1. Introduction
Economic evaluations of new vaccines coming onto the market are often developed and published prior to authorisation and launch, based on summary efficacy data from randomised controlled clinical trials conducted in places where the vaccine will be first administered [
1,
2,
3]. Such economic assessments present potential value estimates with assumptions made about the long-term vaccine effect [
4]. They provide important information with a cost-effectiveness analysis that influences price-setting for the new product at market launch. The evaluations are supported by extended sensitivity analyses of the variables subject to uncertainties. This approach is well established and recommended in guidelines, and the evaluations are applicable in countries that wish to assess the economic value of the new products to reimburse, as local authorities are willing to pay to the vaccine producer a vaccine price worth its economic value [
5]. However, it is surprising to observe that these early economic evaluations are rarely challenged by data collected subsequent to the approval and implementation of the new product [
6,
7]. Moreover, if the initial assessment is simple in its presentation, it is likely that long-term evaluations will not be questioned [
8,
9].
However, effect monitoring of new vaccines in real-life settings is essential to obtain accurate economic value estimates in the short to long term [
10]. Results based on observed data should be compared with projections made at vaccine submission when aiming for reimbursement [
11]. This is particularly relevant for preventative vaccinations, as the potential gain could be affected by many different factors impacting the long-term benefit, which are unknown prior to launch. Rotavirus vaccination provides a perfect example of the need for long-lasting monitoring and evaluation.
Different vaccines are available on the market against rotavirus infection, of which two are predominant in high-income countries: a two-dose human assorted live-attenuated vaccine called Rotarix (GSK), and a three-dose live-attenuated human–bovine assorted vaccine called RotaTeq (Merck) [
12,
13]. It is assumed that the effect over time is equivalent between both vaccines [
14]. Before the start of this vaccination programme against rotavirus, it was generally considered that this disease (diarrhoea in children) was easy to manage in high-income countries with a low mortality rate [
15,
16,
17]. The vaccine had a major positive effect on hospitalisations observed in the clinical trial data [
1]. Its administration was straightforward because of its oral formulation. However, when observed, real-world vaccine effect data were collected and scrutinised in detail, the actual impacts of the vaccination and the disease were difficult to understand. Real-world data were collected in a special study set up in Belgium in 2007, called the Rotavirus Belgium Impact Study (RotaBIS) [
18]. This study showed that there was seasonality in rotavirus infection spread (mostly between January and March); a vaccine herd effect early on; and potentially waning vaccine efficacy to consider when adequately fitting the observed with the modelled data [
19,
20,
21,
22]. Moreover, a vaccine catch-up programme to immunise the entire age group up to the age of 5 years was not possible, because the vaccine has a very low frequency of a serious side effect (intussusception) if the doses are not given within strict time schedules [
23,
24]. Therefore, continuous vaccination of new-born infants with high coverage from the start was needed to obtain control of the infection spread. The follow-up of the observed RotaBIS data identified two key points [
22]. First, if the initiation of the vaccination programme was not optimal, this could lead to low vaccine coverage in the group forming the primary source of infection during the normal rotavirus peak season, with the consequence that the herd effect could be low (15%) in the first year and could disappear in the second year due to greater prominence of secondary sources for infection spread [
21]. Second, with suboptimal vaccination implementation, the primary source of infection shifted after a while from very young children (less than 13 months old) to an older age group, which may result in long-term regular seasonal peaks of the disease at a lower frequency and height than pre-vaccination. However, the reduced herd effect and the appearance of new smaller disease peaks after a while could be altered with optimal initiation of the vaccination programme, with high coverage from the start (around 90%), and an optimal start date for the vaccination programme (at least 6 months before the next seasonal peak). These findings could be deduced from a more in-depth analysis of the rotavirus vaccination with the RotaBIS follow-up data.
The objective of the present analysis is to evaluate the economic value of an optimal vaccine launch, compared with a non-optimal situation such as the one observed in Belgium. The analysis uses an evaluation technique that allows the simulation of different vaccine launch scenarios, with different long-term accumulated outcome results for the economic assessment. It may identify threshold conditions that determine whether an initial vaccination strategy moves to optimal or less optimal long-term cost-impact results.
4. Discussion
Rotavirus vaccination is an interesting case study to illustrate that there may be potentially important differences in economic value between pre-launch model predictions compared with real-world observational data over time. At the beginning, performing a cost-effectiveness analysis for rotavirus vaccination was considered a straightforward exercise, even with the use of dynamic models, to estimate the potential health gain and price-setting [
40,
41]. The reality observed in Belgium by the RotaBIS study indicated much greater complexity in infection spread and the vaccine effect. The vaccine launch in Belgium was, by chance, an intriguing case because it was not optimally implemented, but this was not known at the time of reimbursement in November 2006 [
42,
43,
44]. Comparing observed and predicted data made it possible to identify issues in virus spread in the child population, with primary and secondary sources of infections that early rotavirus disease models did not include [
11]. Most models assumed a single source of infection that reduced over time with vaccination [
45,
46]. In addition, the seasonality of the infection implied that there were clear, annual periods of intense virus transmission that should be targeted at the start of the vaccination programme with a very high vaccine coverage of the population transmitting the infection. This was not achieved in Belgium, with the now known consequences [
19]. Finally, the vaccination programme did not allow for a catch-up strategy, such as vaccinating a whole age group up to 3 years old at the beginning, because of vaccine safety concerns [
23]. This had the consequence of not achieving immediate control of virus spread in children who were older than the target age for vaccination. It was the reason for splitting the evaluation into two periods: a vaccine uptake and a post-vaccine uptake period [
22]. All these elements show the importance of a detailed understanding of the pre-vaccination infectious disease situation and patterns of infection spread, before introducing a new vaccine. The vaccine administration process and the potential constraints and safety concerns should be well-known at the start of the vaccination campaign.
Modelling these elements has helped to clarify the indirect effects of the vaccine that increase or reduce the herd effect, influencing the net vaccine effect and explaining the appearance of new disease peaks over time with a shift to older children as the primary source of infection [
22]. The data from Spain confirm the findings in Belgium with a sub-optimal launch of rotavirus vaccination [
32,
47]. The data from Finland and the UK may prove that initiating the vaccine programme earlier in the year and with an immediately high coverage achieves greater reductions in hospitalisation, compared with what was observed in Belgium [
38,
48,
49]. This suggests that Belgium could have obtained better results by starting the programme differently, although this was not known at the time. After reaching the stage of the post-vaccine uptake period, the modelling results indicate that it would be very difficult to substantially improve the results unless a massive, new intervention shock happened. By chance, such a shock occurred with the lockdowns introduced due to the COVID-19 pandemic in 2020 and 2021, and the rotavirus peaks during those years disappeared in Belgium [
39]. This striking result would have been very difficult to achieve without the lockdowns, as increased vaccine coverage does not immediately reduce the primary infection source that shifts to an older age group not directly under the effect of the vaccine. Only the very young ages are vaccinated. This is critical information because when the vaccine programme is not well initiated, it has long-term negative consequences that are difficult to adjust. It is also the situation of the rotavirus vaccination results currently observed in the US [
50,
51].
A few additional questions could be asked in relation to this economic evaluation. One is about the economic value this vaccine should have pre-launch that defines its price of reimbursement at launch, with a better understanding of the importance of how the vaccine programme is introduced. Analyses relying on simpler models, without taking into account the new knowledge of the optimal method of introducing the vaccination programme, as was carried out in Belgium, may produce a range of cost-effectiveness results in the sensitivity analysis that includes the optimal result. However, such an analysis would not be able to indicate how to achieve the optimal result if not all the necessary details were included in the model construct. In this case, the absence of information on an optimal vaccine introduction to define the price-setting at launch is a risk for both the producer and the paying party. Either may find that they are paying or being paid too much or too little for a vaccine, and it is difficult to readjust the vaccination programme after a non-optimal introduction because of the limitations of readjustment interventions, such as increasing the vaccine coverage rate. Therefore, it is very important to refine the vaccination programme at its introduction to maximise the efficiency of the programme in the long term. Thus, the recommendation is that an economic submission for reimbursement should evaluate different scenarios of vaccine introduction that consider the differences in cost-effectiveness and cost-impact analyses related to the vaccine coverage rate, and the time selected for vaccine introduction, in relation to the expected seasonal disease peak. This approach, with an emphasis on obtaining initial high coverage ahead of the next expected seasonal peak, could be applicable to other diseases with marked seasonality and high contagion. If COVID-19 becomes an endemic disease in infants with seasonal peaks, these findings may be relevant to future research on the design of a potential COVID-19 vaccination programme in this age group. Could this have been foreseen in the Belgium submission file for the rotavirus vaccine? This would have been difficult if the full infection spread was not well understood at the beginning, having identified the importance of an optimal introduction of the vaccination and having discovered the age shift in the primary source of infection after a sub-optimal introduction. In this respect, one should remember that the European randomised controlled trial (RCT) conducted for Rotarix in 2004–2006 had a randomisation process of two vaccinated children for one placebo child [
1]. This type of randomisation increased the herd effect in the placebo group, as the randomisation occurred at local level and not by a cluster site. In cluster site randomisation, regions are divided into clusters. The clusters are randomised to vaccination or no vaccination, thereby reducing the chance of a herd effect occurring in the unvaccinated clusters. In contrast, in local randomisation, the control group is subject to herd effects, resulting in an apparently decreased vaccine effect in the second year of the evaluation. This was not considered when the analysis was conducted and reported because of the lack of baseline information prior to the vaccine introduction. It is also possible that other, additional factors may influence the observed local results, such as the organisation of day-care centres and their potential function as a hub of local epidemics, which would not have good infection control and have poor vaccine coverage. However, there are limitations on the complexity of models that can be constructed and applied in practice. Factors that do not have critical effects on vaccine impact or do not cause important costs or health changes may add little to the economic value generated by the more complex model. The right balance needs to be found between the feasibility of collecting and analysing sufficient data and the wants and needs of the paying parties and producers. The precise balance is likely to vary between specific interventions and settings.
Finally, is this economic model also applicable to other settings such as non-high-income countries? Some critical points that are essential for the optimal functioning of the vaccine in high-income countries but that could be absent in other countries include the seasonality of infection spread, easy contact patterns among very young children (such as day-care centres) that facilitate virus transmission, and the hospitalisation of severe cases leading to a high healthcare cost. It would be a challenge to apply the current model if any of those conditions were not fulfilled. Nevertheless, this analysis indicates that rapidly achieving high coverage at the start of a rotavirus vaccination programme is essential for maximising the health benefit of the vaccine, as this minimises the development of secondary sources of infection that persist over time and are very difficult to correct at a later stage.
The analysis presented here has some limitations. Some cost items, such as first-line support and indirect costs, were not considered, and not all the QALY losses at different disease stages were included in this evaluation. However, the focus of the analysis was to demonstrate that quite different economic value results could be obtained for a vaccine from a predicted pre-launch value assessment and real-life data observations. Vaccination needs data monitoring on its effect over time once approved and implemented, in order to capture deviations from what could be considered an optimal launch. The economic analysis is an additional tool to help in the selection of a vaccine strategy. For instance, some countries like to produce price–volume contracts when introducing new vaccines. These results suggest that such a policy would be a disaster for rotavirus vaccination if the volume is fixed at 40% or 50% for the first year of implementation followed by progressive increases in vaccine coverage over time. With a start at 40–50% vaccine coverage across a country, the present model suggests that it is likely that a limited effect will be seen on hospitalisation reduction, limiting the total value of the vaccine in the short to long term. Conversely, this model suggests that obtaining immediate very high vaccine coverage ahead of the next expected seasonal disease peak would maximise both the health benefit of rotavirus vaccination and its cost impact over the long term.