Discussion

During the 2022–2023 influenza season in France, HD in nursing home residents reduced influenza-related hospitalisations by 30.8% (95%CI, 11.8–45.7) compared to SD. Results remained consistent across the range of sensitive analyses conducted. This is in line with a previous research conducted in France during the 2021–2022 influenza season in community-dwelling populations, which reported a rVE of 23.3% (95%CI, 8.4–35.8) [22]. Additionally, our findings align with a cluster randomised trial showing a 12.7% (95%CI, 1.8–22.4) reduction in respiratory illness-related hospitalisations [19]. Notably, our study population was older (mean age of 88.0 years in our cohort vs. 83.6 years in Gravenstein et al.) and had lower comorbidity rates (diabetes (17% vs. 34%; COPD/asthma 10.6% vs. 20%)) [19]. Results for pertaining to less specific endpoints, including pneumonia and P/I hospitalisation, were non-significant regardless of outcome definitions. Nonetheless, the effect size reported here against influenza-related hospitalisation is clinically meaningful: in a population where hospitalisation usually translates into either severe disability or death following discharge [5, 7], a 30.8% reduction could translate into 260 avoided hospitalisations, fewer complications and reduced burden on families and healthcare systems.

A major strength of this study is its large, nationwide coverage made possible by the access to the SNDS database. This enabled the study to capture most of the HD vaccine doses distributed in France during the 2022–2023 season. We employed published algorithms to identify most of the individuals’ characteristics, comorbidities and medical history. Increased use of PCR testing for influenza since the COVID-19 pandemic [38, 39] likely improved specificity of influenza coding in hospital records and reduced reliance on broader codes potentially explaining the non-significant results for non-specific endpoints compared to the cluster randomised setting [19]. These strengths contribute to the validity and reliability of our findings regarding influenza hospitalisations, as laboratory-confirmed influenza hospitalisation is the most specific endpoint for evaluating vaccine effectiveness [40–42]. The specificity of this endpoint is crucial in our observational study, given the potential for residual confounding and the high SARS-CoV-2 circulation during the study period. Moreover, while unmeasured confounding cannot be ruled out in an observational study, the NCO analyses yielded non-significant IRR regardless of the outcome, suggesting minimal residual bias.

In this setting, vaccine exposure was not comparable to that observed in community. Indeed, it is a reasonable assumption that a specific nursing home decided to use exclusively HD or SD to vaccinate their residents. As mentioned, the variable indicating the nursing home unit was unavailable, making impossible to conduct a cluster-level analysis, which might have been more appropriate in this context. The analysis was affected by the additional consideration that influenza circulation would likely occur in a cluster effect for nursing homes. Consequently, rVE analyses between HD and SD could not account for comparable influenza circulation between different nursing homes and other disparities at the geographical and unit-specific practices levels (including different vaccine coverage rates for healthcare professionals). This bias was limited by including the department of residence in the matching procedure.

Moreover, our study was limited to residents of nursing homes without in-house pharmacies, because in facilities with such pharmacies, individual vaccine deliveries cannot be tracked through the SNDS claims data. According to the 2019 EHPA study by the French Ministry of Health’s DREES, these nursing homes represent up to 77.5% of all such facilities in France, encompassing an estimated 469 369 residents. In our study, we identified 315 239 individuals who received an influenza vaccine. This corresponds to a theoretical vaccine coverage rate of 67.2%, based on the EHPA population [29]. This estimate is lower than the 87% influenza vaccine coverage reported by Santé publique France in a survey of 108 000 residents, which likely included facilities with in-house pharmacies where vaccine coverage may be higher [9]. Consequently, the total number of vaccinated nursing home residents aged 65 and older captured in our dataset is likely underestimated.

According to EHPA data, patient severity, dependency (GIR scale) and in-hospital mortality are broadly similar across facility types [29, 43] supporting generalizability of our findings to the overall nursing home population.

Two thirds (63.4%) of HD recipients were successfully matched to a SD recipient. Analogous pre- and post-matching characteristics make selection bias due to unmatched individuals unlikely. Matching constraints (exact matching variables: sex, age groups, region and vaccination week) prevented the planned 1:4 ratio. A 1:1 ratio was retained for interpretive clarity, with variable ratio matching confirming consistent results.

The limitations related to the use of administrative databases for research purposes also applied to our study. Indeed, we may not have captured all relevant health characteristics of the study individuals. This limits the ability of our matching techniques to adjust for confounding, as without robust estimators of frailty or level of severity for the comorbidity, it is difficult to adjust for these confounding factors, which may lead to an underestimation of vaccine effectiveness. Of note, because the analysis matches HD recipients to SD recipients, the estimated rVE represents the average treatment effect in the treated. These results are therefore most applicable to nursing home residents who received HD vaccination, and may not generalise to all nursing home residents, including those for whom HD was not selected. Other unmeasured confounders missing from administrative databases could have impacted our findings if they were unbalanced between HD and SD recipients. Similarly, the outcome definitions used relied on ICD-10 codes within hospital stays as the database does not include laboratory results (such as influenza or COVID-19 tests results). It can also include hospital-acquired influenza irrespective of SD or HD recipients. Conversely, it does not include patients with influenza who were not hospitalised because deemed too severe. Additionally, mortality during follow-up represents a competing risk for hospitalisation in this population. Future analyses using competing-risk frameworks could further refine these estimates. Furthermore, the 14-day lag period applied to the outcomes means that any very early effect (or differential reactogenicity) of HD versus SD in the first two weeks post-vaccination is not captured by the main analysis. Both numerator (events) and denominator (person-time) correctly exclude this period, but this remains a limitation in terms of generalisability to the earliest post-vaccination window. Furthermore, the exclusion of individuals with a hospitalisation for the outcomes of interest during the pre-vaccination period or lag window (180 736/8 418 163 [2.1%]) means our study population represents a slightly healthier subgroup of nursing home residents. This may limit generalisability of findings to the broader nursing home population. The present analysis focused on effectiveness outcomes and did not assess safety events associated with HD versus SD vaccines. Although existing evidence from clinical trials does not show a clinically significant increase in serious adverse events with HD formulations. Further investigations could also rely on pragmatic trial designs, to confirm these findings and assess broader outcomes (functional decline, quality of life).