Epidemiology and vaccine effectiveness
Influenza incidence estimates were extracted from CDC reports for 5 influenza seasons from 2013 to 2018 (Table S1) (11), and combined with WHO FluNet (12) virological data to obtain yearly incidence per strain for the US. The model assumed 66% of people infected with influenza virus were symptomatic (13), that they had an incubation period of 0.8 days, and remained infectious for 1.8 days (14,15).
We used strain specific QIVe VE estimates obtained by Rolfes et al. for the influenza season 2017-2018 (16) assuming VE against A(H3N2) was a mismatch between the vaccine and the circulating strain. To estimate VE conferred by QIVc, we extracted results from a recent study estimating the relative VE (rVE) of QIVc compared to QIVe from electronic medical records, where patients were matched by propensity score for the influenza season 2017-2018 (17), i.e. the same influenza season as in Rolfes et al. This study estimated an overall rVE of 19.3% (95% CI [9.5%;28.0%]) for QIVc compared to QIVe, and age-adjusted rVE for adult population (Table 1). In order to compute the specific QIVc increased effectiveness against A(H3N2) (in 2017-2018, only A(H3N2) antigens are cell-based), we used the total rVE estimated on a US cohort comparing QIVc and QIVe for the same influenza season, and then, we recomputed, the QIVc VE against A(H3N2) using the following equations: see equations 1, 2, and 3 in the supplementary files.
Epidemiological and economic model
A four strains compartmental transmission model was developed to provide estimates of the epidemiological impact of the switch from QIVe to QIVc. The model is a classic SEIR model where the population can be either Susceptible to infection with Influenza strain I (Si), Exposed to the strain (Ei), Infected and infectious (Ii) or Recovered from infection (Ri) (Figure 1). The vaccinated population could still be infected and contribute to the infection dynamic but with a reduced probability corresponding to the vaccine effectiveness against the given influenza strain. The model simulates independently the epidemiological dynamics of A(H1N1)pdm09, A(H3N2), B/Victoria and B/Yamagata for a given influenza season. The model is structured by age-group (6-23mo, 2-4yo, 5-12yo, 13-17yo, 18-49yo, 50-64yo, and more than 65yo) and uses a contact matrix to account for assortative rate of contacts between age-groups. In our analysis, we used in our base case analysis the matrix from Mossong et al.(18) and have conducted sensitivity analysis using Zagenhi et al.(19). Both matrices have provided qualitatively similar results. Probabilities of influenza transmission per influenza strain are estimated for each influenza season to match US reported strain specific attack rate (2013 to 2018 influenza seasons). We assume a pre-immunity of a third of the population based on estimations from Baguelin et al. (20). The estimation process uses the non-linear Nelder-Mead simplex algorithm(21) to maximize a likelihood function.
The economic model is based on a decision tree model published in De Boer et al. (22) whose inputs are given in Table S3. We used the same methodology and same health outcomes computed on the whole population. The number of cases per age-group estimated by the epidemiological model, shown in Table 2, are taken as inputs of the economic model. Probabilities of general practitioner visit, hospitalization, and death are applied on the number of cases attributed to a high or low risk group, and then translated into public health outcomes and costs. The economic analysis is performed from a societal perspective without taking into account productivity loss due to death. We consider a willingness to pay per QALY threshold of US$50 000 to consider a strategy as cost-effective (23).
Economic data
Disease costs and QALYs were extracted from a recent influenza health economic analysis performed in the US context (22). Cost of a work day for the pediatric population are assumed to be related to parental work loss. Vaccine cost for QIVe and QIVc were set at $17.22 and $24.22 respectively(24) . We do not consider administration costs since they are assumed to be the same across the different vaccination strategy (no difference in vaccination coverage). Details of the costs per age-groups are given in Table S3.
Scenarios
As a reference strategy, we assume that the US population is vaccinated with conventional QIVe for those aged under 65 years of age and TIV HD for those aged 65 years and above. Then we compare this strategy to a scenario where QIVe is replaced by QIVc for people aged 18–64 years, other age-groups keeping their baseline vaccination. For both scenarios, we use age-based vaccination coverage rate documented by CDC(25), in particular we consider that 34.9% of people aged 18 to 49yo, 47.30% of people aged 50 to 64yo, and 68.10% of people older than 65 years are vaccinated against influenza (Table S2). In this base case comparison, we consider that a seasonal mismatch between the QIVe A(H3N2) strain and the circulating strain due to egg adaptation occurred during the last 3 years out of 5 years in the analysis scope (26). As a sensitivity analysis we also assessed the impact of QIVc when the mismatch due to egg adaptation occurred over a varying number of years from 1 to 5 years, and randomly picking the influenza season concerned by the mismatch.
We also performed a stochastic probabilistic sensitivity analysis in order to assess the robustness of our results regarding uncertainties in vaccine effectiveness, economic inputs (primary care and hospitalization costs), probability of outcomes, listed in Table S3 and Table S4, with their probability distributions. In this analysis, 1,000 sets of the above-mentioned parameters are randomly drawn from distributions indicated in the Table S2 and Table S3. Clinical and economic results are averaged over the 5 influenza seasons.