Modelling approach

We used the same mathematical model as the one in [3] and [12], used to recommend the introduction of the paediatric seasonal influenza vaccination programme in the UK. The model brings together surveillance data from a variety of primary care, secondary care and sentinel data sources in addition to data on social contact patterns and seroepidemiological data and uses these data in a Bayesian approach, specifically adaptive Markov chain Monte Carlo (MCMC) techniques, to reconstruct epidemics using a transmission model for three influenza subtypes over 14 years. The model simulated seasonal influenza outbreaks in seven age groups (<1 year, 1–4, 5–14, 15–24, 25–44, 45–64 and 65+ years) and two at-risk groups (low risk, clinical risk). All parameters for both the transmission model and economic evaluation model remained the same from both publications. We assumed that the paediatric programme had been implemented for all primary schools only and simulated epidemics using the model through the fluEvidenceSynthesis R package [13].

The model follows a modified Susceptible-Exposed-Infectious-Removed (SEIR)-type structure with gamma-distributed latent and infectious periods. At the beginning of an influenza outbreak, a small fraction of individuals in each age class is infectious with the remainder susceptible to infection. Mixing between clinical risk groups is random within each age group, with the population mixing patterns described via a resampling of the POLYMOD matrix for the population of Great Britain [14]. Costs and health benefits from the perspective of the health care provider were discounted an annual rate of 3.5%. The full model and cost-effectiveness framework are described in greater detail in both [3] and [12].

We simulated four scenarios in the model, described below and shown in Fig. 1.

Model parameters

Vaccination coverage data for all age and risk groups were taken from the monthly coverage data published by Public Health England [17]. The data for the final coverage achieved are shown in Table 1. The primary and tertiary care costs from [3] were updated.

We assumed that the new quadrivalent vaccine administered to the clinical risk groups and elderly individuals would cost the same per dose as the quadrivalent vaccine administered to school children. The maximum incremental cost per dose of the quadrivalent vaccines over the trivalent vaccines was estimated using a threshold analysis on the mean incremental cost-effectiveness ratios for each proposed vaccination programme for a series of potential willingness-to-pay (WTP) thresholds. The incremental vaccine costs were varied from no difference to £15. The criteria to be cost-effective were in line with recommendations from the JCVI that 90% of simulations had an incremental cost-effectiveness ratio below the WTP threshold [18].

Epidemics for each vaccination programme were simulated 1000 times. We estimated the mean savings in primary care and tertiary care, reporting 95% CIs of the means based on 1000 bootstrap replications.

Sensitivity analysis

We conducted a sensitivity analysis using the highlighted parameters (indicated with §) in Table 1 to assess the sensitivity of our estimates to parameter uncertainty. The efficacy of the quadrivalent vaccines in preventing infection from influenza B is unclear, so we assumed a variation of ±20% in absolute terms. The uptake for all three target cohorts was varied by ±10% in absolute terms. In addition, we varied the proportion of infections that attend primary care consultations with influenza-like illness (ILI) symptoms by ±50% of that reported in [1]. The impact of these uncertainties was assessed against the most cost-effective incremental cost of the quadrivalent vaccines with a WTP threshold of £20,000 per QALY.

Additional results assuming all children aged 2–16 years are eligible

If the paediatric vaccination programme is extended to all children of ages 2–16 years, then the maximum incremental costs of the quadrivalent vaccines in each proposed programme would be £4.04, £1.03 and £0.06 respectively.

Model sensitivity

It is clear from the results presented in Table 3 that the programme that allows the largest variation in costs between the trivalent and quadrivalent vaccines is the paediatric programme. The cost-effectiveness of this programme is most sensitive to the efficacy of the paediatric QLAIV/QIIV to the B strains. If the vaccine used for children was more efficacious at preventing influenza B infection, then the maximum incremental cost of the vaccine would be £9.01, whilst a less efficacious vaccine would need to be £2.42 more expensive than the paediatric LAIV/TIIV to remain cost-effective (Fig. 3). Variation in the uptake of the vaccine in this age group by ±10% in absolute terms is also a source of uncertainty for the maximum incremental cost of the QLAIV. In comparison, the rate of GP consultations attributable to ILI had little impact on the cost-effectiveness of the programmes as did the uptake of trivalent vaccines in the older age groups.

Cost per QALY	Programme 1	Programme 2	Programme 3
£10,000	£3.69	£1.03	£0.11
£15,000	£5.04	£1.44	£0.16
£20,000	£6.36	£1.84	£0.20
£25,000	£7.58	£2.25	£0.26
£30,000	£8.89	£2.66	£0.31

Other parameters such as the efficacy of the QIIV for the adult age groups cannot have an impact on the cost-effectiveness of the paediatric programme. Their impact on the cost-effectiveness of the adult QIIV programmes is explored in Additional file 1.

Strengths and weaknesses of this study

This study uses a dynamic transmission model that was fitted to 14 years of surveillance data, and it uses the same cost-effectiveness framework that was used to support the decision to implement a nationwide seasonal influenza vaccination programme for healthy children [12], providing a consistent approach to understanding the impact of an evolving vaccination programme. This approach, as outlined in the previous publications, accounts not just for the direct protection inferred on vaccinated individuals but also the indirect protection inferred on unvaccinated individuals in the population. The model also uses the same data [14] on social contacts to account for mixing patterns in the population, an important inclusion to understanding infectious disease dynamics [20].

We assumed that existing trivalent vaccines were poorly matched to the dominant circulating B strain in each influenza season. Whilst this assumption allowed us to maintain a consistent modelling approach with that of Baguelin et al., we recognise that it presents the best-case scenario for introducing quadrivalent vaccines. In addition, other countries have reported that some vaccines were well matched against the dominant circulating influenza B strains with little activity for the unmatched B strain [21, 22], which would reduce the maximum incremental cost per dose of quadrivalent vaccines in our analysis. However, our approach allows us to report the maximum incremental cost in the best possible scenario for quadrivalent vaccines with the understanding that any previous years with a better strain match in vaccines will reduce the cost-effectiveness of quadrivalent vaccines. Indeed, the estimated proportion of influenza B infections between 2000 and 2010 caused by vaccine mismatched strains was 52.4% [23], presenting an encouraging opportunity for quadrivalent vaccines to reduce the public health impact of seasonal influenza.

We did not consider extending eligibility to children aged younger than 2 years old, in contrast to paediatric vaccination programmes in some other countries. The live attenuated vaccine is not licensed for children in this age group; and the effectiveness of inactivated vaccines is lower [24]; and implementation of a programme using an injectable vaccine would represent a considerable additional workload for a policy that has not been recommended in the UK.

Our economic evaluation of the vaccination programmes followed the guidelines recommended by the National Institute for Health and Clinical Excellence (NICE) in evaluating the costs of the programme from the perspective of the health care provider [25]. We took a conservative approach in assessing the potential additional expenditure appropriate to reducing the disease burden caused by influenza B strains. We considered vaccination programmes with at least 90% of all simulations below the WTP threshold as cost-effective, with a WTP threshold of £20,000 per QALY as recommended by NICE. We also reported our results for other WTP thresholds should that advice change in the future.

Our sensitivity analysis reported that our model estimates are most sensitive to the estimated efficacy of the QLAIV in preventing influenza B infection, but the manufacturers of the vaccine have not confirmed these estimates and have instead hypothesised that the similarity in manufacturing processes between the LAIV and QLAIV implies that there is no difference. Greater clarity on the estimated efficacy of the QLAIV in preventing influenza B infection would greatly improve the accuracy of our estimates.

A recent systematic review of economic evaluations assessing the impact of quadrivalent influenza vaccines compared the modelling approaches and results of 16 analyses published before September 2016 [26]. The authors reported that 13 of these analyses were funded by vaccine manufacturers, two did not specify the funding source and the one remaining study was publicly funded. Our independent study is therefore a helpful addition to the literature on the cost-effectiveness of quadrivalent vaccines that is so dominated by industry-funded studies. In addition, the review of de Boer et al. calls for more extensive use of dynamic transmission models to fully understand the impact of quadrivalent vaccines in influenza vaccination programmes, an approach that we adopted here using the fluEvidenceSynthesis R package.

Relation of this study to other studies

The review of de Boer et al. reported that the range of incremental cost per dose of quadrivalent vaccines over existing trivalent equivalents for all studies was $1.25 to $7.14 in 2015 US dollars [26], though there was variation in the perspective for which economic evaluations were performed, the WTP thresholds for each country and the requirement of the proportion of simulations that should be below those WTP thresholds. Many studies reported that quadrivalent vaccines are cost-effective with an emphasis of the sensitivity of these estimates to the parameters for the cost of vaccines and the efficacy of vaccines considered against the circulating influenza B strains. We adopted a conservative approach to our economic evaluation and feel that our conservative incremental cost-per-dose estimates for each proposed programme compare well to those of the studies featured in this review, with similar findings from our sensitivity analysis to those featured in the review.

Possible explanations and implications for clinicians and policymakers

Extending a QIIV programme to include clinical risk individuals younger than 65 years old in England is more likely to be cost-effective than extending further to also include all elderly individuals. However, the maximum incremental cost per dose of the quadrivalent vaccines is just £1.84, meaning that the current policy of reimbursing GPs and pharmacists for administering QIIVs to clinical risk groups is likely to be cost-effective if the incremental reimbursement cost is less than this value and less than £0.20 per dose if extended further to all elderly individuals. This result is influenced by the heterogeneous burden of influenza B infection in England [1], with children shouldering the largest proportion of this burden; therefore, directly protecting the elderly from infection with QIIV offers a smaller return than the QLAIV programme in schools.

Unanswered questions and future research

We did not address the issue of adverse events arising from vaccination using the QIIV and QLAIV and assumed that these risks were negligible, as clinical trials suggest that the quadrivalent vaccines have similar safety profiles to those of the trivalent vaccines [11, 27, 28]. We did not consider the implications of infection from natural exposure to influenza B, nor did we attempt to estimate the impact of repeated vaccination on the immunogenic response in the patient.