key: cord-0994989-rvce4x5x authors: Kissling, Esther; Hooiveld, Mariëtte; Brytting, Mia; Vilcu, Ana‐Maria; de Lange, Marit; Martínez‐Baz, Iván; Sigerson, Debbie; Enkirch, Theresa; Belhillil, Sylvie; Meijer, Adam; Castilla, Jesus; William, Naoma; Carnahan, AnnaSara; Falchi, Alessandra; Hendriksen, Janneke; Casado, Itziar; Murray, Josie; Enouf, Vincent; Dijkstra, Frederika; Marques, Diogo F. P.; Valenciano, Marta title: Absence of association between 2019‐20 influenza vaccination and COVID‐19: Results of the European I‐MOVE‐COVID‐19 primary care project, March‐August 2020 date: 2021-01-22 journal: Influenza Other Respir Viruses DOI: 10.1111/irv.12839 sha: 7b8c7742a38886d300ac26432035392a0bdf7354 doc_id: 994989 cord_uid: rvce4x5x BACKGROUND: Claims of influenza vaccination increasing COVID‐19 risk are circulating. Within the I‐MOVE‐COVID‐19 primary care multicentre study, we measured the association between 2019‐20 influenza vaccination and COVID‐19. METHODS: We conducted a multicentre test‐negative case‐control study at primary care level, in study sites in five European countries, from March to August 2020. Patients presenting with acute respiratory infection were swabbed, with demographic, 2019‐20 influenza vaccination and clinical information documented. Using logistic regression, we measured the adjusted odds ratio (aOR), adjusting for study site and age, sex, calendar time, presence of chronic conditions. The main analysis included patients swabbed ≤7 days after onset from the three countries with <15% of missing influenza vaccination. In secondary analyses, we included five countries, using multiple imputation with chained equations to account for missing data. RESULTS: We included 257 COVID‐19 cases and 1631 controls in the main analysis (three countries). The overall aOR between influenza vaccination and COVID‐19 was 0.93 (95% CI: 0.66‐1.32). The aOR was 0.92 (95% CI: 0.58‐1.46) and 0.92 (95% CI: 0.51‐1.67) among those aged 20‐59 and ≥60 years, respectively. In secondary analyses, we included 6457 cases and 69 272 controls. The imputed aOR was 0.87 (95% CI: 0.79‐0.95) among all ages and any delay between swab and symptom onset. CONCLUSIONS: There was no evidence that COVID‐19 cases were more likely to be vaccinated against influenza than controls. Influenza vaccination should be encouraged among target groups for vaccination. I‐MOVE‐COVID‐19 will continue documenting influenza vaccination status in 2020‐21, in order to learn about effects of recent influenza vaccination. The end of 2019 saw the emergence of a novel severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), which causes coronavirus disease 2019 . Europe has been heavily affected, with 14 750 809 cases and 336 422 deaths reported from Europe between 31 December 2019 and 18 November 2020. 1 The second wave in autumn is causing increasing trends in hospital and intensive care unit admissions. 2 In the average winter season, hospital bed occupancy is higher, due to influenza patients. 3, 4 In the 2020-21 winter season with the ongoing COVID-19 pandemic, it is crucial to reduce the burden on hospitals, as much as possible. Influenza vaccination, despite its sometime moderate vaccine effectiveness, is an important factor in reducing hospitalisations 5 and is specifically important for the 2020-21 season with SARS-CoV-2 circulation. However, claims of influenza vaccination increasing risk are circulating, particularly on social media. An article on influenza vaccination increasing the risk of seasonal coronaviruses 6 has fuelled these claims, despite its results having been refuted. 7 Additionally, ecological studies correlating influenza vaccination uptake rates and COVID-19 mortality, published in scientific journals as articles or rapid responses to articles sustain these hypotheses. 8, 9 Conversely, several articles have been published in scientific journals on the negative association or no association between influenza vaccination and COVID-19 mortality, hospitalisation and infection, 10, 11 many of which are summarised in a recent systematic literature review. [12] [13] [14] [15] The authors of the latter paper recommend that further studies are carried out to validate these preliminary findings across settings. The I-MOVE-COVID-19 consortium is a network of European sites and countries, many of which are part of the I-MOVE influenza vaccine effectiveness studies. 16 One component of the I-MOVE-COVID-19 project is a primary care multicentre study on Within this study, we aimed to measure the association between 2019-20 influenza vaccination and COVID-19. We conducted a multicentre test-negative case-control study at primary care level, in study sites in five European countries: France (FR), the Netherlands (NL), Scotland (SC), Spain (Navarra, NA) and Sweden (SE). The systems for data collection vary across sites. Briefly, ambulatory (non-hospitalised) patients with acute respiratory infection (ARI) were swabbed for SARS-CoV-2 and demographic and clinical information was collected, including 2019-20 influenza vaccination status. Sentinel general practitioner (GP) networks in well-established influenza surveillance systems in FR, NL and SE identified ARI patients. In SC, dedicated COVID-19 assessment centres were used to identify patients. In NA, ARI patients were identified from data from the Navarra Health Service. Clinical staff collected respiratory specimens face-to-face (NL, SE, NA, FR and SC), or specimens were received through self-swab (SC). Information was obtained from the patient by the GP during the consultation (NL, SE and FR), by clinical staff in COVID-19 centres (SC), by patient self-report (SC) or through electronic medical records (Navarra). The study period ranged from the ISO week of swab of the first SARS-CoV-2-positive case detected by the system, until the ISO week of swab of the last SARS-CoV-2-positive case up to the end of August 2020. We restricted the study until the end of August to reduce ambiguity as to which influenza vaccination to record: 2019-20 or 2020-21 season vaccination. For France, the study period ended at and including ISO week 20 of swab, as from week 21-40, influenza vaccination status was no longer collected. Cases were patients presenting with ARI at primary care level testing RT-PCR positive for SARS-CoV-2 and controls were those testing RT-PCR negative. Institutionalised patients, patients with inconclusive test results, missing swab date, not agreeing to participate and those testing positive to influenza were excluded. In sites where information on influenza testing was not available, we started the study period from 1 April onwards, under the assumption that little or no influenza was circulating (one site). All specimens were tested by real-time reverse transcription quantitative (RT-qPCR) methods. case-control study, COVID-19, influenza vaccination, multicentre study, SARS-CoV-2, testnegative design Seasonal influenza vaccination campaigns in Europe start in early autumn. 17 As the study period began in March 2020, after the peak of influenza in Europe, we assumed that all persons vaccinated with influenza vaccine in the study were vaccinated more than 14 days before symptom onset and could be considered fully vaccinated. For the primary analysis, we included only patients swabbed within 7 days of symptom onset, as PCR sensitivity is higher within this time range. 18 Additionally, only sites with <15% of missing 2019-20 influenza vaccination information (France, the Netherlands and Sweden) were included, to avoid a potential substantial bias due to missing data. In the primary analysis, we carried out a complete case analysis, dropping records with missing values for covariates. In a secondary analysis, we included all sites regardless of the number of days between onset of symptoms and swabbing or missing 2019-20 influenza vaccination information (all five sites). We pooled individual data and compared the odds of 2019-20 influenza vaccination between SARS-CoV-2 RT-PCR-positive and SARS-CoV-2 RT-PCR-negative patients, with study site as a fixed effect in the model, using logistic regression. We adjusted for a priori confounders to measure adjusted odds ratios (aOR). Potential confounders included age (as measured in 5-year age groups), sex, calendar time of swabbing, and presence of chronic conditions, including diabetes, heart disease, chronic lung disease and immunodeficiencies. The functional form (categories, continuous variable or restricted cubic spline) of 5-year age groups and calendar time was determined using the Akaike Information Criterion (AIC). We stratified the analysis by age group (<20 years, 20-59 and ≥60 years) and sex. To obtain the aOR by 5-year age groups, we added an interaction between 2019-20 influenza vaccination and 5-year age groups modelled as a restricted cubic spline, adjusting for other confounders, with location of knots as defined by Harrell 19 . The number of knots was determined by the AIC, avoiding any models where the standard errors of a coefficient were greater than the coefficient themselves. In order to test the robustness of our results, we carried out several sensitivity analyses. In sensitivity analyses in the primary analysis, we computed the aOR between 2019-20 influenza vaccination and COVID-19 for other delays between onset of symptoms and swabbing: <10, <5 days, and any delay between onset of symptoms and swabbing. In a sensitivity analysis, we used multiple imputation using chained equations to account for missing data. Briefly, we estimated missing data for 2019-20 influenza vaccination status and covariates using multiple multivariate imputation using chained equations. We used missing at random assumptions. We used all predictors together to impute the missing values and independently analysed 20 copies of the data using 200 cycles of regression. In the primary analysis, we did not take clustering by GP practice into account. Despite large numbers of clusters (GP practices), the number of units within clusters (patients) was low in some study sites (20% of all GP practices included <5 patients). In a sensitivity analyses, we carried out all analyses in the primary analysis using a multilevel model with GP practice as a random intercept, in order to take this clustering into account. To avoid overfitting the logistic regression model, we did not attempt to measure the aOR if there were fewer than 10 cases or controls per number of parameters within the study site variable (N − 1) in the logistic regression model. If there were fewer than 10 cases or controls per number of all parameters, we carried out a sensitivity analysis using Firth's method of penalised regression. Sweden, France and the Netherlands were included in the primary analysis. These countries and Scotland and Navarra were included in secondary analyses. The latter two study sites were included in secondary analyses only, as Scotland had >15% of missing 2019-20 influenza vaccination information and data from Navarra did not include date of onset, so delay between onset of symptoms and swabbing could not be determined. In the primary analysis, 4106 records were received with swab dates between the 1 March 20 202 and the 31 August 2020 (see Figure 1 ). We included in the analysis data set 257 SARS-CoV-2-positive cases and 1631 SARS-CoV-2-negative controls. In the primary analysis, 4106 records were received with swab dates between the 1 March 2020 and the 31 August 2020 (see Figure 1 ). We included in the analysis data set 257 SARS-CoV-2-positive cases and 1631 SARS-CoV-2-negative controls (see Figure 2 ). Among controls, 24% were aged <20 years, compared to 9% of cases. Fifty per cent of controls were aged 40 and over compared to 68% of cases ( Table 1 ). The proportion of controls vaccinated with the 2019-20 influenza vaccination was 25%, compared to 30% of cases. Overall, there was Before restricting to those swabbed within seven days of symptom onset and where delay between onset of symptoms and swabbing was known (3418/3658), the median swab delay was 5 days among cases (IQR: 3-8) and 7 days among controls (IQR: 3-17). among those aged 20-59 and 0.92 (95% CI: 0.51-1.67) among those aged 60 years and older. When modelling the aOR by 5-year age group, the aOR varied between 0.41 and 1.06, but the differences were not statistically significant ( Figure 3 ). The aOR was 0.92 (95% CI: 0.60-1.42) when restricting the days between onset of symptoms and swabbing to 5 or fewer days ( Table 2 ). The aOR was 0.98 (95% CI: 0.71-1.36) when restricting the days between onset of symptoms and swabbing to 10 or fewer days. When permitting any delay between onset of symptoms and swabbing, the aOR was 1.05 (95% CI: 0.78-1.42). The aOR of the imputed analysis among all ages (N = 1888) was 0.90 (95% CI: 0.64-1.27). The imputed analysis point estimates differed by <0.05 to those of the complete case analysis for overall analyses and those stratified by age and sex (Table S1 ). In the primary analysis, 322 GP practices were included with a median number of three patients per GP (IQR: 1-6) and range of 1 to 75 patients per GP practice. The aOR among all ages was 0.77 (95% CI: 0.51-1.15). The aOR for all other estimates were lower by 0.01-0.19 taking clustering by GP practice into account compared to the unclustered analysis (Table S2 ). The secondary analysis included France, Navarra, the Netherlands, Scotland and Sweden, presented for any delay between onset and swab, as onset date was not available in Navarra. The analyses are imputed, due to a high proportion of missing data for SC (45% after applying exclusion criteria). We received 76 146 patients with swab dates between the 1 March 2020 and the 31 August 2020 (see Figure S1 ). We included Confidence intervals in the primary analysis included the null. This was due on one hand to a reasonably low sample size (wide confidence intervals), but also due to the fact that point estimates themselves, so the anticipated biological effects, were not far from The lower aOR among males than among females, which was consistent across all analyses, may have been a spurious finding. The interaction term between sex and 2019-20 influenza vaccination was not significant. Alternatively, we cannot exclude that unmeasured confounding differed by sex, and one of the estimates was (more) biased than the other. As this is an observational study, we cannot exclude in general that there is unmeasured confounding in our study. The inci- In the primary analysis among all ages, the aOR among those swabbed within 5 or 7 days is similar (0.92 and 0.93, respectively), but closer to the null (0.98) among those swabbed within 10 days of symptom onset. Allowing any number of days between onset of symptoms and swab increases the aOR even further to 1.05. With increasing days between onset of symptoms and swabbing, we expect to include more false-negative COVID-19 patients in the analysis. 18 The change in aOR in this direction is expected, if there is a true negative association between 2019-20 influenza vaccination and COVID-19. Based on this, we expect the aOR in the secondary analysis, where onset date is not available from one study site, to underestimate the association. Among all ages, the aOR in the secondary analysis allowing any number of days between onset of symptoms and swab is 0.86. Following the above logic, this may be an While a better understanding of who is swabbed early and late will help with interpretation of analyses, good data quality is key in studies to be able to make inferences. The three study sites in the primary analysis include a low proportion (<10%) of missing data for key variables. These study sites are based on well-established sentinel influenza surveillance networks and are used to high-quality data collection. In the secondary analysis, we include a site (Scotland) that has up to 50% of missing data for 2019-20 influenza vaccination. Our approach of using multiple imputation to account for these missing data is an attempt to deal with missing data. While this statistical approach is useful, having fewer missing data would be preferable. In another study site in the secondary analysis (Navarra), onset date was not available. This study site increased our sample size greatly, but there is some uncertainty around the estimate as we cannot restrict to 7 days between onset of symptoms and swabbing. On the flipside, in this study sites the data quality for key variables such as vaccination status and chronic conditions was extremely high (<1% of missing data). Systems for data collection on patients with respiratory illness have been disrupted due to the COVID-19 pandemic. In this article, we include systems from five different countries, some with data collection limitations. Despite these limitations, the results among the primary analysis sites, with high data quality, were similar to the results among all sites, including those with issues around missing data. Additionally, some of these limitations will be improved in the near future, for example, Scotland has launched a new digital survey data collection platform 1 for patients and clinicians to complete. In Navarra, missing data were due to an exceptional work overload due to the pandemic and have subsequently been resolved. As we move into the winter season, more I-MOVE-COVID-19 study sites will be providing data. As with the sites included in this article, we will assess data quality and potential heterogeneity between sites. Despite all the limitations of data collected during a public health emergency, we believe it is important to provide results that can contribute to the scientific evidence on risk and preven- Navarra: Ethical approval was granted on the 8 May 2020 with approval number PI_2020/45. Document available upon request. Oral consent is required in the Netherlands, France, Sweden and Scotland. Patient informed consent procedures are available upon request. Informed consent is not required in Navarra. No reproduced material is included. The peer review history for this article is available at https://publo ns.com/publo n/10.1111/irv.12839. Data available on request due to privacy/ethical restrictions: The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions. European Centre for Disease Prevention and Control. COVID-19 situation update worldwide, as of 18 Increased Transmission of COVID-19 in the EU/EEA and the UK-thirteenth update GBD 2017 Influenza Collaborators. Mortality, morbidity, and hospitalisations due to influenza lower respiratory tract infections Influenza outbreaks and hospital bed occupancy in Rome (Italy): current management does not accommodate for seasonal variations in demand Impact of influenza vaccination programmes among the elderly population on primary care, Portugal, Spain and the Netherlands: 2015/16 to 2017/18 influenza seasons Influenza vaccination and respiratory virus interference among Department of Defense personnel during the 2017-2018 influenza season Influenza vaccine does not increase the risk of coronavirus or other non-influenza respiratory viruses: retrospective analysis from Canada Positive association between COVID-19 deaths and influenza vaccination rates in elderly people worldwide Covid-19 immunity, covid vaccines, and influenza vaccines Influenza vaccination and risk of SARS-CoV-2 infection in a cohort of health workers Association between exposure to influenza vaccination and COVID-19 diagnosis and outcomes Individualizing risk prediction for positive coronavirus disease 2019 testing Baseline chronic comorbidity and mortality in laboratory-confirmed COVID-19 cases: results from the PRECOVID study in Spain Male gender and kidney illness are associated with an increased risk of severe laboratory-confirmed coronavirus disease The association between influenza vaccination and the risk of SARS-CoV-2 infection, severe illness, and death: a systematic review of the literature on behalf of the I-MOVE study team C. I-MOVE study team. I-MOVE: a European network to measure the effectiveness of influenza vaccines European Centre for Disease Prevention and Control. Timing of influenza vaccination. Timing of influenza vaccination Clinical sensitivity and interpretation of PCR and serological COVID-19 diagnostics for patients presenting to the hospital Regression modeling strategies: with applications to linear models, logistic regression, and survival analysis The effect of influenza vaccination on trained immunity: impact on COVID-19. medRxiv Recent endemic coronavirus infection is associated with less-severe COVID-19 The test-negative design for estimating influenza vaccine effectiveness The effect of small sample size on twolevel model estimates: a review and illustration Robustness issues in multilevel regression analysis Absence of association between 2019-20 influenza vaccination and COVID-19: Results of the European I-MOVE-COVID-19 primary care project