increase the seroprevalence of MSM who attend SHS by 1 to 6% (over 5 years). Estimates listed are for the first 10 years. Without pre-emptive vaccination, the seroprevalence or immunity level reduces from 65¢8% to 64¢2% at the end of a 5-year period. The incoming outbreak is assumed to be induced by importation of infections as for the 2016/18 outbreak.
saves 2283 QALYs (compared to no control measures) over 10 years and reduces the total costs (outbreak plus vaccination costs) from £120 million to £11 million. Vaccination strategies that increase the seroprevalence further become less cost-effective because they go beyond the herd immunity threshold (1-1/R 0 = 68¢7%). Additional model projections suggest that higher rates of pre-emptive vaccination become cost-effective if the initial seroprevalence was lower (Table S7).

Reactive vaccination strategies
For the same modelled outbreak in 2023, adding reactive vaccination in SHS clinics and primary care once the outbreak starts shortens the outbreak to 1597 days with 3770 cases (Figure 3 and Table 2). This initiative saves 2139 QALYs over 10 years compared to having no outbreak control measures and reduces the total outbreak cost to £14 million. Comparing to pre-emptive vaccination (Table 2), our results suggest that if the preemptive vaccination increases seroprevalence among SHS attendees by 9% in absolute terms over 5 years then reactive vaccination saves less costs than pre-emptive vaccination and gains less QALYs (Table 2), whereas this is reversed if the pre-emptive vaccination increases seroprevalence among SHS attendees by ≤7% over 5 years. This indicates that pre-emptive vaccination is the best first choice if it achieves sufficient levels of vaccination, otherwise reactive vaccination is better.

Combined vaccination strategies
Our analyses of combined vaccination strategies show that it is cost-saving to add reactive vaccination to preemptive vaccination if the pre-emptive initiative does not increase the seroprevalence among MSM attending SHS by over 8¢0% in absolute terms over 5 years (Table 2). For instance, if the pre-emptive vaccination initiative increases seroprevalence among SHS attendees by 7% over 5 years, then adding reactive vaccination would further reduce the total outbreak costs by £48 million and save 793 QALYs. This suggests that reactive vaccination should be added to pre-emptive vaccination initiatives if the pre-emptive initiative has not increased seroprevalence by such an extent that reactive vaccination achieves little additional impact (Table 2).

Sensitivity analysis
The effects of our sensitivity analyses on the cost-effectiveness of vaccination initiatives are generally small ( . Reactive vaccination also remains cost-saving across all sensitivity analyses, with the cost-savings and/or QALYs gained of reactive vaccination only varying by more than 20% in relative terms if mildly symptomatic infected MSM transmit HAV, utilities for HAV are decreased by 10%, the contact rate decreases during the outbreak, or if immunity does not decrease between the outbreaks. Lastly, for a pre-emptive vaccination initiative that increases seroprevalence among SHS attendees by 7% over 5 years, adding reactive vaccination remains costsaving (Table 3) for all sensitivity analyses except if the contact rate decreases during the outbreak or immunity remained steady between the outbreaks without preemptive vaccination. This again changes if the pre-emptive vaccination initiative increases seroprevalence among SHS attendees by 9% in absolute terms over 5 years, whereupon adding reactive vaccination is no longer cost-effective unless mildly symptomatic MSM still transmit HAV or both vaccination strategies only targeted high-risk MSM (Table S9). Interestingly, increasing the time between outbreaks from 5 to 10 years does increase the overall levels of vaccination needed but has little effect on the cost-effectiveness results (Table S14).
The sensitivity analysis where we assumed a lower number of cases in the 2016/18 outbreak did not result in major changes to the resulting parameters for the 2023 outbreak model (Table S8) and so was not considered further.

Discussion
Our modelling of the HAV outbreak among MSM in England in 2016/18 suggested that despite HAV being highly transmissible (R 0 =3¢19), the outbreak was limited by existing high levels of immunity (69¢1%) and a modest reduction (16¢9%) in contact rates during the outbreak. Going forward, our modelling of the hypothetical outbreak in 2023 show that reactive vaccination of MSM during future outbreaks can be a cost-saving strategy for substantially reducing its magnitude. However, pre-emptive vaccination of MSM in sexual health services (SHS) is likely to be more cost-effective if the vaccination rate is sufficiently high, suggesting that this is the optimal first choice. Importantly, it can also be cost-effective to add reactive vaccination to pre-emptive vaccination initiatives if the coverage of pre-emptive vaccination is not too high and MSM contact rates do not  Table 3: Sensitivity analysis on the costs (in millions of GB£), QALYs and incremental cost-effectiveness ratio (ICER in cost per QALY saved) of pre-emptive (PV) or reactive vaccination (RV) strategies, solely or in combination. The ICERs of the single interventions are compared to a counterfactual of no vaccination, while the combined scenario considers the ICER of adding reactive vaccination to the pre-emptive vaccination initiative that increases seroprevalence among SHS attendees by 7% over 5 years.
CS denotes cost saving where the option is cheaper compared to the comparator and QALYs are gained. KPC is the daily number of reactive vaccinations given in primary care per infection on the previous day and KSHS is the daily number of reactive vaccinations given in sexual health services per infection on the previous day. * The baseline scenario is defined by: an outbreak start criterion of >=30 cases reported within 3 month and ends when there is <=1 case within one month; the gap between outbreaks is 5 years; immunity decreases between outbreaks due to inflow of susceptible and outflow of immune MSM; mildly symptomatic MSM are not sexually active; weak assortative mixing (b=3¢5%); equal initial immunity levels among 4 MSM sub-groups (high and low risk and SHS attendance or not); the utility weights are 0¢83 for asymptomatic, 0¢64 for symptomatic, 0¢26 for fulminant, 0¢73 for post liver transplant cases, and 0¢90 for other health states. Baseline vaccination is to all SHS attendees equally with RV starting immediately once the outbreak starts (if RV is done) and no change in contact rate during the outbreak. Effect of first vaccine dose is assumed to last 7 years and 50% return for second dose.
decrease during the outbreak due to public health messaging. During the 2016/18 HAV outbreak, Public Health England (PHE) recommended reactive vaccination of MSM through SHS. 11 Concurrently, PHE also raised awareness of the outbreak and followed up cases and their close contacts to minimise the risk of transmission by providing hygiene advice and post-exposure vaccination. 9 Our modelling suggests these public health measures may have reduced contact rates among MSM, reducing the outbreak by a third and being the main cause for its downturn. The reactive vaccinations had much less impact because they started late in the outbreak (after July 2017) due to a national vaccine shortage and other challenges. This contrasts with our projections going forward that show reactive vaccination could have considerable impact if initiated early in an outbreak. This emphasises the importance of early recognition of outbreaks and having vaccine supply and delivery systems in place to respond rapidly. 26 Our estimates of R 0 for HAV among MSM in England (R 0 =3¢19) are in agreement with estimates from persons experiencing homelessness or who use drugs, 26 larger but comparable to estimates from MSM elsewhere (R0=1¢4-3¢1), 10,27,28 and much larger than estimates for other groups, 29,30 emphasising the increased HAV risk in MSM. There are limited studies of the costeffectiveness of HAV vaccination among MSM. Our findings are consistent with a US study reporting that Hepatitis A and B vaccination of individuals attending STI clinics is cost-effective, although they did not distinguish MSM from non-MSM, 31 and another study reporting that HAV vaccination is cost-effective among MSM in USA. 32 Although limited, this evidence agrees with our study that targeted HAV vaccination of individuals with heightened sexual risk is value for money.

Strengths and limitations
Strengths of our analysis include calibrating our model to a recent large and prolonged HAV outbreak among MSM. This provided unique and rich information about the transmissibility of HAV among MSM in England and gave insights into why the outbreak died out. We also stratified MSM by levels of sexual risk behaviour and their attendance of SHS, which allowed us to assess the impact of vaccinating specific MSM subpopulations.
Limitations include uncertainty in numerous model parameters. Firstly, we assumed the same baseline seroprevalence among all MSM sub-groups. Although, it is likely that high-risk MSM and MSM attending SHS may have greater immunity due to increased chances of being infected or vaccinated, our sensitivity analyses suggest this should not have changed our conclusions. Additionally, we assumed the seroprevalence would decrease over time following the 2016/18 outbreak due to population turnover, and that the next outbreak would occur in 2023, which is highly uncertain. Although future outbreaks will tend to be bigger if there is a longer time gap, with vaccination having greater impact, it did not significantly affect the cost-effectiveness. Indeed, vaccination would remain cost-saving even if seroprevalence remained stable between the outbreaks. We also did not incorporate vaccination of MSM travelling to high-risk countries or for other reasons (»60-70,000 vaccinations a month through primary care in 2016, NHS data), 5,33 or immigration of MSM born in endemic countries who are likely to have preexisting immunity, which may result in seroprevalence declining less between outbreaks. We also did not consider the potential emergence of vaccine-escape variants, which may possibly occur if some MSM receive post exposure-vaccination, 34 because there is little data on its clinical or public health significance.
Secondly, we did not include age structure in our model, which can be associated with HAV seroprevalence, exposure risk and clinical presentation. 14 Our modelling shows that targeting pre-emptive vaccination to high-risk MSM attending SHS clinics reduces the cost savings and QALYs gained compared to vaccinating all SHS attendees, suggesting that prioritising younger MSM may have no added benefit.
Thirdly, there were uncertainties around behavioural parameters such as levels of assortative mixing and whether mildly symptomatic MSM are sexually active. Our sensitivity analyses suggest these uncertainties do not affect our findings.

Implications
Our study has implications for preventing HAV outbreaks. We found that although the transmissibility of HAV is high among MSM in England, existing high levels of immunity have prevented large HAV outbreaks. This immunity is best maintained in the future through pre-emptive vaccination strategies, as recommended by UKHSA, 5 and then possibly reactive vaccination strategies if the pre-emptive vaccination coverage is insufficient to prevent an outbreak. Although our analyses suggest this can be achieved through solely vaccinating in sexual health services, we would also recommend that MSM should be vaccinated in other settings to ensure high coverage is achieved. Going forward, robust surveillance is needed to ensure that reactive vaccination is initiated promptly when an outbreak starts, but is not initiated due to natural variation in cases. Acting quickly requires a stable supply of vaccines, pre-agreed commissioning arrangements, and logistics and staff to rapidly scale-up. Although this can be challenging, our vaccine response to COVID-19 shows that it is possible. Lastly, improved monitoring of HAV seroprevalence among MSM is also needed to ensure that immunity levels are not decreasing so increasing the likelihood of future outbreaks.

Contributors
XSZ, SM, PV, and JJO conceptualized and developed the analysis methodology. XSZ developed the model and performed all model analyses. All authors undertook data analyses for parameterising and calibrating the model. PV, JJO, ME and SM supervised the overall analysis, with PV, LM and JJO supervising the modelling and economic evaluation. All authors reviewed the analysis methodology and model results. XSZ and PV wrote the original draft of the manuscript. All authors reviewed and edited the manuscript.

Data sharing statement
The model code and projections for this paper will be shared with interested parties upon reasonable request, which will be decided by Peter Vickerman and Xu-Sheng Zhang.

Declaration of interests
PV has received unrestricted research grants from Gilead not related to the submitted work. This research was funded in whole, or in part, by the National Institute for Health Research Health Protection Research Unit for Behavioural Science and Evaluation and the Wellcome Trust [WT 220866/Z/20/Z].