Impact of proactive and reactive vaccination strategies for health-care workers against MERS-CoV: a mathematical modelling study

Background Several vaccine candidates are in development against MERS-CoV, which remains a major public health concern. In anticipation of available MERS-CoV vaccines, we examine strategies for their optimal deployment among health-care workers. Methods Using data from the 2013–14 Saudi Arabia epidemic, we use a counterfactual analysis on inferred transmission trees (who-infected-whom analysis) to assess the potential impact of vaccination campaigns targeting health-care workers, as quantified by the proportion of cases or deaths averted. We investigate the conditions under which proactive campaigns (ie vaccinating in anticipation of the next outbreak) would outperform reactive campaigns (ie vaccinating in response to an unfolding outbreak), considering vaccine efficacy, duration of vaccine protection, effectiveness of animal reservoir control measures, wait (time between vaccination and next outbreak, for proactive campaigns), reaction time (for reactive campaigns), and spatial level (hospital, regional, or national, for reactive campaigns). We also examine the relative efficiency (cases averted per thousand doses) of different strategies. Findings The spatial scale of reactive campaigns is crucial. Proactive campaigns outperform campaigns that vaccinate health-care workers in response to outbreaks at their hospital, unless vaccine efficacy has waned significantly. However, reactive campaigns at the regional or national levels consistently outperform proactive campaigns, regardless of vaccine efficacy. When considering the number of cases averted per vaccine dose administered, the rank order is reversed: hospital-level reactive campaigns are most efficient, followed by regional-level reactive campaigns, with national-level and proactive campaigns being least efficient. If the number of cases required to trigger reactive vaccination increases, the performance of hospital-level campaigns is greatly reduced; the impact of regional-level campaigns is variable, but that of national-level campaigns is preserved unless triggers have high thresholds. Interpretation Substantial reduction of MERS-CoV morbidity and mortality is possible when vaccinating only health-care workers, underlining the need for countries at risk of outbreaks to stockpile vaccines when available. Funding UK Medical Research Council, UK National Institute for Health Research, UK Research and Innovation, UK Academy of Medical Sciences, The Novo Nordisk Foundation, The Schmidt Foundation, and Investissement d'Avenir France.


Introduction
SARS-CoV-2 has demonstrated the threat of novel coronaviruses, the need for effective vaccines, and the challenges in vaccine deployment. First identified in Saudi Arabia in 2012, 1 MERS-CoV, the virus that causes Middle East respiratory syndrome (MERS), remains a major public health concern, with 2591 laboratoryconfirmed cases and 894 deaths reported in 27 countries as of August, 2022. 2 The infection fatality ratio based on laboratory-confirmed cases only is 35%, although the true ratio is lower. 3 Human-to-human transmission occurs primarily nosocomially but is otherwise uncommon, 4 and dromedary camels (Camelus dromedarius) constitute the animal reservoir. 5 However, the threat of major outbreaks from human-to-human transmission should not be underestimated: a single introduction of the virus to South Korea caused 186 cases and 38 deaths in May-July, 2015. 6 No vaccines against MERS-CoV are yet licensed, although several are in development. [7][8][9][10] The most advanced are the ChAdOx1 MERS recombinant viral vector vaccine (developed by University of Oxford and Janssen Vaccines), 11 which phase 1b trials have shown to be safe and to elicit antibody and T-cell immune responses in humans, 12 and the INO-4700 nucleic acid vaccine (Inovio Pharmaceuticals), which is currently in phase 2 trials. 13 In this study, we build on previous transmission tree inference 5 (who-infected-whom analysis) that estimated the MERS-CoV reproduction number, serial interval, and contributions of the animal reservoir and human-to-human transmission to the 2013-14 epidemic in Saudi Arabia. This dataset is a detailed, individual-level, line list of a well described epidemic, allowing us to address several questions about the potential impact of a viable vaccine: could vaccinating health-care workers substantially reduce overall morbidity and mortality; would the highest-impact strategy to limit cases and deaths depend on vaccine efficacy or its duration of protection; if a vaccine does not maintain its efficacy permanently, would reactive vaccination campaigns, whereby vaccinations respond to an unfolding outbreak, reduce case numbers more than prophylactic proactive campaigns that might occur years before the next outbreak; and to what extent would control measures targeting the animal reservoir mitigate an epidemic? We only consider vaccination of health-care workers because nosocomial transmission increases their risk and patients' risk, and they can feasibly be identified and vaccinated.

Data
We use a line list 5 including day of symptom onset, hospital, region, health-care worker status, and clinical outcome of 681 individuals infected with MERS-CoV during the 2013-14 outbreak in Saudi Arabia. Cases with symptom onset between Jan 1, 2013, and July 31, 2014, are included. These data were previously used 5 to probabilistically infer transmission trees, which we prune to evaluate a given counterfactual strategy, considering what would have happened had a MERS-CoV vaccine been available, by deleting cases (and their secondary cases) and then recording the outcome. Health-care worker and capacity data by region were obtained from Saudi Arabian Government statistics 14 (appendix p 8).

Transmission tree inference
Inferred transmission trees preserve spatial properties of the epidemic, quantifying nosocomial, within-region, between-region, and animal reservoir transmission. The tree inference procedure has been described previously, 5 but is reproduced in the appendix (pp 1-4) for completeness. We used Markov chain Monte Carlo sampling for parameter inference and data augmentation for infectors. Each posterior sample includes, in addition to transmission parameters, a transmission tree. 55 000 iterations were done with a 5000 iteration burn-in, thinning every five iterations, giving 10 000 posterior samples per model run. The model was coded in C++

Research in context
Evidence before this study There are no vaccines against MERS-CoV, which currently has lower transmissibility but considerably higher lethality than SARS-CoV-2, and so poses a continuing threat to public health. However, several candidate vaccines are in development. The most advanced are the ChAdOx1 MERS recombinant viral vector vaccine (developed by University of Oxford and Janssen Vaccines), which phase 1b trials have shown to elicit both antibody and T-cell immune responses in humans, and the INO-4700 nucleic acid vaccine (Inovio Pharmaceuticals), which is currently undergoing phase 2 trials. Although there has been substantial work developing candidate vaccines and reporting on their progress, to our knowledge, there are no studies examining the potential population-level impact of MERS-CoV vaccines or investigating strategies for their deployment. Human-to-human transmission is primarily nosocomial, meaning that health-care workers are at high risk of infection, and importantly constitute a group that could be targeted and vaccinated relatively easily and cost-effectively.

Added value of this study
We investigate the potential effect of vaccination of health-care workers on MERS-CoV outbreak size. Using data from the 2013-14 Saudi Arabia outbreak, we infer transmission trees consistent with the spatiotemporal dynamics of that outbreak to evaluate the potential relative performance of a variety of health-care worker vaccination strategies using the proportions of cases and deaths averted as our outcome metrics. We find that reactive vaccination campaigns, initiated in response to an unfolding outbreak, outperform proactive vaccination campaigns, unless waning of vaccine efficacy is low, which recent experience of SARS-CoV-2 suggests is unlikely. The spatial scale at which a reactive vaccination campaign is implemented is key. If each hospital reacts only to its own hospital-level outbreak, the effect is greatly curtailed, although this is the most efficient strategy (as measured by cases averted per thousand doses). Conversely, reacting to an outbreak at the regional or national level gives the largest overall reductions in mortality and morbidity (almost regardless of vaccine efficacy), but these strategies are less efficient. Increasing the threshold required before a reactive campaign is launched affects the proportions of cases and deaths averted according to spatial level: nationallevel impact is largely preserved, whereas hospital-level impact is essentially nullified. Our analysis shows that, even when vaccinating health-care workers only, it is possible to achieve substantial case reductions that are disproportionately large compared with the number of health-care workers as a result of downstream cases being successfully averted.

Implications of all the available evidence
Large-scale MERS-CoV outbreaks are rare and vaccine efficacy and duration of protection will thus prove challenging to measure empirically with traditional clinical trials. Therefore, in preparation of the next large epidemic, it is important to have an evidence-based MERS-CoV vaccination policy that is consistent across values of vaccine efficacy and duration. Our analysis provides a quantitative and data-driven foundation for such policy. and R version 4.0.5, using the ggplot2 15 and igraph 16 packages. Priors are flat and fitted on a log scale. Parameter estimates, reproduced from a previous study, 5 are given in the appendix (p 9).

Counterfactual transmission trees with vaccination
We consider two categories of vaccine campaign: proactive and reactive. Under proactive campaigns, vaccination of health-care workers occurs before the next outbreak, so vaccinees gain at least some protection at its outset. However, it is impossible to predict when the next outbreak will occur, and efficacy might have waned when the vaccine is needed. Recent experience with SARS-CoV-2 vaccines [17][18][19][20][21][22] suggests that waning is probable. For a given strategy, the primary outcome measures are the proportions of cases and deaths averted in the population (and not only among health-care workers), which we decompose into direct and indirect protection. Simulated strategies and parameters are summarised in the table.
We also consider campaigns that react to an unfolding outbreak. Although initial cases are unprotected, suboptimal protection from a waning vaccine might outweigh slow reaction times. Alternatively, downstream effects could mean that stopping early cases, even with a vaccine of diminished efficacy, is more important.
The impact of a reactive campaign depends on vaccine efficacy and duration, speed of implementation, and delay to protection. We consider three spatial scales of reactive campaign: 1) health-care workers in each hospital are vaccinated in response to the first case (health-care worker or otherwise) in that hospital; 2) health-care workers in each region are vaccinated in response to the first case in that region; and 3) health-care workers nationally are vaccinated in response to the first case nationally.
Let case k have infector i(k) and symptom onset at time t k . We set i(k)=0 if case k was infected by the animal reservoir. We assume protection from disease protects against transmission, deleting all downstream cases. Let VE denote initial vaccine efficacy. Assuming exponential waning, for mean duration of protection D (referred to as simply duration), efficacy after t * years is VE*(t*)VEexp(-t*/D).
We alternatively consider, as a sensitivity analysis, waning using the sigmoidal Hill function: where Y is the half-life and a governs the rate of decline. This function declines more slowly until its half-life, after which decline accelerates. We set a=4 to allow reasonably smooth waning (appendix p 10). We consider initial vaccine efficacies of 0-100% in 5% intervals, and mean durations (or half-lives) of 1, 2, 5, 10, 15, and 20 years, as well as no vaccine waning. For proactive campaigns, we simulate waiting times between vaccination and the next outbreak (henceforth referred to as wait) of 6 months, and 1-10 years.
If S denotes the set of health-care workers and P c is the coverage achieved, then, under a proactive strategy, case k is vaccinated with probability P v,k =δ s (k)P c , where Case k is protected and deleted from the transmission tree with probability P v,k VE * (t * k ), where t * k is case k's time post vaccination. We assume full health-care worker coverage (P c =1), although results for reduced coverage can be obtained through scaling (eg, 45% vaccine efficacy with 100% coverage is equivalent to 90% vaccine efficacy with 50% coverage).
For reactive campaigns, let τ I denote the reaction time between the first case (in a hospital, region, or country, among health-care workers or otherwise) and vaccination. Let τ P denote the lag between vaccination and protection.
At hospital level, T 0,h is the time of symptom onset of hospital h's first case. At regional or national levels, T 0,h is the time of symptom onset of the first case in region h or the entire country, respectively. Case k is protected and deleted from the tree with probability δ P (k)P v,k VE * (t * k ), where We vary τ I from 0 to 28 days in 2-day intervals. ChAdOx1-MERS T-cell responses peaked at 14 days; while antibodies peaked at 28 days, titres were still high

Proactive strategy
Wait between vaccination and next outbreak 1-10 years in 1-year intervals, and 6 months

Reactive strategy
Spatial scale Hospital, regional, or national Reaction time (τ I ) 0-28 days in 2-day intervals between start of outbreak (hospital, regional, or national) and vaccination Time between vaccination and immunity (τ P ) 14 days (no immunity assumed for days 0-13 post vaccination)  12 Therefore we set τ P =14 days. The effect of longer or shorter values of τ P can be inferred from the analysis of sensitivity to reaction times, given the overall delay to vaccine-induced protection is the sum of the reaction time and the time to protection.
Modelling animal reservoir (ie, camel) control measures, if i(k)=0, we delete case k and their downstream cases with probability ζ, considering values between 10% and 50%, in 10% intervals.

Analysis of policy efficiency
As a step towards a cost-effectiveness analysis, we examine the relative efficiency of proactive and reactive strategies. Using Saudi Ministry of Health data 14 (appendix p 8) on regional health-care capacity (numbers of hospitals and health-care workers), we calculate MERS-CoV cases averted per dose (assuming a singledose regimen). We assume all hospitals within a given region have equal numbers of doctors, nurses, and midwives, who combined give the number of health-care workers.
We assume that proactive and national reactive campaigns require vaccination of health-care workers nationally upon detection of a single case, whereas regional-level and hospital-level campaigns require health-care workers within their region or hospital, respectively, to be vaccinated upon detection of a single case there. Using counterfactual transmission trees, we calculate the expected number of doses required as the probability of each hospital or region containing at least one case, multiplied by the number of health-care workers there, summed over all hospitals or regions.

Role of the funding source
The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results
Between Jan 1, 2013, and July 31, 2014, 681 cases of MERS were recorded, of which 534 (78%) were symptomatic and 276 (41%) were fatal. 187 (27%) cases were recorded in health-care workers (mean age 39 years [SD 12] in health-care workers vs 51 years [SD 21] in non-health-care workers), 15 (8%) of which were fatal, comprising only 5% of all 276 deaths. The case fatality ratio was 8% among health-care workers and 53% among non-healthcare workers. National, regional, and hospital epidemic curves are shown in figure 1A, figure 1B, and the appendix (pp [11][12]. Example transmission trees, showing within-hospital, between-hospital within-region, between-region, and reservoir transmission contributions, are shown in figure 1C-F. Counterfactual epidemic curves for several illustrative scenarios are provided in the appendix (p 13), along with discussion of the effect of interventions targeting only the animal reservoir (appendix pp 4, 31) and the change in transmission contributions by campaign (appendix pp 5-6, 33-34).
The impact of proactive vaccination campaigns targeting health-care workers depends on vaccine efficacy, duration of vaccine protection, and the wait to the next MERS-CoV outbreak (which cannot be known in advance). Without animal reservoir control measures, an optimistic scenario of 90% efficacy with a 20-year mean duration and a 6-month wait would avert 64% (95% credible interval 54-74) of cases. However, an 8-year wait would avert only 54% (41-67) of cases. The 2013-14 outbreak remains the largest of its scale; 23 therefore, the wait until the next large outbreak, when vaccination is most needed, is likely to be long, perhaps several years.
The proportions of cases and deaths averted with different values of vaccine efficacy, duration, wait, and reservoir control measure effectiveness are shown in figure 2 and the appendix (p 14). The impact of a proactive campaign, on both cases and deaths, increases with vaccine efficacy and duration, and decreases with wait. Wait is largely irrelevant if duration is long (eg, 20 years), whereas duration matters less for short waits, when success depends on vaccine efficacy alone. Low vaccine efficacies (≤25%) have little effect unless wait is short or duration is long, averting at most 31% (18-51) of cases and 23% (10-42) of deaths. Short durations (≤2 years) similarly have small effects on cases and deaths, unless waits are short (<1 year) and vaccine efficacy is moderate (≥50%). On average, the proportion of deaths averted by proactive campaigns is 0·68 that of cases (appendix p 14).
30% effective reservoir control measures improve the above optimistic scenario (90% vaccine efficacy, 20-year mean duration, and 6-month wait) to 75% (63-87) of cases and 66% (49-83) of deaths averted. 50% effective measures improve this further to 82% (69-92) of cases and 75% (57-90) of deaths (figure 2), with similar trends with respect to vaccine efficacy, duration, and wait. Lower uncertainty bounds (of 95% credible intervals) show diminished effect in averting cases and deaths, whereas upper bounds suggest that all combinations of vaccine efficacy, duration, and wait could practically eliminate the epidemic (appendix pp [15][16]. With reactive campaigns, assuming 90% vaccine efficacy, 20-year mean duration, and 0-day reaction time at the hospital level would avert 56% (46-66) of cases, illustrating the disproportionate effect of removing downstream cases, as only health-care workers (27% of cases) are vaccinated ( figure 3). Assuming a 14-day reaction time, the proportion of cases averted decreases to 50% (39-60), and for a 28-day reaction time to 48% (37-59). With 50% vaccine efficacy and 20-year mean duration, cases are reduced by 40% (27-54) with a 0-day reaction time, 36% (23-50) with a 14-day reaction time, and 34% (21-50) with a 28-day reaction time (figure 3). On average, the proportion of deaths averted by reactive campaigns is 0·75 that of cases (appendix p 17). Credible www.thelancet.com/lancetgh Vol 11 May 2023 e763 intervals are wide (appendix pp [18][19]. For reactive campaigns, impact on both cases and deaths are insensitive to reaction times between 10 and 28 days. The small reduction in effect for reaction times greater than 8 days is attributable to whether a particular case in a health-care worker in hospital 13 or region 7 (Aseer) in our dataset would be vaccinated in time (appendix pp 11-12). Compared with hospital-level campaigns, regionallevel campaigns have a greater effect on cases and deaths (figure 3, appendix p 17) and reaction times matter less. A 90% efficacious vaccine with 20-year mean duration deployed instantly averts 65% (56-75) of cases. 14-day and 28-day reaction times each avert 62% (54-72) of cases. Reacting at the national level offers little further improvement beyond regional-level campaigns (figure 3), although the impact is the same for 0-day, 14-day, or 28-day reaction times, reducing cases by 65% (56-75) and deaths by 51% (39-65).
The effect of reactive campaigns on cases and deaths increases with vaccine efficacy and duration, and decreases with reaction time ( figure 3, appendix p 17). However, excluding vaccine efficacy, the greatest factor is spatial scale: reacting at hospital level usually has less effect than reacting at regional level, and national level is slightly better still. Although short durations reduce the effect (eg, with 90% vaccine efficacy and 1-year duration,

Figure 1: MERS-CoV incidence and inferred transmission trees for the 2013-14 outbreak in Saudi Arabia (A-B) Weekly incidence of MERS-CoV cases (A) and deaths (B) among health-care workers and overall. (C-F) Inferred transmission trees.
Each tree is a posterior sample of the infectors of each case, including the animal reservoir, for a particular modelled scenario. Examples are shown for the following scenarios: no vaccination (C); a reactive campaign at the regional level with a 60% efficacious vaccine, 20-year mean duration of protection, and 14-day reaction time (D); a proactive campaign with a 60% efficacious vaccine, 20-year mean duration of protection, and 6-month wait before the next outbreak (E); and considering only control measures targeting transmission from the animal reservoir, with 50% effectiveness (F).  a hospital-level reaction within 14-days reduces the effect to almost half that of a 20-year duration), the effect is relatively robust to durations of 5 years or more ( figure 3). Notably, very short durations (1 year) combined with short reaction times change the spatial scale rank order: national level is worst, but regional level is best. 30% effective reservoir controls improve the maximum proportion of cases averted by a reactive campaign to 78% (68-88), whereas 50% effective reservoir controls avert 84% (74-93) of cases (appendix pp [20][21]. Differences between spatial scales vary by vaccine efficacy, duration, and reaction time (appendix pp 5, 32). When comparing the two campaign types, if vaccine efficacy remains constant, proactive campaigns outperform reactive campaigns. Otherwise, regional-level reactive campaigns with 28-day reaction times outperform proactive campaigns, unless vaccine duration is long (≥15 years) and the wait is short (≤1 year; figure 4). National-level reactive campaigns were superior in all scenarios considered. Proactive campaigns require less stringent conditions to outperform hospital-level reactive campaigns, although even here are inferior in approximately half our simulations, and less often if a shorter reaction time is assumed (figure 4; appendix p 22).
Trends hold almost irrespective of vaccine efficacy and are relatively insensitive to reservoir controls (appendix p 23), although these factors decrease the differences between proactive and reactive campaigns. Trends further hold for deaths (appendix p 24), although

Figure 4: Comparison of proactive and reactive strategies
Plots show the ratio of cases averted with proactive campaigns versus reactive campaigns (proactive divided by reactive), with varying values of duration of vaccine protection and wait between vaccination and next outbreak. Proactive campaigns are compared with reactive campaigns at the hospital, regional, and national levels. A 28-day reaction time and a 14-day lag between vaccination and immunity are assumed in all plots. Ratios less than 1 (left of black contour line) indicate that a reactive campaign averts more cases than a proactive campaign.
hospital-level and regional-level campaigns outperform proactive campaigns slightly less often. The relative performance of proactive and reactive campaigns is asymmetrical. Assuming 10% vaccine efficacy, proactive campaigns avert 44% more cases than hospital-level campaigns (34% assuming 90% vaccine efficacy). However, regional-level and national-level campaigns each avert many times more cases than proactive campaigns, albeit with small numbers being compared.
To analyse policy efficiency, we calculated the expected number of cases averted per 1000 doses required (appendix p 25). Hospital-level campaigns are most efficient (particularly for short reaction times), with little difference between the regional and national levels, which slightly outperform proactive campaigns, especially for long waits. Efficiency increases with vaccine efficacy, and is reasonably robust for durations of 5 years or more, but declines with reaction time, especially at the hospital level for shorter durations.
We decomposed predicted effects into direct protection (ie, cases averted because they were vaccinated) and indirect protection (cases averted because some healthcare workers were vaccinated, preventing their infection). Direct protection accounts for at most 40% of the cases averted, increases with residual vaccine efficacy (ie, the aggregate of vaccine efficacy and waning), decreases with reaction time (for reactive scenarios), and decreases with waning and wait (for proactive scenarios; appendix p 26). Direct protection is highest for regional-level and national-level campaigns and lower for proactive and hospital-level campaigns. The high indirect impact implies that health-care workers contribute substantially to overall infection.
We can also evaluate impact on cases when reacting using trigger thresholds for vaccination based on case numbers larger than the single-case threshold used above. At the hospital and national levels, the precise trigger definition is unimportant: impact on cases is greatly reduced for thresholds above one case for hospital-level campaigns, but is much less sensitive to higher thresholds for national-level campaigns. At the hospital level, even sensitive triggers (eg, five infections within 7 days) yield only 10-20% of the impact obtained when reacting to a single case. Nationally, unless triggers are very stringent (eg, ≥20 infections within 14 days or less), impact remains largely unchanged. At the regional level, triggers have variable effects: low thresholds over longer timeframes yield almost identical impacts, whereas higher thresholds over shorter timeframes drastically reduce impact. These effects are largely robust to vaccine efficacy and vaccine durations of 5 years or more, and show only mild sensitivity to reaction times (appendix p 27).
Lower thresholds over longer timeframes initiate vaccination at an earlier stage of the epidemic, while higher thresholds can result in no vaccination, especially at the hospital or regional levels (appendix pp 28-30).
The sensitivity of the results to the choice of waning function and vaccination timeframe is discussed in the appendix (pp 6-7, 35-40).

Discussion
No human vaccine against MERS-CoV has yet been licensed. When vaccines become available, maximising impact is non-trivial. In this study, we analysed vaccine campaign strategies as a function of vaccine efficacy and duration. Each strategy was evaluated by estimating then pruning transmission trees to generate counterfactual epidemics and determine the resulting reduction in cases and deaths. All strategies considered could, in principle, be implemented, and are therefore relevant for policymakers. We considered vaccination of health-care workers only, as they are easily vaccinated, most at-risk, and, therefore, cost-effective to target.
The fundamental difference between proactive and reactive campaigns is whether to vaccinate in anticipation of, or in response to, the next outbreak. Proactive campaigns depend on vaccine efficacy, duration, and wait. Reactive campaigns depend on vaccine efficacy, duration, reaction time, and spatial scale. In all scenarios examined, vaccination averts more cases than deaths, probably because health-care workers are younger (mean age 39 years vs 51 years) and constitute 27% of cases.
Short durations or long waits diminish the effect of proactive campaigns. Although the time until the next major outbreak and its magnitude cannot be known, the 2013-14 Saudi Arabia outbreak was unique in its scale, 23 so the wait until the next outbreak will probably be long. However, even if large MERS-CoV outbreaks become more frequent (eg, every 2 years), regional-level and national-level campaigns are still superior. While additional doses could surmount the problem of waning vaccine efficacy in proactive campaigns, this approach would incur greater costs than required for reactive campaigns.
The spatial scale of reactive campaigns is crucial. If hospitals react individually, many cases occur before vaccination or immunisation. However, regional-level or national-level campaigns are affected less by these delays and avert more cases than their proactive equivalents, even if the wait is short and the duration is long. Conversely, hospital-level reactive campaigns are most efficient (in terms of cases averted per thousand doses) and, for very short durations combined with quick reaction times, avert more cases than campaigns at the regional and national levels. For very short durations, quickly reacting to local outbreaks is preferable to reacting too quickly nationally and thereby allowing vaccine efficacy to wane. This situation effectively demonstrates the pitfalls of proactive campaigns in microcosm: vaccinating too early risks substantial waning. Using triggers based on a set number of cases in a given timeframe rather than single cases before launching hospital-level campaigns practically nullifies the effect of those campaigns on cases and deaths, whereas the effect is preserved at the national level for all but the most stringent triggers.
Notably, the relative performance of reactive and proactive campaigns depends little on vaccine efficacy, and the rank order of campaigns is unaffected by reservoir control measures. Essentially, regional-level and nationallevel reactive campaigns offer an opportunity to outpace the epidemic, acting as a proactive campaign with a more certain, shorter wait time.
Our analysis is reasonably robust to the protection waning model, but more sensitive to the choice of timeframe. Hospital-level reactive campaigns rarely outperform proactive campaigns when simulated on subsets of the data. However, our conclusions that spatial scale is crucial and that national-level reactive campaigns best reduce cases are strengthened. Furthermore, focusing on single smaller outbreaks is misguided: the objective is achieving maximum possible impact over the widest possible timeframe. Effectively, a reactive campaign against one outbreak can be considered proactive against the next.
Our analysis has some limitations. We assume that downstream cases of inoculated vaccinees would not be infected, whereas, in practice, downstream cases could be infected by another infector, and vaccination could affect disease and transmission differently, although we have attempted to quantify the maximum possible extent of this bias by decomposing the vaccine impact into direct and indirect protection. We assume no agedependency in vaccine efficacy, although SARS-CoV-2 suggests that such dependency is likely. [24][25][26] We have not accounted for behavioural changes (ie, risk compensation) due to available vaccines, which might affect reservoir spillover or other transmission contributions. We have not explicitly modelled vaccine availability, hesitancy, or logistical issues that would affect campaigns, although these issues would probably result in the higher reaction times we do simulate. We have modelled simple reductions in reservoir transmission, without specifying what these measures would entail (eg, animal vaccination, 27,28 improved hygiene, or reduced physical contact). We have not considered new MERS-CoV variants, beyond modelling multiple vaccine efficacy values. If new variants transmit more slowly or quickly, the effectiveness of vaccinating health-care workers within a given reaction time could be affected. Without appropriate data informing more precise analyses, attempting to account for these issues would be overly speculative. We assumed no immunity until 14-days post vaccination, perhaps leading to slight underestimation of the effect of reactive campaigns because, in practice, there might be some protection earlier. The effects of different delays from vaccination to protection can be inferred; if the time to protection is, for example, reduced to 10 days, the results presented hold for reaction times 4 days longer than those quoted above. Finally, our analysis cannot predict the epidemic curve beyond the line list period: it is a data-driven approach that models counterfactuals on a real epidemic, and does not estimate transmission or other quantities beyond the dates in the dataset.
Because MERS-CoV outbreaks are quite infrequent, traditional randomised controlled trials might be infeasible 29 and vaccine efficacy and duration might be difficult to measure. It is therefore useful to have, in preparation for the next large epidemic, maximally effective strategies that are robust to such uncertainty. Regional-level or national-level reactive campaigns outperform proactive campaigns when targeting healthcare workers; therefore, these strategies are preferred if vaccine stockpiles are sufficiently large. Substantial reduction of MERS-CoV cases and deaths is possible, even when vaccinating health-care workers only, underlining the need for at-risk countries to stockpile vaccines when available.