Evaluation of the epidemiological outlook of the influenza A/H3N2 clade K in England during the 2025-26 season

Evaluation of the epidemiological outlook of the influenza A/H3N2 clade K in England during the 2025-26 season James A Hay*,1, Punya Alahakoon†,1, Alexander Greenshields-Watson†,1, Michelle Kendall1, Mahan Ghafari1,2, Chris Wymant1, Robert Hinch1, Luca Ferretti1, Jasmina Panovska-Griffiths1,3,4, Christophe Fraser1 1. Pandemic Sciences Institute, Nuffield Department of Medicine, University of Oxford, Oxford, UK 2. Department of Biology, University of Oxford, Oxford, UK 3. The Queen's College, University of Oxford, Oxford, UK 4. UK Health Security Agency, London, UK * Correspondence to [email protected] † Contributed equally Key findings ● England is currently experiencing a high growth rate of infections caused by the influenza A/H3N2 K clade. Antigenic change from the previously dominant clade, a rapid selective sweep evident in genomic data, and an unusually early start to the season have raised concerns about the potential severity of this year’s influenza season. ● Analysis of publicly available data sources from the current season suggests that the effective reproduction number has been largely consistent with previous severe seasons. However, the modelled peak growth rate in the current season to date is slightly higher than previous peak growth rates from the past 10 years when subsetting for A/H3N2 cases, but comparable when aggregating all influenza cases. ● Scenario analyses using an age-stratified compartmental model compared to the previous A/H3N2 season in 2022/23 suggest that substantial immune escape is unlikely given current epidemiological trends. Current trends are compatible with small levels of immune escape in all ages, or slightly greater immune escape in children, or a 10-20% higher R0, or an earlier seed date with no change in virus fitness or immune escape. In almost all scenarios, an earlier and faster epidemic growth rate leads to depletion of susceptibles before the Christmas period with a dampening effect due to the half term school holiday. ● To support understanding and exploration of model outputs, an interactive visualisation tool was devised and made available online: https://hay-idd.shinyapps.io/ModelFluUk-H3N2/ ● This rapid analysis is intended to support situational awareness. It provides quantitative comparisons of early epidemic growth rates with previous seasons and qualitative insights into plausible epidemic dynamics. Funding: JAH and PA are supported by a Wellcome Trust Early Career Award (grant 225001/Z/22/Z). AGW is supported by a Wellcome Trust directed call (309152/Z/24/Z). MG is supported by a Wellcome Trust Early Career Award (grant 309205/Z/24/Z). LF acknowledges support from a PSI Career Development Fellowship, the European REA, Marie Skłodowska-Curie Actions (grant agreement no. 101131463 SIMBAD), and UK Research and Innovation (UKRI) under the UK government’s Horizon Europe funding guarantee (grant number EP/Y037375/1). JPG's work was supported by funding from the UK Health Security Agency and the UK Department of Health and Social Care (DHSC). The views expressed in this article are those of the authors and not necessarily those of the UK Health Security Agency or the UK Department of Health and Social Care. RH and CF were supported by research grants from CEPI. Conflict of interest: JAH, PA, AGW, LF, RH, MK, JPG, CW, MG and CF declare no competing interests. Data availability: All code and data required to reproduce the analyses are available at https://github.com/hay-idd/influenza_H3N2_k_clade Acknowledgements: We thank Steven Riley, Simon Cauchemez, Ben Cowling, Oliver Eales, Freya Shearer, and Juliette Paireau for helpful discussion and for providing global context for the analyses. We also thank Richard Neher for helpful discussions around growth rate advantages estimated from genomic data. Use of AI: AI tools including ChatGPT and Copilot were used for literature review, sourcing data and developing code. Introduction The 2025/26 influenza season in the northern hemisphere appears to be dominated by an antigenically drifted clade of A/H3N2 viruses. This clade—named K (previously J.2.4.1)—is descended from the J.2 clade on which the vaccine strain selection was based, and has swept to dominance from a frequency of <1% on 2025-07-02 to 99% in Europe as of 2025-11-21 [1]. This suggests a substantial fitness advantage over other J.2 viruses [2]. The combination of rapid growth, multiple antigenic substitutions in the haemagglutinin (HA) protein, antigenic mismatch with the vaccine strain, typically higher morbidity and mortality amongst the elderly during A/H3N2 seasons [3,4], and an unusually early start to the season have raised concerns over the potential for a severe season in the United Kingdom and Europe more broadly [5–7]. Furthermore, the 2025 flu season in Australia was one of the worst on record [8,9], and Japan is suffering from an early epidemic leading to school closures and increased hospitalisations [10]. These indicators suggest, though not definitively, an unusually severe influenza season ahead [11]. The interaction of season timing, antigenic drift, other subtype dynamics, vaccine efficacy and coverage, non-HA-mediated immunity, immune waning from previous seasons, climate factors, and contact pattern changes around school holidays is complex and leads to highly varied cumulative and peak seasonal burden [11,12]. The potential outlook of the 2025/26 influenza season is therefore uncertain, though historical seasons with early onset and peaks have often been severe [4]. One of the key concerns around clade K is that it has an estimated growth rate advantage of ~20% over competing strains based on clade frequency data. Multinomial logistic regression was used to estimate the relative fitness of clade K based on its coefficient in the regression model fitted to clade frequencies over time [2], giving an estimated selection coefficient of around 1.2. How this growth rate advantage translates to an estimate for the effective reproduction number Re depends on the choice of reference strain and the Re of that strain. A systematic review found that the median effective reproduction number for seasonal influenza was 1.28 (IQR: 1.19-1.37), with the estimated reproduction number in 2009 (pandemic H1N1) at 1.46 (IQR: 1.30-1.70) [13]. Real-time epidemiological assessment in England is based on weekly influenza surveillance data from the Second Generation Surveillance System (SGSS) [14]. SGSS compiles test results for infectious diseases including COVID-19, influenza, RSV and others from multiple data streams: the Royal College of General Practitioners Research & Surveillance Centre (primary care), the Respiratory DataMart system (hospital testing) and other hospital testing. The main surveillance indicator in the UK Health Security Agency weekly surveillance reports is the percentage of tests that are positive for influenza, out of those patients presenting with influenza-like illness (ILI). This metric has increased much earlier than in other recent seasons following the decline of a previous COVID-19 wave, particularly among the 5-14 year old age group. Note that Scotland, Wales and Northern Ireland have separate data reporting systems and have also recently reported substantial increases in influenza activity, though with a slight lag to England [15–17]. Despite its widespread use, metrics based solely on the percentage of tests positive for influenza are known to be biased by the dynamics of other ILI-causing pathogens which makes interpretation difficult. It is therefore recommended to use an ILI+ indicator which multiplies ILI cases by the percentage of tests positive for influenza, ideally stratified by subtype [18]. This metric can be generated from publicly available data sources from UKHSA. The WHO also provides a useful global data source for influenza dynamics in England, FluNet, which combines samples from non-sentinel (e.g., outbreak investigation, POC testing etc) and sentinel surveillance sites to give subtype-specific case counts over time [19]. We focus here on these sources of influenza case counts, either overall or by subtype, and stratified by age for recent years, to understand the transmission rate of the current clade K viruses in the UK based on traditional measures of absolute growth rate and Rt. In this report, we analyse publicly available epidemiological data from the UK and present modelling analyses using an age-stratified Susceptible-Infected-Recovered model to address two main questions: 1. Do the epidemiological data indicate a more transmissible A/H3N2 strain than previous seasons? 2. Under plausible scenarios of a virus with increased transmissibility, earlier seeding, and/or substantial immune evasion, what is the potential impact on cumulative and peak healthcare burden for the rest of the season? To allow visualisation and exploration of different parameter combinations within the model runs and constructed analyses, we embedded our model into an interactive webtool using a service provided by shinyapps.io (https://www.shinyapps.io/). The tool (accessible at https://hay-idd.shinyapps.io/ModelFluUk-H3N2/) enables users to reproduce our results, but also allows exploration of different parameter combinations. Results The maximum growth rate of influenza A/H3N2 cases in the current season is slightly higher than in previous seasons We calculated weekly exponential growth rates of cases from WHO FluNet, case counts in England from the Respiratory DataMart system, and an ILI+ indicator constructed from the Royal College of General Practitioners Research and Surveillance Centre data (Figure S1-2). We compared raw and modelled growth rates from the current influenza season to previous seasons, using both generalised additive models (GAMs) and a penalised spline model described by Eales et al. (Figure 1A, Figure S3-S5) [20]. Using data from WHO FluNet, and aligning the epidemic curves to the epidemic peak across different seasons, the 2025/26 season to date shows an earlier peak (epidemiological week 40 vs. 49, 49, and 47) and a faster A/H3N2 peak weekly growth rate (posterior means and 95% CrI: 0.680 [0.503-0.871] vs 0.0886 [-0.0577-0.238], 0.388 [0.264-0.515] and 0.592 [0.493-0.708]) compared to the 2024/25, 2023/24 and 2022/23 seasons (respectively) (Figure 1B). Model fits overlaid on the empirical growth rates are shown in Figure S6. We note that the choice of smoothing model is has a relatively large impact on the estimated growth rates due to the amount of noise in the data, and thus we present an alternative fit using a Gaussian random walk, noting that the peak estimated A/H3N2 growth rate for the current season is higher under this model (Figure S7). Similar relative trends were observed using the RCGP RSC data (Figure S8). Comparing growth rates using all influenza cases (i.e., aggregating A/H3N2, A/H1N1pdm09, unsubtyped influenza A, and influenza B) suggests that the overall peak influenza growth rate is comparable to previous seasons (Figure S9&S10). The 2025/26 season to date has a peak on epidemiological week 40 compared to weeks 49, 49 and 49 in the 2024/25, 2023/24 and 2024/25 seasons respectively. The 2025/26 season a peak weekly growth rate for ‘all influenza’ of 0.482 (posterior mean; 95% CrI: 0.385-0.584) compared to 0.528 (0.476-0.582), 0.584 (0.526-0.647) and 0.548 (0.505-0.593) for the 2024/2025, 2023/2024 and 2022/2023 seasons (respectively). The dominant subtype from each prior season back to 2014/15 is shown in Table 1. Figure 1. Growth rate estimates of A/H3N2 cases from the WHO FluNet database using the penalised spline model. Coloured lines and shaded regions show posterior mean and 95% credible intervals for the model-estimated weekly growth rate. Colouring distinguishes the current season from post, pre and during COVID-19 pandemic seasons. The doubling time corresponding to the growth rate is shown on the right hand y-axis. Table 1. Summary of dominant influenza A subtypes by season (derived from UKHSA hospital typing data [21], Influenza | UKHSA data dashboard). Season Dominant subtype 2014/15 A/H3N2 2015/16 A/H1N1pdm09 2016/17 A/H3N2 2017/18 A/H3N2 2018/19 A/H1N1pdm09 2019/20 A/H3N2 2020/21 N/A – COVID-19 pandemic 2021/22 A/H3N2 2022/23 A/H3N2 2023/24 A/H1N1pdm09 2024/25 A/H1N1pdm09 2025/26 (current) A/H3N2 Figure 2. Growth rate estimates using a penalised spline model. School holidays are marked with vertical grey bars. The autumn half-term break in 2025 is marked in red. Solid lines and ribbons show posterior means and 95% credible intervals. We note that this is the date of the autumn half-term across most English schools, but there is a slight variation in some regions and schools Weekly log growth rates calculated directly from the data are shown in faint coloured lines to visually evaluate model fit. Age-stratified growth rates highlight much higher growth in children than adults We observed substantially higher recent growth rates of A/H3N2 cases in the 1-4 and 5-14 year old age groups compared to adults and adolescents (Figure 2, Figure S11). The difference in A/H3N2 weekly growth rates between children and the 15-44 year old age group was far higher than has been observed in the prior two influenza seasons (Table 2, Figure S12). A substantial drop in growth rates was observed preceding and during the recent Autumn half-term school break, consistent with previous years and with the known role of school contacts in driving influenza spread (Figure 2). However, we have so far only been able to obtain an age-stratified ILI+ indicator for A/H3N2 cases going back to the start of the 2023/2024 influenza season, which limits our ability to compare age-specific trends to previous A/H3N2 seasons. Comparison of these age-stratified growth rates to previous seasons should therefore be interpreted cautiously until historical data can be included. Table 2. Values shown are the absolute difference in the posterior mean estimated peak weekly growth rate between the shown age group (years of age) and the 15-44 year old age group. Season 1-4 5-14 45-64 65+ 2023 to 2024 0.20 0.31 0.10 0.25 2024 to 2025 0.15 0.19 0.10 0.14 2025 to 2026 0.49 0.58 0.11 0.19 Estimates for the timevarying reproduction number suggest a similar peak infection rate to previous seasons We used the WHO FluNet data from 2015 to 2025 to estimate the timevarying reproduction number, Rt. Note that here we distinguish Rt from the effective reproduction number, Re, which we define as the reproduction number at the start of the season. Across the influenza seasons 2015/16, 2016/17, 2017/18, 2018/19, 2021/22, 2022/23, 2023/24, 2024/25, and 2025/26, the estimated time-varying reproduction number Rt generally remained close to 1, with intermittent fluctuations dropping below 1 and occasionally reaching values above 1.4 (Figure 3). When trajectories were aligned relative to the pre-Christmas period (before December 25), peak Rt values consistently occurred between early October and approximately December 20 across all seasons examined (Figures 4, Figure S13). Notably, post-COVID-19 pandemic seasons exhibited relatively stable Rt trajectories, and the 2025/26 season was in line with previous seasons, with Rt estimates not exceeding 1.4 (posterior mean; 95% credible intervals: 1.22, 1.47). This suggests that the overall dynamics for the most recent season (2025/26) were broadly similar to those of previous seasons. For comparison we also estimated the current Re in Japan, which at time of writing has been experiencing a longer period of exponential growth driven by clade K than England has. We obtained point estimates of 9.16 days for the doubling time and 1.26 for Re. Figure 3: Top panel: Daily incidence combining all influenza subtypes for influenza seasons 2015-16, 2016-17, 2017-18, 2018-19, 2022-23, 2023-24, 2024-25, and 2025-26. The daily incidence curves were derived by smoothing the weekly case counts obtained from the WHO FluNet dashboard. Bottom panel: Estimated effective reproduction numbers for the same seasons using the data in the top panel with the EpiEstim R package. Horizontal dashed lines show Rt=1, 1.2 and 1.4 for reference. All lines are shaded by the dominant subtype that season as shown in Table 1. Figure 4: Effective reproduction number (Rt ) for each influenza season arranged relative to peak pre-Christmas Rt. As in Figure 3, but aligning the curves to the date of peak Rt identified between June and before December 25 of that year. All seasons were aligned relative to their peak date by calculating the number of days since the peak. The date and estimated value of the peak Rt are shown in each subplot. Plots are coloured by their timing with respect to the COVID-19 pandemic. Note that we have excluded the COVID-19 pandemic years (2019/20 and 2020/21) where influenza activity was severely disrupted. holidays – the half-term holiday can act as a circuit breaker, dampening the peak and spreading cases over a longer period of time. All of our scenarios place peak incidence prior to the Christmas holiday when increased mixing across age groups is expected, reducing the opportunity for transmission into older individuals. This work presented here aims to make two contributions to situational awareness. First, we aimed to address two key unknowns early in the influenza season in England: 1) whether early epidemiological data are consistent with predictions from immunological and genomic data; and 2) identifying the range of potential epidemic dynamics compatible with current incidence trends. These types of analyses were used widely during the COVID-19 pandemic [34,35], but have since become less common for seasonal respiratory diseases. The analyses performed here are relatively straightforward and quick, using data which are already routinely collected. Second, to understand and support the model outcomes, we have set out to design an interactive visualisation tool that we made available online. Given that the model has a large number of parameters, a large number of scenario analysis can be explored. Devising a compact visualisation tool to capture the possibility of different model outcomes allows users to explore scenarios of interest that can enable insights to be drawn on possible preparedness measures and inform policymakers of different response options. Regular updating of data analyses of respiratory virus dynamics would achieve three benefits: improvements of data flows and schemas; improvements of analytical methods; and most importantly, easy contextualisation of sudden change, such as the appearance of the current K strain of influenza. For understanding the significance of such changes to health and health systems, systems and decision making tools such as the Shiny app presented here should be maintained regularly and updated early in the season to rule in or out different scenarios. Maintaining the toolkit shown here requires low-resource but consistent input from technical specialists, and real-time access to absolute reported case numbers rather than just percentage of tests positive for influenza. Our analyses have a number of limitations, some due to limitations in data, and some due to the rapid nature of our analysis. For the epidemiological analyses, we combined data from multiple sources (UKHSA reports, RCGP RSC, WHO FluNet, Respiratory DataMart). These data sources each have their own limitations in representativeness and reporting algorithms [36], and our analyses required additional assumptions to align age groups and dates. Growth rate and Rt estimates are also possibly biased by changes in testing intensity and reporting rates between age groups. When estimating growth rates, we used a penalised spline model to smooth the incidence curve, which may mask true sudden changes in growth rates. For the Rt analyses, we used WHO FluNet non-sentinel surveillance data, which also have caveats around representativeness and testing intensity, and used all influenza positive samples, which will mask subtype-specific transmission rates. We also smoothed the incidence data to help with model convergence, which again might mask genuine fluctuations in transmission rates. We also note that overconfident estimates from small sample sizes are a known issue with EpiEstim. All of the data used were reported weekly, which might obscure model finely resolved effects. Direct access to the UKHSA and RSGP influenza data would have potentially given us a more granular dataset, and if this were available in future, our analyses can be easily repeated. The model used for scenario analyses is also heavily caveated, as we did not perform a formal model fit due to time constraints. Instead, we chose fixed parameter values based on commonly assumed influenza parameters (R0, infectious period, final size), and then manually calibrated other parameters (e.g., reporting rates, age-specific symptomatic fraction, seed date and size) to achieve a reasonable visual fit to the 2022/23 influenza incidence data. Assumptions regarding age-specific immunity, immune escape, and all-or-nothing immunity in two classes rather than stratified immunity ([37]) all have a large impact on the projected incidence curves, and we therefore recommend using the tool to inform a general understanding of the system rather than predictions. A key omission is vaccination, which we did not include due to severe challenges in parameterising age-specific vaccine efficacy against infection and disease. The potential interaction of the early season with the time taken to achieve high vaccine coverage in high risk populations may be important. Our model is also limited by the use of outdated contact data, and also strong assumptions surrounding behaviour changes in school holidays and the Christmas period. Although we are confident that these assumptions capture general trends, more recent and well-calibrated parameter values would likely result in different epidemic trends. It is important to note that in this report we have only considered data from England. Scotland, Wales and Northern Ireland have all experienced a later start to the influenza season, and thus our scenarios with half-term acting as a circuit breaker may be less relevant since the timing of half-term dates differ between the four UK nations. Furthermore, a later start to the season with moderate immune escape or increased transmissibility may result in much higher incidence during the Christmas period, which may translate into higher burden in the elderly. Within the visualisation interactive tool, users can upload alternative datasets and change the population size as well as alter different model parameters to create the scenario analyses. Hence it is possible to explore additional scenarios to understand possible dynamics in Scotland, Wales and Northern Ireland, which will be important to explore in the future through collaboration with policymakers in these nations. In summary, we have combined publicly available influenza data for England with an age-stratified Susceptible-Infected-Recovered model and an interactive visualisation webtool. Our analysis shows that compared to previous years, the epidemic growth rate is high but not exceptional. Modelling analysis suggests that the epidemic trajectory so far is consistent with at most modest reduction in population immunity compared to previous years. Increases in transmission do not always translate to bigger epidemics because of the interaction between epidemic dynamics and school holidays. We provide a range of scenarios and an interactive scenario-explorer app, that show the impact of different assumptions on the epidemic curve for the coming months. Methods Epidemiological analyses Data summary The ideal dataset for understanding influenza epidemiological trends would be an unbiased measure of symptomatic A/H3N2 infection incidence stratified by age. However, it is not possible to obtain this exact dataset due to biases in testing behaviour and coverage, limitations of data reporting, and restrictions on public sharing. Instead, influenza surveillance in England consists of multiple indicators from different sources, ranging from primary care through to emergency department visits. Following recommendations from [18], our main aim was to develop an ILI+ indicator for each age group of interest, calculated as: 𝐼𝐿𝐼+=𝐼𝐿𝐼*𝑝+𝑣𝑒*𝑝𝐴/𝐻3𝑁2 Where is the proportion of all tests done which are positive for influenza and is the 𝑝+𝑣𝑒 𝑝𝐴/𝐻3𝑁2 proportion of all positive influenza tests which are attributed to A/H3N2. In the time frame of this analysis, we were unable to construct one consistent dataset to use in all analyses and we therefore pieced together different data sets to generate as close to an ILI+ indicator by age as possible. We intend for future iterations of this analysis to simplify this process. In brief, we used data on influenza cases by age group from the Royal College of General Practitioners (RCGP) Research & Surveillance Centre (RSC), data on ILI from the Respiratory DataMart system, and data for the percentage of influenza cases positive for A/H3N2 influenza from the Second-Generation Surveillance System (SGSS) (Figure S1). We multiplied these datasets together to obtain: 1) an ILI+ indicator by age group going back to 2023; 2) an estimate of absolute influenza cases (all subtypes) over time going back to 2009; 3) and an an estimate of age-stratified influenza cases (all subtypes) for the 2022/23 season (see: https://hay-idd.shinyapps.io/ModelFluUk-H3N2/). Further details on data processing are provided in the Supplementary Material. As a comparator dataset, we also obtained weekly counts of reported influenza specimens stratified by subtype obtained from the WHO FluNet platform [19]. All data used are aggregated weekly and we used the final date of the epidemiological week as the reported date. For the calculation of Re for Japan, we used weekly cases counts reported at https://weathernews.jp/news/202511/210136/ extracted on 24/11/2025 (values translated as “The number of reports from fixed-point medical institutions over the past week”). We included values reported as being from 28/09/2025 (before which growth appeared sub-exponential) until 16/11/2025 (the latest available) inclusive. We took the generation time distribution to be gamma distributed with the mean and standard deviation reported in (Chan et al. 2024): 3.2 days and 2.1 days respectively. Weekly growth rate calculations Weekly growth rates were calculated for each influenza season and aligned by calendar week as: 𝑦=𝑙𝑜𝑔( 𝑖(𝑡) 𝑖(𝑡−1)) Where is the reported incidence over week t. We then produced smoothed weekly growth rate 𝑖(𝑡) curves using two methods: 1. Fitting a penalised smoothing spline model to reported incidence data using the Bayesian method described by Eales et al [20]. We used the EpiStrainDynamics package using penalised splines with degree 3 and 3 weeks per knot [38]. We found that the model was unstable with low case counts, and thus for age-stratified growth rate estimates we first artificially inflated the counts 10-fold before fitting the model. Thus, uncertainty is likely unrepresented. 2. Fitting a Generalised Additive Model (GAM) using the mgcv R package predicting log weekly growth rate as a function of time, placing a penalised smoothing spline a basis dimension of up to 5 [39]. Time-varying reproduction number estimation We estimated the time-varying effective reproduction number (Rt ) using the EpiEstim package in R, which implements the method described by Cori et al. [40]. We used the overall weekly influenza incidence data for the period 2015–2025 from the WHO FluNet database for this analysis. To approximate daily incidence, weekly counts were disaggregated by distributing the cases evenly across days per week, followed by application of a 14-day rolling mean to smooth short-term fluctuations. The serial interval distribution was assumed to follow a distribution with a mean of 3.6 days and a standard deviation of 1.6 days, based on [41]. Rt was computed over sliding weekly windows to capture temporal variation in transmission potential. The resulting Rt trajectories were visualised to compare epidemic dynamics across seasons, highlighting differences in transmissibility and potential shifts in seasonal patterns over the study period. In this analysis, we assumed that a typical flu season starts on September 1st each year. For the estimation of Re for Japan, we first estimated the exponential growth rate r using a simple linear regression model of log(cases) against date reported. We then estimated Re using the relationship implied by the renewal equation [42]. 𝑅𝑒=1/0 ∞ ∫ω(τ)𝑒−𝑟τ𝑑τ Compartmental model and scenario analyses Model overview We simulated seasonal influenza transmission dynamics for England using an ageand immunity-structured deterministic Susceptible-Infected-Recovered model. We divided the population into four age groups (0-4, 5-18, 19-64 and 65+ years; Note that infants under 1 year were excluded, as we did not have age-stratified A/H3N2 ILI+ data for this age group) and two immunity classes (fully susceptible and partially immune). Contact rates within and between age groups were derived from the POLYMOD study using the socialmixr R package [43,44]. To account for changes in contact rates outside of school terms, we resampled from the original POLYMOD data to scale contact rates during the half term holiday, pre-Christmas period, and the Christmas school holidays (described below). For each age group, we tracked the overall incidence of infections, symptomatic cases and infections per week. We informally calibrated the model to reported overall influenza case data from the 2022/23 season to provide a baseline, and performed scenario analyses varying key model parameters to generate plausible scenarios for the 2025/26 season. Model structure The force of infection in group i was defined as: λ𝑖(𝑡)=β𝑗=1 𝑚 ∑𝐶𝑖,𝑗(𝑡)𝐼𝑗(𝑡) Where β is the overall transmission rate (not age-stratified), Ci,j is the contact rate between group i and group j, and Ij is the number of infected individuals in group j at time t. Transition rates between the three compartments were defined by the following set of ordinary differential equations: 𝑑𝑆𝑎,𝑘 𝑑𝑡 =−α𝑘𝑆𝑎,𝑘(𝑡)λ𝑎,𝑘(𝑡) 𝑑𝐼𝑎,𝑘 𝑑𝑡 =α𝑘𝑆𝑎,𝑘(𝑡)λ𝑎,𝑘(𝑡)−𝐼𝑎,𝑘(𝑡) 𝑇𝑔 𝑑𝑅𝑎,𝑘 𝑑𝑡 =𝐼𝑎,𝑘(𝑡) 𝑇𝑔 Where ɑk denotes the relative susceptibility of immune class k and Tg is the infectious period. Note that ɑk was set to 0 for the immune population, representing all-or-nothing immunity. Leaky immunity (ɑk>0) leads to very dynamics. We solved the model in daily timesteps using the deSolve R package [45]. Immunity, initial conditions and seeding We set the population size of the model to 60,000,000, corresponding to the population size of England. We distributed the population into age groups based on the age distributions returned by the socialmixr package using the POLYMOD data. Each age group was then stratified into the susceptible or fully immune class using varying proportions to capture different levels of immune escape for particular age groups. An additional immune escape multiplier, δ, was introduced to scale the overall population immune proportion (δ=0 corresponds to complete immune escape, whereas δ=1 corresponds to no loss of population immunity). The epidemic was seeded by setting Ia’(0) = s and Sa’(0) = Sa(0) - s, where s is the initial seed size and Sa(0) is the number of individuals of age group a who are initially susceptible. Note that it is possible to vary the seed size, date and age group. We kept seeding in the youngest age group at 1000 initial infections in all scenarios, varying only the initial seed size. Contact matrices over time Base contact matrices were calculated from the POLYMOD UK data aggregated to the four age groups (0 to 5, 5 to 18, 18 to 65 and 65+ years). We generated four contact matrices for different periods of time: 1) regular school term-time; 2) half-term with no school contacts and reduced school contacts; 3) pre-Christmas shopping period (1-15 December) with an overall increase in non-school contacts; and 4) the Christmas school holiday period with no school contacts, a reduction in all contacts, and a substantial increase in at-home contacts. These four matrices were constructed by resampling the original POLYMOD contact diary entry data from the socialmixr R package with replacement and applying multipliers for home, work and other contact types. Holiday dates were based on the Oxfordshire school holiday period, though we note that holiday dates vary over the country, and even more so over the UK. For all non-school contacts, the relative changes in contacts over half-term holidays and the Christmas period (shopping and holiday) were extrapolated from 2021-2023 data from Kendall et al. Science 2024 (see Figures 5A and 7B in the paper) [33]. Absolute changes in overall non-school contact rates were extrapolated from the same source (see Figures 1A, 7A in the paper). To transition between these four different contact matrices, we generated a single contact matrix for each day, taking a weighted average of the four matrices depending on the time. For example, contacts in the middle of the school term were entirely governed by the term-time matrix (weighting of 1, all other matrices weighting 0), whereas contacts in the middle of the Christmas school holiday were governed entirely by the Christmas school holiday matrix. To ensure smooth transitions between these contact patterns, we smoothed the transition of weightings over 7 days before and after the holiday period using a cosine function. Baseline model calibration Model parameters were chosen based on standard seasonal influenza parameter values (R0 of 2, infectious period of 4-5 days, and final size of around 15% [46]), intuition and manual calibration to generate seasonal dynamics similar to what was seen in the 2022/23 season. We note that this is a complex model with a large number of parameters, making formal model fitting extremely difficult. Some of the parameters are hard to identify and interpret, such as the overall fraction of symptomatic cases reported, the symptomatic fraction by age combined with age-specific reporting rates, and the level of immune escape of the seed virus. Parameter values used for the baseline scenario are shown in Table 4. Scenario analyses Scenario analyses were chosen to illustrate potential hypotheses for the early and rapid growth of A/H3N2 cases in England for the 2025/26 season. We also varied the immune escape scaling parameter δ, the basic reproduction number R0, the proportion of the 0-4 and 5-18 year old population initially immune, and the seed date were varied invariably to generate Figure 6. Implementation All analyses were implemented and run in R version 4.2.2. The compartmental model was also implemented as a Shiny app with user-friendly sliders to change key parameter values, available at: https://hay-idd.shinyapps.io/ModelFluUk-H3N2/. Table 4. Model parameters assumed for the baseline scenario. These are the default parameters in the interactive web tool. Parameter Assumed value R0: basic reproduction number 2 Tg: infectious period 4 days γ: immune escape multiplier 1 Seed date 10th September Seed size in age group 1 (0-4 yrs) 1000 Population size 60,000,000 Proportion of work contacts kept in school holidays 0.75 Multiplier for home contacts in school breaks 1 Multiplier for non-school and non-work contacts in school breaks 1.1 Multiplier for all non-school contacts in Christmas period (1-15 December) 1.3 Proportion of all contacts kept over Christmas 0.67 Multiplier for home contacts over Christmas holiday 3 Proportion initially immune (0-4 yrs) 0.3 Proportion initially immune (5-18 yrs) 0.6 Proportion initially immune (19-64 yrs) 0.7 Proportion initially immune (65+ yrs) 0.75 Symptomatic fraction and reporting rate (0-4 yrs) 0.05 Symptomatic fraction and reporting rate (5-18 yrs) 0.2 Symptomatic fraction and reporting rate (19-64 yrs) 0.35 Symptomatic fraction and reporting rate (65+ yrs) 0.45 References 1. auspice. [cited 19 Nov 2025]. Available: https://nextstrain.org/seasonal-flu/h3n2/ha/2y?dmin=2024-11-29&p=grid 2. Huddleston J, Chang J, Lee J, Steinberg P, Neher R, Bedford T. Seasonal influenza circulation patterns and projections for September 2025 to September 2026. Zenodo; 2025. doi:10.5281/ZENODO.17281059 3. Allen JD, Ross TM. H3N2 influenza viruses in humans: Viral mechanisms, evolution, and evaluation. Hum Vaccin Immunother. 2018;14: 1840–1847. 4. Fleming DM, Elliot AJ. Lessons from 40 years’ surveillance of influenza in England and Wales. Epidemiol Infect. 2008;136: 866–875. 5. Kirsebom FCM, Thompson C, Talts T, Kele B, Whitaker HJ, Aziz NA, et al. Early influenza virus characterisation and vaccine effectiveness in England in autumn 2025, a period dominated by influenza A(H3N2) subclade K. 2025. Available: https://assets.publishing.service.gov.uk/media/691349457a0ccd6a3aad7fa9/_Flu_interimVE_2 526_.pdf 6. Iacobucci G. Flu in numbers: NHS faces one of worst winters ever, officials warn, amid concern over mutated strain. BMJ. 2025;391: r2391. 7. European Centre for Disease Prevention and Control. Threat Assessment Brief - Assessing the risk of influenza for the EU/EEA in the context of increasing circulation of A(H3N2) subclade K - 20. ECDC. 2025. 8. Dashboard · NINDSS Portal. [cited 18 Nov 2025]. Available: https://nindss.health.gov.au/pbi-dashboard/ 9. Australia posts record-breaking flu numbers as vaccination rates stall. [cited 18 Nov 2025]. Available: https://www.racgp.org.au/gp-news/media-releases/2025-media-releases/october-2025/australi a-posts-record-breaking-flu-numbers-as-vac 10. Fieldhouse R. Japan declares a flu epidemic - what this means for other nations. Nature. 2025 [cited 18 Nov 2025]. doi:10.1038/d41586-025-03367-z 11. Perofsky AC, Huddleston J, Hansen C, Barnes JR, Rowe T, Xu X, et al. Antigenic drift and subtype interference shape A(H3N2) epidemic dynamics in the United States. 2024. doi:10.7554/elife.91849.1 12. Dalziel BD, Kissler S, Gog JR, Viboud C, Bjørnstad ON, Metcalf CJE, et al. Urbanization and humidity shape the intensity of influenza epidemics in U.S. cities. Science. 2018;362: 75–79. 13. Biggerstaff M, Cauchemez S, Reed C, Gambhir M, Finelli L. Estimates of the reproduction number for seasonal, pandemic, and zoonotic influenza: a systematic review of the literature. BMC Infect Dis. 2014;14: 480. 14. UK Health Security Agency. National flu and COVID-19 surveillance reports: 2025 to 2026 season. GOV.UK; 2025. Available: https://www.gov.uk/government/statistics/national-flu-and-covid-19-surveillance-reports-2025-t o-2026-season 15. Weekly Influenza and Acute Respiratory Infection Report. In: Public Health Wales [Internet]. [cited 20 Nov 2025]. Available: https://phw.nhs.wales/topics/immunisation-and-vaccines/fluvaccine/weekly-influenza-and-acut e-respiratory-infection-report/ 16. Public Health Scotland. Influenza. [cited 20 Nov 2025]. Available: https://publichealthscotland.scot/population-health/health-protection/infectious-diseases/influen za/data-and-surveillance/ 17. Respiratory surveillance report. 12 Nov 2025 [cited 20 Nov 2025]. Available: https://www.publichealth.hscni.net/services-and-teams/public-health-services/health-protection /surveillance-data/respiratory-0 18. Eales O, McCaw JM, Shearer FM. Biases in routine influenza surveillance indicators used to monitor infection incidence and recommendations for improvement. Influenza Other Respi Viruses. 2024;18: e70050. 19. flunetchart. [cited 7 Nov 2025]. Available: https://worldhealthorg.shinyapps.io/flunetchart/ 20. Eales O, Windecker SM, McCaw JM, Shearer FM. Inferring temporal trends of multiple pathogens, variants, subtypes or serotypes from routine surveillance data. Am J Epidemiol. 2025; kwaf119. 21. UKHSA data dashboard. [cited 24 Nov 2025]. Available: https://ukhsa-dashboard.data.gov.uk/respiratory-viruses/influenza 22. Vega T, Lozano JE, Meerhoff T, Snacken R, Mott J, Ortiz de Lejarazu R, et al. Influenza surveillance in Europe: establishing epidemic thresholds by the moving epidemic method. Influenza Other Respi Viruses. 2013;7: 546–558. 23. Timing of influenza seasons. In: GOV.UK [Internet]. 28 Nov 2022 [cited 23 Nov 2025]. Available: https://www.gov.uk/government/publications/respiratory-virus-circulation-england-and-wales/ti ming-of-influenza-seasons 24. Bedford T, Riley S, Barr IG, Broor S, Chadha M, Cox NJ, et al. Global circulation patterns of seasonal influenza viruses vary with antigenic drift. Nature. 2015;523: 217–220. 25. Sabaiduc S, Kaweski SE, Separovic L, Gao R, Ranadheera C, Bastien N, et al. Emergence of seasonal influenza A(H3N2) variants with immune escape potential warrants enhanced molecular and epidemiological surveillance for the 2025–2026 season. Journal of the Association of Medical Microbiology and Infectious Disease Canada. 2025. doi:10.3138/jammi-2025-0025 26. Kikawa C, Huddleston J, Loes AN, Turner SA, Lee J, Barr IG, et al. Near real-time data on the human neutralizing antibody landscape to influenza virus to inform vaccine-strain selection in September 2025. bioRxivorg. 2025. doi:10.1101/2025.09.06.674661 27. Scott J, Abers MS, Marwah HK, McCann NC, Meyerowitz EA, Richterman A, et al. Updated evidence for Covid-19, RSV, and influenza vaccines for 2025-2026. N Engl J Med. 2025. doi:10.1056/NEJMsa2514268 28. Rose AM, Lucaccioni H, Marsh K, Kirsebom F, Whitaker H, Emborg H-D, et al. Interim 2024/25 influenza vaccine effectiveness: eight European studies, September 2024 to January 2025. Euro Surveill. 2025;30. doi:10.2807/1560-7917.ES.2025.30.7.2500102 29. te Beest DE, van Boven M, Hooiveld M, van den Dool C, Wallinga J. Driving factors of influenza transmission in the Netherlands. Am J Epidemiol. 2013;178: 1469–1477. 30. Shaman J, Pitzer VE, Viboud C, Grenfell BT, Lipsitch M. Absolute humidity and the seasonal onset of influenza in the continental United States. PLoS Biol. 2010;8: e1000316. 31. Cauchemez S, Valleron A-J, Boëlle P-Y, Flahault A, Ferguson NM. Estimating the impact of school closure on influenza transmission from Sentinel data. Nature. 2008;452: 750–754. 32. van Noort SP, Águas R, Ballesteros S, Gomes MGM. The role of weather on the relation between influenza and influenza-like illness. J Theor Biol. 2012;298: 131–137. 33. Kendall M, Ferretti L, Wymant C, Tsallis D, Petrie J, Di Francia A, et al. Drivers of epidemic dynamics in real time from daily digital COVID-19 measurements. Science. 2024;385: eadm8103. 34. Howerton E, Contamin L, Mullany LC, Qin M, Reich NG, Bents S, et al. Evaluation of the US COVID-19 Scenario Modeling Hub for informing pandemic response under uncertainty. Nat Commun. 2023;14: 7260. 35. Abbott S, Hellewell J, Thompson RN, Sherratt K, Gibbs HP, Bosse NI, et al. Estimating the time-varying reproduction number of SARS-CoV-2 using national and subnational case counts. Wellcome Open Res. 2020;5: 112. 36. Data quality report: national flu and COVID-19 surveillance report. In: GOV.UK [Internet]. [cited 22 Nov 2025]. Available: https://www.gov.uk/government/publications/sources-of-surveillance-data-for-influenza-covid-1 9-and-other-respiratory-viruses/data-quality-report-national-flu-and-covid-19-surveillance-repor t#about-this-report 37. Yuan H-Y, Baguelin M, Kwok KO, Arinaminpathy N, van Leeuwen E, Riley S. The impact of stratified immunity on the transmission dynamics of influenza. Epidemics. 2017;20: 84–93. 38. EpiStrainDynamics. Github; Available: https://github.com/acefa-hubs/EpiStrainDynamics 39. Wood S. Mixed GAM Computation Vehicle with Automatic Smoothness Estimation [R package mgcv version 1.9-4]. In: Comprehensive R Archive Network (CRAN) [Internet]. 7 Nov 2025 [cited 22 Nov 2025]. Available: https://cran.r-project.org/web/packages/mgcv/index.html 40. Cori A, Ferguson NM, Fraser C, Cauchemez S. A new framework and software to estimate time-varying reproduction numbers during epidemics. Am J Epidemiol. 2013;178: 1505–1512. 41. Cowling BJ, Fang VJ, Riley S, Malik Peiris JS, Leung GM. Estimation of the serial interval of influenza. Epidemiology. 2009;20: 344–347. 42. Wallinga J, Lipsitch M. How generation intervals shape the relationship between growth rates and reproductive numbers. Proc Biol Sci. 2007;274: 599–604. 43. Social Mixing Matrices for Infectious Disease Modelling [R package socialmixr version 0.4.0]. In: Comprehensive R Archive Network (CRAN) [Internet]. 18 Oct 2024 [cited 22 Nov 2025]. Available: https://cran.r-project.org/web/packages/socialmixr/ 44. Mossong J, Hens N, Jit M, Beutels P, Auranen K, Mikolajczyk R, et al. Social Contacts and Mixing Patterns Relevant to the Spread of Infectious Diseases. Riley S, editor. PLoS Med. 2008;5: e74. 45. Solvers for Initial Value Problems of Differential Equations (’ODE', “DAE”, “DDE”) [R package deSolve version 1.40]. In: Comprehensive R Archive Network (CRAN) [Internet]. 27 Nov 2023 [cited 22 Nov 2025]. Available: https://cran.r-project.org/web/packages/deSolve/index.html 46. Yang W, Lipsitch M, Shaman J. Inference of seasonal and pandemic influenza transmission dynamics. Proc Natl Acad Sci U S A. 2015;112: 2723–2728. 47. RCGP. Public health data. [cited 22 Nov 2025]. Available: https://www.rcgp.org.uk/representing-you/research-at-rcgp/research-surveillance-centre/public -health-data 48. Virology Dashboard. [cited 22 Nov 2025]. Available: Figure S5. Comparison of weekly growth rate estimates using the EpiStrainDynamics R package fitted to UKHSA (Respiratory DataMart) and WHO FluNet samples. Shaded ribbons show posterior means, 50% and 95% credible intervals. Estimates during the COVID-19 pandemic are not shown. The top panel shows weekly growth rates of A/H3N2 samples. The bottom panel shows the weekly growth rate of all influenza samples. Figure S6. Fits of the penalised spline model to empirical weekly growth rates from the WHO FluNet data. Purple lines and ribbons show posterior mean and 95% CrI. The blue line shows the raw log weekly growth rates. The yellow line shows the raw log weekly growth rates smoothed over 4-week intervals (right aligned). Figure S7. Fits of the random walk model to empirical weekly growth rates from the WHO FluNet data using only A/H3N2 cases. Purple lines and ribbons show posterior mean and 95% CrI. The blue line shows the raw log weekly growth rates. The yellow line shows the raw log weekly growth rates smoothed over 4-week intervals (right aligned). Figure S8. Growth rate estimates using a penalised spline model matching Figure 2, but instead aligned by the week of peak growth rate as x=0. Figure S9. Growth rate of all influenza cases from the WHO FluNet database. Coloured lines and shaded regions show posterior mean and 95% credible intervals for the model-estimated weekly growth rate. Colouring distinguishes the current season from post, pre and during pandemic seasons. Doubling time is shown on the right hand y-axis. Figure S10. Fits of the random walk model to empirical weekly growth rates from the WHO FluNet data using all influenza cases. Purple lines and ribbons show posterior mean and 95% CrI. The blue line shows the raw log weekly growth rates. The yellow line shows the raw log weekly growth rates smoothed over 4-week intervals (right aligned). Figure S11. Growth rate estimates of age-stratified A/H3N2 cases from the ILI+ indicator matching Figure 2. Shown are the empirical growth rates and model fits using the GAM. School holidays are marked with grey bars. The autumn half-term break (i.e., the recent school holiday) is marked in red. Solid lines and ribbons show mean estimates and 95% confidence intervals. Figure S12. Comparison of age-stratified growth rates from the ILI+ indicator by season. Lines show the posterior mean estimates from the penalised spline model in Figure 2. Estimates from each season are aligned to the week since the start of the influenza season. The growth rate difference is calculated as the absolute difference between the growth rate in the focal age group to the 15-44 year old age group, such that a line significantly above 0 indicates much higher growth rates than younger adults. Figure S13. Effective reproduction number (Rt ) for each influenza season arranged relative to peak pre-Christmas Rt. Estimates shown are identical to Figure 4, but plotted together to enable visual comparison. Figure S14. Comparison of symptomatic influenza incidence in 65+ from the scenario analyses shown in Figure 5.