We used a retrospective cohort design involving computerized medical record data from the Royal College of General Practitioners (RCGP) Research and Surveillance Centre (RSC) sentinel network. The RCGP RSC is a nationally representative primary care sentinel network with over 50 years of history in influenza and respiratory disease surveillance [5]. At the time of this study, the RCGP RSC network contained more than 8 million patient records from 336 member general practices across England. Clinical encounters of patients visiting their general practitioners (GPs) for adverse events are recorded into a GP electronic medical record system using 5–byte Read or Clinical Terms Version 3 codes.

We extracted data from all patients who had received a seasonal influenza vaccine between September 1, 2018, and April 30, 2019. We excluded patients older than 100 years, those who attended practices involved in EMA enhanced surveillance programs [6-8], and those who received monovalent pandemic influenza vaccines. Patients were followed up retrospectively for occurrence of a list of AEIs prespecified by the EMA, and those who presented with compatible symptoms were included in this study [9].

We extracted the following data: age, sex, self-reported ethnicity, index of multiple deprivation (IMD), vaccination date, vaccine manufacturer, vaccine valency (number of viral strains included), route of administration, date of any AEI, type of AEI, and dates of registration and deregistration at the practice.

First, we used prescription data (where available) and clinical event data to assign drug name, manufacturer, valency, and route of administration. Second, for the records without prescription data, batch numbers were collated for each drug name to assign manufacturer, valency, and administration route. For records with both prescription and clinical data available, conflicting data were excluded from the analysis. We coded vaccinations where we had sufficient information to identify the vaccination as having taken place within a practice.

We used a self-controlled case series approach [10,11], a method typically used to investigate adverse events following administration of medications or vaccinations. It is a case-only method for investigating the association between a time-varying exposure and an outcome event, in which any time-invariant confounding is automatically controlled for as each patient acts as their own control.

We computed descriptive statistics to provide an overview of the demographic characteristics of the study sample.

We conducted three separate models to address the three aforementioned study aims. The observation period used in all models was from September 1, 2018, to April 30, 2019. Where an individual registered with the practice after September 1, 2018, or deregistered with the practice before April 30, 2019, their observation period was defined as the number of days they remained registered. In model 1, we defined the exposure risk periods relative to the day of vaccination (day 0) as days –7 to –1, days 0 to 6, days 7 to 13, and days 14 to 45, and defined seasonal periods within the observation period as days 0 to 29, days 30 to 59, days 60 to 89, days 90 to 119, days 120 to 149, days 150 to 179, days 180 to 209, and days 210 to 241 from the beginning of the influenza season (see Figure 1 for an illustration). The time outside of the defined exposure risk periods is used as the control for each individual patient. Where an individual received the vaccination very early or very late in the season, parts of the exposure risk periods that fell outside of our defined observation period were not included. We calculated the RI of AEIs following vaccination for the different exposure periods and for the different seasonal periods. In models 2 and 3, we focused on the first 7 days post vaccination as the exposure risk period and modeled potential modification effects of vaccine type and vaccination settings by including an interaction term in the model [12].

All statistical analysis was performed using R 3.4.4 (R Foundation for Statistical Computing) [13] with the packages tidyverse version 1.2.1 [14], SCCS version 1.1 [15], lubridate [16], and tableone [17]. Graphical output was generated using the packages ggplot2 [18] and ggthemes version 3.5.0 [19].

All potentially identifiable data were pseudonymized as close to the source as possible and not made available to researchers; data were not extracted for patients who opted out of data sharing. All data are stored and processed at the RCGP RSC secure data and analytics hub, the University of Surrey. According to the Health Research Authority and Medical Research Council Regulatory Support Centre’s online decision tool, this study falls under the category of service evaluation and does not require further ethical review. This study was approved by RCGP.

A total of 1,024,160 unique individuals who received seasonal influenza vaccinations were identified, of which 165,723 individuals presented with symptoms compatible with vaccine-related AEIs. Baseline demographic characteristics of the study participants are presented in Table 1. The median age of the cohort was 66 years, with slightly more women than men, and the majority of the study participants were of White ethnicity. The age-sex profile showed a peak between the ages of 2 and 9 years, and a marked increase in uptake from the age of 65 years (Figure 2). The IMD distribution, however, showed a slight overrepresentation of patients from less deprived neighborhoods (see Table 1).

We were able to identify the vaccine administered in 77.8% (797,285/1,024,160) of the records. The main types of vaccines used in 2018/2019 included aTIV, QIV, and LAIV. A small number of trivalent vaccines (TIVs) were also identified (n=1526); as they were not recommended in the national influenza vaccination program, we excluded them from the analyses.

The incidence rates of AEIs for which patients sought consultation in the 7 days post vaccination are listed in Table 2, grouped by category of surveillance condition. We observed AEIs in every EMA category, from the most common ones of cough, myalgia, rash, and headache to the more severe ones such as Guillain-Barré syndrome (n=4) and anaphylaxis (n=6).

We observed a slight reduction in the incidence of AEIs in the 7 days prior to vaccination (RI 0.91, 95% CI 0.89-0.94). The 7 days following vaccination showed an elevated incidence (RI 1.88, 95% CI 1.84-1.91). The period of day 8 to 14 was not associated with a significant increase in incidence, and the period of day 15 to 45 only showed a marginally increased incidence (RI 1.01, 95% CI 1.00-1.03; Table 3).

We also observed a seasonal pattern of AEIs, with increasing incidence as the influenza season progresses, reaching a peak around February, and then the incidence declines until the end of the season. The exception to this pattern is the 30-day period that encompasses the end-of-year holidays, which showed a lower incidence compared to the preceding and succeeding 30-day periods (Table 3).

To compare the incidence of AEIs associated with the different types of vaccines, we incorporated an interaction term in the model. The results showed that relative to aTIV, QIV was associated with more AEIs (RI 1.46, 95% CI 1.41-1.52), whereas LAIV was associated with fewer AEIs (RI 0.78, 95% CI 0.73-0.83; Table 4).

Similarly, we added an interaction term to the model to explore whether vaccination setting had an impact on rates of AEIs. There appeared to be no significant difference in rates of AEIs whether or not the vaccination took place within a practice (RI 0.97, 95% CI 0.66-1.43; Table 5).

In this study, we examined the incidence of AEIs following seasonal influenza vaccination, and the moderation effects of vaccination type and setting. We demonstrated that AEIs most occurred within 7 days post vaccination, although a small increase in incidence was also detected between days 15 and 45. Similar to observations from earlier influenza seasons, we also found a small “healthy vaccinee” effect in the 7 days leading up to vaccination, possibly due to patients who felt unwell deferring vaccination [20]. The seasonal pattern we observed was expected, as many of the AEIs are more common in winter.

Vaccination type had a significant effect on the incidence of AEIs in the 7 days post vaccination; we found that LAIV was associated with the lowest incidence, followed by aTIV and QIV. The incidence of AEIs following immunization with LAIV was 22% lower than that with aTIV, which in turn had a 46% lower incidence of AEIs than QIV. Data from earlier influenza seasons also showed a similar pattern where LAIV was associated with lower incidence of AEIs compared to QIV [20]. The finding that aTIV was associated with lower rates of AEIs than QIV was unexpected, as previous clinical trials and postlicensure studies have either reported higher reactogenicity with aTIV compared to TIV or similar safety profiles between the two [21-23].

Vaccination setting did not influence the incidence of AEIs; receiving the influenza vaccine outside of a practice was not associated with higher rates of AEIs. In the United Kingdom, most of these vaccinations not administered in a practice would have taken place in a pharmacy. In recent years, several countries have introduced pharmacy-based influenza vaccination services to enhance vaccination coverage. Their convenience and accessibility are thought to address some of the factors that contribute to vaccine hesitancy. Previous studies have reported that patient experiences are generally positive [24] and that vaccine delivery was safe [25]. Our findings are in accordance with prior research that influenza vaccines administered outside of a practice are safe, and leveraging this to enhance vaccination coverage may be particularly important in a pandemic, and commissioning other providers in addition to pharmacists may be an option to consider. The only downside of pharmacist vaccination is that there is no electronic transmission of administration into GP computerized medical record systems [26].

The RCGP RSC is a nationally representative sentinel network, and practices receive feedback about data quality, so only data that meet a quality threshold are included in our report. However, it is important to acknowledge that the data used in this study is based on GP consultations, and patients are unlikely to seek medical attention for commonly expected or minor AEIs but tend to self-manage instead, or they may have directly attended the hospital for more severe AEIs. Furthermore, we were unable to identify brand-specific information for a proportion of the vaccinations; most often this is where vaccines are not administered in general practice [26]. Given that different types of vaccines were offered to different age groups in the 2018/2019 season, it is possible that the differences in rates of AEIs that we observed reflect differences in reactogenicity.

To date, most vaccines have had egg-based manufacturing, but in the 2019/2020 season cell-based manufactured vaccines were introduced [27]. A similar comparison of AEIs should be conducted to compare cell-based with egg-based QIV. Additionally, the sentinel system has had to adapt for the COVID-19 pandemic [28,29]; this has resulted in the expansion of the network size and the strengthening of its infrastructure [30]. The United Kingdom has ordered over 90 million COVID-19 vaccine doses [31], and the sentinel system could be used to monitor AEIs associated with its administration.

The incidence of AEIs varied between different types of vaccines and can be compared using routine sentinel network data. Here, we report that aTIV are associated with fewer AEIs than QIV, but more AEIs than LAIV. Rates of AEIs were similar whether a vaccine took place in or outside of a practice. These findings should be reassuring for patients, address some of the factors that contribute to vaccine hesitancy, and may help improve vaccination coverage in future influenza seasons.