For TND studies, the start of the study period for IVE analyses for the DRIVE studies was defined as the first week of two consecutive weeks when influenza viruses were detected at the study contributor level (based on the data provided to DRIVE), and the end as the week prior to the first of two consecutive weeks when no influenza viruses were detected at the study contributor level (based on the data provided to DRIVE) or 30 April 2022, whichever occurred first. In the population-based cohort study, the study period was defined a priori from 4 October 2021 to 30 April 2022.

In 2021–2022, a bimodal influenza season was observed, with different influenza peaks: a smaller one in December 2021–January 2022 and a larger one in March–April 2022. As a consequence, three scenarios with respect to the study period for analysis were observed: (i) the first peak did not lead to enough cases to trigger the study start definition but only the second peak did (e.g., Italy CIRI-IT BIVE); (ii) the first peak triggered the study start definition and the occasional influenza cases observed between the first and the second peaks ensured that the study end definition was not triggered before the second peak (e.g., FISABIO); and (iii) there was a very minor first peak that triggered the study start definition with no cases until the second wave started; in this case, only the second wave was considered (e.g., NIID).

For the TND studies, the study population consisted of non-institutionalized subjects ≥6 months of age with no contraindication for influenza vaccination, no prior positive influenza test in the same season, and a swab taken <8 days after ILI/SARI onset. In hospital settings, subjects hospitalized <48 h prior to symptom onset or with symptom onset ≥48 h after hospital admission were excluded. For each subject, data on covariates (at least age, sex, and date of symptom onset), vaccination status, and vaccine brands for vaccinated subjects were collected. Cases and controls were considered fully vaccinated if they had received seasonal influenza vaccination >14 days before the start of the ILI/SARI episode.

For the population-based cohort, the study population consisted of all registered Finnish residents aged 6 months−6 years and 65–100 years. The case definition was laboratory-confirmed influenza, as registered in the National Infectious Diseases Register. Using the Care Register for Health Care, it was possible to identify laboratory-confirmed influenza cases who were inpatients for any reason on the day of laboratory confirmation.

Study contributor-specific IVE and 95% confidence intervals (CIs) for TND studies were calculated using logistic regression as IVE = (1 – OR) × 100%, where OR is the odds ratio comparing the odds of influenza among vaccinated and unvaccinated subjects. As supported by an exploratory analysis of the 2018–2019 and 2019–2020 DRIVE datasets, a parsimonious confounder adjustment was performed, which included sex, a smooth function of age, and a smooth function of the calendar week or date of symptom onset (5).

Study contributor-specific IVE estimates from the TND studies were pooled through a random-effects meta-analysis, which assumes that the observed effect estimates can vary across study sites because of differences in the treatment effect in each study site (e.g., due to differences in population, healthcare utilization, circulating influenza strains) in addition to sampling variability (6, 7). Study contributor-specific IVE estimates that were both outlying (studentized deleted residuals |r| >2.5) and influential (standardized |DFBETAs| >2/√n) were excluded from the pooled analysis; a sensitivity analysis including these estimates was conducted.

In the present study, we focused our attention on significant IVE estimates and those with a CI width <40% (if any), although for the sake of transparency, the rest of the IVE estimates can be consulted in the DRIVE 2021–2022 results report and the WebAnnex.

For the population-based cohort study, contributor-specific semi-crude estimates (adjusted only for calendar time) and confounder-adjusted estimates (adjusted as in the TND studies) were obtained. IVE and 95% CIs were estimated using Poisson regression. IVE was calculated as IVE = (1 – IRR) × 100%, where IRR is the incidence rate ratio comparing the incidence of influenza in vaccinated subjects to the incidence of influenza in unvaccinated subjects.

Each local study was approved by national, regional, or institutional ethics committees, as appropriate. All the studies were approved without the need for protocol revision; therefore, ethics committee approval was given in due time for the start of the DRIVE study. Written informed consent was obtained in two-thirds of the participating study sites; in the remaining one-third of study sites (i.e., THL, MUV, GTPUH, EL, HVUH), informed consent was not needed, as the DRIVE study was nested within the national/regional influenza surveillance systems.

A total of 12 influenza vaccines were licensed and marketed in the EU/European Economic Area (EEA)/UK for the season 2021–2022, and eight of these vaccines were captured in the DRIVE dataset. Details on vaccine characteristics, the approved age indication, and, for each age group, the countries that reported the vaccine brand in the 2021–2022 studies are listed in Table 2.

The recommended composition of influenza vaccines in the Northern Hemisphere for the 2021–2022 season can be consulted at the World Health Organization website (8).

The circulation of influenza was relatively low, and co-circulation of influenza and SARS-CoV-2 was observed. In most of the participating countries, the influenza season initially overlapped with the SARS-CoV-2 Omicron wave (from November 2021 to February 2022) and presented two peaks of higher activity: the first one in December 2021–January 2022 (peaking in week 52/2021) and a higher and unusually late peak in March–April 2022 (peaking between weeks 10 and 15 of 2022), which declined until May 2022 (9, 10).

Within the 2021–2022 DRIVE study, a total of 1,039 influenza cases were included in the analysis for the TND studies and 331 in the population-based cohort study. The number of influenza A cases exceeded the number of influenza B cases at all sites. Among influenza A cases with a known subtype, the most frequently identified subtype was A(H3N2) at all the participating sites (ranging from 47 to 100% of the subtyped viruses), in line with the data reported by the ECDC (9). Influenza B was only detected in Austria (0.6% of their influenza cases) and Finland (7%). All B cases included were of B/Victoria lineage, and no cases of B/Yamagata lineage were identified (11). Supplementary Table 1 describes the distribution of influenza cases by type and subtype for each study contributor.

The predominant A(H3N2) subclade was 3C.2a1b.2a.2, with a haemagglutinin (HA), with several substitutions that made it antigenically different from the A(H3N2) Cambodia lineage (3C.2a1b.2a.1) included in the 2021–2022 influenza vaccine. Therefore, a mismatch between the circulating A(H3N2) influenza strain and the strain included in that season's vaccine composition was observed.

A total of 12 influenza vaccine brands were marketed in the EU/EEA/UK for the 2021–2022 season, eight of which were included in the DRIVE dataset, highlighting the ability of the DRIVE study network to cover a representative sample of the European influenza vaccine market. All except one of the vaccines observed were quadrivalent, reflecting the transition from trivalent to quadrivalent vaccines. The two most frequently observed vaccine brands per age group were the same as for the 2020–2021 season (12). The brands that were not identified were Afluria Tetra (Seqirus), Chiroflu (Seqirus), Influvac (Abbott), and Supemtek (Sanofi; Table 2). Afluria Tetra was only distributed in Germany, a country not included in DRIVE, and Supemtek was only available in the UK, where only a very limited number of subjects were enrolled. Chiroflu and Influvac are both trivalent vaccines that are being phased out. The inclusion of study contributors that together capture a sufficiently diverse range of vaccine availability is a particular challenge for a network aiming to study brand-specific IVE.

DRIVE conducted TND studies to produce pooled, confounder-adjusted influenza vaccine effectiveness estimates against LCI overall and per type and subtype/lineage, by setting and age group. The most significant overall and brand-specific IVE estimates showed a protective effect and were obtained among older adults in the primary care setting and adults in the hospital setting for any influenza and influenza A types. Despite CDC's recommendation to not report VE estimates if the CI exceeds 50% in order to avoid reporting uninterpretable estimates (13), DRIVE partners decided to report the most significant overall and brand-specific VE estimates even though their CI exceeded 50%. The less significant VE estimates are also available in the DRIVE 2021–2022 results report and the WebAnnex for full transparency of the analysis performed.

The population-based cohort study in Finland generated IVE estimates for the two influenza vaccine brands used in Finland and overall IVE. Due to the small number of influenza cases in Finland, all IVE estimates presented a wide CI and, therefore, were not informative. Interestingly, the two different settings, the hospital setting and the mixed primary care and hospital setting yielded similar results, which is in line with a review by Feng et al. (14).

To the best of our knowledge, DRIVE is the only multi-country IVE study in Europe that generates brand-specific IVE. Nevertheless, several studies focusing on overall IVE estimates for the 2021–2022 season were identified.

From 2007–2008 until 2021–2022, the I-MOVE (Influenza Monitoring Vaccine Effectiveness in Europe) consortium (15, 16) has conducted, in collaboration with the ECDC, annual studies to evaluate the effectiveness and impact of the influenza vaccines in Europe. The primary care-based I-MOVE study produced overall IVE estimates from 10 primary care study sites between October 2021 and mid-May 2022 (17). I-MOVE estimates among adults ≥65 years were 26% (95% CI −22 to 55) against influenza A(H3N2) and 23% (95% CI −21 to 51) against any influenza A, whereas, for DRIVE, the IVE estimate at the primary care setting for ≥65 years was 76% (95% CI 23–93) against any influenza A and A(H3N2). While the point estimates obtained from DRIVE and I-MOVE were different, the 95% of CIs overlapped, preventing meaningful comparisons.

Moreover, in children aged 2–6 years in Denmark, VE was estimated at 64.2% (95% CI 50.5–74.1) against non-hospitalized influenza A and 63% (95% CI 10.9–84.4) against hospitalized influenza A; 92% of vaccinated children had received the live-attenuated influenza vaccine (LAIV) (18). In the UK, the VE of LAIV against influenza requiring an emergency department visit was 72% (95% CI 50–85) (19). In DRIVE, the point estimates for LAIV were 63% (95% CI 23–82) against influenza A in the TND primary care studies and 40% (95% CI −2.9 to 64.5) against any influenza in any setting among children 2–6 years in the Finnish population-based cohort study. Significant comparisons across these studies are hindered due to variations in study design, the outcomes examined, and the overlap in the 95% CIs of the point estimates.

Finally, VE against all-age outpatient influenza was estimated at 50% (95% CI 14–71) for any influenza and 31% (95% CI −29 to 64) for A(H3N2) in France (20); and VE against laboratory-confirmed influenza was estimated at 47% (CI not reported) for adults ≥65 years in Sweden (mixed inpatient and outpatient setting) (21), which is higher than the point estimates for this age group in DRIVE.

The discrepancies between the DRIVE estimates and those obtained from different studies may be partially explained by the differences among the participating countries in each study (different vaccine recommendations and virus circulation per country) and study settings. In addition, most of the abovementioned VE estimates had large CIs, so comparison across studies is complex. These caveats underline the necessity of collaboration between VE platforms and the harmonization of the methods used for larger and more representative European studies.

The sample size required for brand-specific IVE studies is large, due to the numerous vaccine brands and the multiple stratifications required. Given the limitations in sample size, overall and brand-specific estimates produced by DRIVE had wide confidence intervals, limiting their potential to be informative. In addition, the COVID-19 pandemic led to reduced influenza virus circulation, a shift in interest, and an overload of staff dedicated to the studies. Furthermore, the inability of potential collaborators to collect information on influenza vaccine brands, the presence of other overlapping VE networks, and the hesitancy toward the public–private governance model perceived in certain sectors of public health institutions were also limiting factors.

Other than the sample size, multiple factors affect the precision of the estimates, such as vaccination coverage and influenza attack rate, but also the true IVE, test sensitivity and specificity, statistical methods, and the heterogeneity of study contributor-specific IVE estimates. Given the width of the CIs of the majority of the IVE estimates generated in the 2021–2022 season, these should be interpreted with caution.

The availability of information on the presence of chronic conditions was limited for most of the (vaccinated) subjects and did not allow a solid assessment of the impact of additional adjustment for this variable. However, in an exploratory analysis of previous DRIVE data, additionally adjusting for the presence of at least one chronic condition did not have an important impact on VE estimates (5). Furthermore, given the low number of coinfections with SARS-CoV-2 (n = 29), no sensitivity analysis stratified by SARS-CoV-2 status was conducted.

In addition, VE may change during the influenza season due to waning immunity, but we could not study this in DRIVE due to the limited sample size.

Due to the observational nature of the studies, selection bias cannot be ruled out. In TND studies, the relationship between influenza vaccination and testing positive for influenza among those who are tested is studied. The use of a TND, therefore, reduces confounding due to healthcare-seeking behavior but introduces selection bias and may reduce the generalizability to the general population (22).

Healthcare-seeking behavior has changed during the COVID-19 pandemic, through different periods of the pandemic, and the changes are probably not uniform across countries. Therefore, it is not clear how this affects the TND and the subsequent IVE estimates. Finally, residual confounding may be present due to measurement errors in the confounders and other unmeasured confounders.

DRIVE started with a pilot season (2017–2018) to establish the multi-country platform for TND and cohort studies using a limited number of sites (Table 6). The initial study contributors participated with data collected using their own protocols (23). As of 2018–2019, generic protocols were implemented by all sites and, along with the statistical analysis plan, were iteratively improved each season. An IT platform where study contributors could perform data quality checks and upload their anonymized dataset to a secure environment and where data were centrally analyzed was developed. The size of the study network increased from five sites in four countries in 2017/2018 to 13 sites in eight countries in 2021–2022. As of 2018–2019 (24), the network was sufficiently geographically diverse to capture the majority of influenza vaccine brands marketed in Europe (Table 6). From the 2018/2019 season onwards, DRIVE also supported the implementation and assessed the feasibility of using point-of-care PCR testing (POCT) to improve sampling numbers from primary care in England (25–27).

As a result of a post hoc analysis based on the 2018–2019 data (24), and as recommended by Lane et al. (28), a parsimonious set of confounders was defined to simplify the analysis. This approach is also supported by a recent exploratory analysis conducted with DRIVE data (5). The reduced set of confounders was used in the DRIVE studies from the 2019–2020 season and allowed the participation of study contributors with limited data on confounders, in addition to limiting the data discarded due to missing values and avoiding potential over-adjustment. Suggested confounding variables for the main analysis included age, sex, and calendar time, but sensitivity analyses including all potential confounders available (presence of chronic disease, number of hospitalizations and GP visits in the past 12 months, pregnancy) were also performed.

Throughout the existence of DRIVE, attaining a sufficient sample size to obtain meaningful estimates for brand-specific estimates, with their relevant strata, has been a challenge. DRIVE aimed to obtain precise estimates, defined as a CI width of <40%, although the level of precision required for decision-making by regulators and public health bodies remains unclear. Precise brand-specific estimates were obtained for Vaxigrip Tetra, Fluarix Tetra, and Fluad in the 2019/20 TND studies and for Fluenz Tetra and Vaxigrip Tetra in 2018–2019 and 2019–2020 in the population-based cohort (24, 29).

The relatively low influenza circulation, as a consequence of the non-pharmaceutical interventions implemented to fight the COVID-19 pandemic and the shift of attention and resources to COVID-19, largely impacted the 2020–2021 and 2021–2022 seasons, obstructing DRIVE from generating robust brand-specific IVE estimates in its last two seasons (30). The DRIVE study platform was adjusted to the COVID-19 pandemic context, and the core protocols and SAP were tailored with the support of the DRIVE partners, study contributors, and the Independent Scientific Committee. Furthermore, new variables related to SARS-CoV-2 testing, infection, treatment, and vaccination were collected.

Finally, the DRIVE infrastructure, study network, and governance model were leveraged to launch COVIDRIVE in 2021, a public–private partnership that currently brings together 12 partners and aims to conduct multi-country European studies to monitor brand-specific COVID-19 vaccine effectiveness (CVE) in real-world conditions (31). The COVIDRIVE partnership was set up in only 9 months thanks to the existence of the DRIVE study platform and partner collaborations. As of February 2023, the COVIDRIVE study is active in 15 hospitals (Belgium, Austria, Italy, and Spain), which have enrolled over 7,000 SARI patients and more than 3,000 COVID-19 cases.