In this study, we compare the PCSC with national data for population representativeness and vaccine uptake by formulation and provide a descriptive report by week of vaccine uptake. These elements were included because vaccine studies can either be analyzed at the national population level or run in representative samples. It is important to know if there is sufficient representation of the different influenza vaccine formulations and to understand the timings of vaccine administration for studies of adverse events of interest or those of vaccine effectiveness compared with known dates of cocirculating strains of influenza [11].

This study was conducted during the 2019-2020 influenza season. The 2020-2021 season was not used as it was considered by authors as atypical due to the COVID-19 pandemic and associated changes in vaccination schedules (inclusion of persons aged 50 to 64 years and an absence of circulating influenza as a result of stay-at-home mandates) [12]. This study was conducted using the PCSC of the Oxford Royal College of General Practitioners (RCGP) Research and Surveillance Centre (RSC). At the time of this study, the RSC was, and remains, one of the largest and longest-established primary care sentinel networks in Europe [13,14].

At the time of this study, the RSC had been recording respiratory conditions, including influenza and notifiable diseases, and had provided data to support clinical and public health research (most notably on vaccine effectiveness) for over 50 years [3,13-16]. Data from the RSC were, and still are, hosted within the Oxford RCGP Clinical Informatics Digital Hub, a trusted research environment [17]. The RSC as a whole comprises over 1800 general practices in England, and the total current registered list includes over 19 million people—approximately 32% of the national population (Figure 1).

The RSC adapted and grew over the course of the COVID-19 pandemic when it was divided into 2 divisions [18]. First, the PCSC is the long-standing sentinel group of general practices (n=693) that conduct virology sampling (n=240) and serosurveillance (n=273) [19,20]. This group of practices has a long history of receiving feedback about data quality, historically at an annual visit and, in recent years, through dashboards, webinars, and web-based visits [21,22]. They were recruited to be nationally representative [13]. The second group, the Syndromic Surveillance General Practices (SSGPs), was recruited to support the UK Health Security Agency’s (UKHSA; formerly Public Health England) syndromic surveillance [23]. These practices were recruited to complement other data sources and were not nationally representative.

The study population included participants of all ages in England who were registered with a general practitioner (GP) in the PCSC between September 1, 2019, and January 29, 2020. We describe the demographic characteristics of the study population in terms of the following:

To evaluate the representativeness of the PCSC in terms of the demographic characteristics above, we undertook a quantitative comparison of the age, sex, and IMD profile of the PCSC against published data for 2019-2020 from the Office for National Statistics (ONS) for England [29]. We compared the ethnic profile of the PCSC against ONS ethnic population estimates extrapolated from the 2011 census [30]. We also compared the broad geographic distribution of the PCSC population in terms of NHS regions with the data from the ONS on the geographical distribution of the population by government office regions [27,31]. Last, we compared the distribution of people with risk factors for influenza defined by the Chief Medical Officer in the PCSC with data from the Quality and Outcomes Framework 2019-2020 primary care disease prevalence figures published by NHS Digital [28,32]. We made 1 near-match using the Quality and Outcomes Framework prevalence of chronic obstructive pulmonary disease to match with chronic lung disease in our data. The latter has a broader definition including conditions such as bronchiectasis.

We took extensive steps to avoid denominator inflation within our data processing. Denominator inflation occurs when patients fail to be deregistered. The existence of a unique national identifier (NHS number) helps address this problem [33], as does the national demographic service, which facilitates the linkage of primary care and secondary care data as well as death certificate data. These include removing cases where there is a completely empty medical record and, based on pseudonymized NHS numbers, only keeping the most recent record where duplicates are identified.

The exposure of interest was influenza vaccination. Influenza vaccination was either recorded as a clinical term for vaccine administration or as a prescription. A large number of doses of influenza vaccines are administered by pharmacies and other community health care workers, although the majority of these vaccines are recorded in the primary care computerized medical record (CMR). Where vaccination was recorded as a clinical term in the CMR, it may not have been specific to the formulation, brand, or batch; when vaccination was prescribed, it was brand and batch specific. Prescriptions were only issued for adult and high-risk children who were vaccinated in their general practice [34]. A list of Systematized Nomenclature of Medicine clinical terms used to identify influenza vaccinations from the CMR of patients in the PCSC is listed in Table S1 in Multimedia Appendix 1 [27,35].

Vaccine coverage in the study population was described by providing the count and proportion (expressed as a percentage) of the PCSC vaccinated against influenza overall and by the following:

We could not precisely match the geographic locations. Our nearest match was to add East and West Midlands government regions to be equivalent to our Midlands region; we also combined Yorkshire and the Humber and the North East to be equivalent to our North East and Yorkshire regions.

To evaluate the representativeness of influenza vaccine coverage in the PCSC, a quantitative comparison was made to the influenza vaccine coverage for the winter season—September 1, 2019, to January 29, 2020—reported by the UKHSA [35]. We compared the influenza coverage in the PCSC according to the following inclusive age groups reported in national data by the UKHSA: 6 months to 1 year, 2-4 years, 5-15 years, 16-64 years, and ≥65 years. For influenza coverage by type of vaccine, coverage in the PCSC was compared with national data from the UKHSA in the following age groups only: 16-64 years and ≥65 years. We again made some near-matches for at-risk groups: coronary heart disease is compared with chronic heart disease, liver disease with chronic liver disease, neurological disease with chronic neurological disease, and obesity with morbid obesity.

In order to evaluate the timing of influenza vaccinations given in the PCSC, we calculated the vaccine coverage overall in the study population by week and day of the week and recorded this information in line with its relevant International Standard Organization (ISO) week number.

For categorical measures, we calculated the frequency and percentage of total study participants observed in each category. For continuous and count variables, we presented the mean, SD, median, and range.

For population proportions, we calculated 95% CI using the normal approximation to the binomial calculation [36]. Paired 2-tailed t tests were used to compare populations [37].

Statistical analyses were undertaken using R (version 3.5.1; R Core Team) [38].

This study was classified as a service evaluation (measuring what standard of care this service achieved) by the Medical Research Council/Health Research Authority decision tool [39], so it did not require formal ethical approval. It was reviewed by the RCGP Approval Committee on January 8, 2021.

A total of 7,010,627 people (male: n=3,488,789, 49.8%; female: n=3,521,838, 50.2%) were registered with a primary care practice from the PCSC that submitted data for this study (n=693) between September 1, 2019, and January 29, 2020. The study population had a greater proportion of people aged 18-49 years (2,982,390/7,010,627, 42.5%; 95% CI 42.5%-42.6%) compared with the 2019 midyear population estimates from ONS (23,219,730/56,286,961, 41.3%; 95% CI 4.12%-41.3%; P<.001; Table 1) [29]. Populations in the most deprived quintile, namely those in IMD quintile 1, were underrepresented in this cohort, and those in the least deprived quintile, namely those in IMD quintile 5, were overrepresented.

A total of 1,731,062 (24.7%; 95% CI 24.7%-24.7%) out of 7,010,627 people within the PCSC received an influenza vaccine (male: 790,660/1,731,062, 22.7%; female: 940,402/1,731,062. 26.7%). We subdivided vaccine coverage in the PCSC population by sociodemographic characteristics (Table 2 and Table S2 in Multimedia Appendix 1) and calculated the proportion of the PCSC population who were vaccinated in each subpopulation stratum. For example, 72.3% (913,695/1,264,700; 95% CI 72.2%-72.3%) of people aged 65 years or older in the PCSC were vaccinated, whereas the national data showed that 72.4% (7,621,505/10,523,854) of individuals aged 65 years or older received seasonal influenza vaccine [35], with a coverage of 24.2% (14,468,665/59,764,928; 95% CI 24.2%-24.2%; P<.001). National coverage data only provided information on limited age bands, making direct comparisons between cohorts challenging (see Table S2 in Multimedia Appendix 1). Coverage of risk groups was slightly higher in our study than in comparable national data; for example, 69.8% (284,280/407,228; 95% CI 69.7%-70%) of people with diabetes in our study received an influenza vaccine whereas 61.2% (983,727/1,607,996; 95% CI 61.1%-61.3%; P<.001) of people with diabetes received an influenza vaccine according to national data (Table 3).

In the PCSC, vaccine type and brand information were available for 71.8% (358,365/498,923; 95% CI 71.7%-72%) of influenza vaccines administered to people aged 16 to 64 years (Table 4 and Table S3 in Multimedia Appendix 1) and 81.9% (748,312/913,695; 95% CI 81.8%-82%) of those administered to people aged 65 years or older (Table 5 and Table S3 in Multimedia Appendix 1). This compares with national data where only 23.6% (696,880/2,900,000) of seasonal influenza vaccinations for people aged 16 to 64 years (Table 4) and 17.8% (1,385,888/7,700,000) for those aged 65 years or older were coded with a defined vaccine type in 2019-2020 (Table 5) [35].

During the 2019-2020 influenza season in the PCSC, influenza vaccination started in ISO week 35 and continued until ISO week 3. Peak of vaccinations occurred in ISO week 41 (Figures 2 and 3). Weekly peaks in vaccination coverage occurred on Saturdays, especially among those aged 65 years or older. People aged 65 years or older received their influenza vaccinations earlier in the season than other age groups, with a peak around ISO week 40.

The population of patients in the English national PCSC is broadly comparable to the general population in terms of demographic features. However, there is a greater proportion of people aged 18-49 years in the PCSC, and people who are more deprived are underrepresented and those in the least deprived quintile overrepresented. It is possible this may result in denominator inflation in the younger half of the working-age population for any influenza vaccine coverage estimates.

Overall, influenza vaccine coverage is comparable in the PCSC to national data, although coverage of risk groups was higher in the PCSC than in comparable published national data. Influenza vaccine type was recorded for a higher proportion of vaccines recorded in the PCSC than available from national data, although there are still a substantial number of vaccinations in the PCSC for which no influenza vaccine type or brand information was recorded.

The timing of influenza vaccination during the 2019-2020 season showed that influenza vaccination started in ISO week 35, with people aged 65 years or older receiving their vaccinations earlier in the season than other age groups. Weekly peaks in vaccination coverage were seen on Saturdays, especially among those aged 65 years or older.

Our study demonstrated the representativeness of the PCSC in describing vaccine exposure, including timing, vaccine type, and brand-specific exposure. However, although we took steps to avoid denominator inflation, it is possible that our findings may be affected by an inflated denominator in the younger half of the working-age population (age 18 to 49 years). Denominator inflation may be a result of systematic differences in PCSC practices versus those practices outside the PCSC that have not been possible to quantify in this study, including opening times, accessibility, range of services offered, and patient satisfaction.

PCSC data about individual vaccine exposure, if combined with data on infection status, could be used to estimate vaccine effectiveness. The timing of vaccination in relation to the start of the season would allow comparison of individual vaccine exposure with circulating influenza virus strains. Where a larger population is required for a study, we could also include the RSC’s SSGPs. As the aggregated PCSC and SSGP population is over 19 million (32% of the national population; Figure 1), their combined representativeness would support a national study for which a larger sample size was needed. The RSC could also link individual exposure data to clinical records or conduct enhanced surveillance to monitor adverse events of interest.

The PCSC data did not capture the influenza vaccine type and brand for all vaccinations. Steps should be taken to improve data quality to enable comparative influenza vaccine effectiveness studies to be conducted. Of note for researchers looking to collect further information at the time of vaccination, many general practices run their influenza vaccination clinics on Saturdays, particularly for older people.

Strengths of the study included its large sample size from a well-established sentinel network that includes primary care practitioners who have been recording information on influenza incidence and vaccinations for over 50 years [14]. Additionally, given the completeness of our data on influenza vaccine type, we were able to include a wider range of influenza vaccine types in the PCSC than comparable national data [35]. However, the limited amount of comparable national data also curtailed our quantitative comparison of influenza vaccine coverage. It was especially difficult to compare the proportions of those vaccinated within at-risk groups as there was no detailed information available in national data. We have presented the proportion of people in at-risk groups vaccinated against influenza as a proportion of all the people in the at-risk group who were registered that year. For example, 284,280 people with diabetes received an influenza vaccine versus 407,228 people with diabetes. Where we identified a higher proportion of risk groups, we think this likely indicated better data quality; however, this may be a statistical artifact of differences in disease prevalence. Finally, we could not precisely match some risk groups.

The COVID-19 pandemic has resulted in interruptions of health services since early 2020. The indirect effects this had on routine vaccinations in primary care may have also affected the results of this study. However, the influenza vaccination programs had largely been completed by early 2020, and we felt confident in including 2019-2020 in our analysis. That said, this limited us to including only 1 year of data in our analysis. Furthermore, a large number of doses of influenza vaccines are administered by pharmacies and other community health care workers [40]. Although the majority of these vaccines are recorded in the CMR, communications from pharmacists to GPs around vaccination records have been described as suboptimal, especially with regard to data on manufacturer or batch number [34]. This may have affected the completeness and accuracy of our data, specifically, incomplete data about influenza vaccinations provided by pharmacies and other community health care workers would serve to underestimate the vaccine coverage rate we present in this paper.

Other authors have reported strengths and limitations in the personal demographics service (described in detail in our method) and NHS number as a unique identifier [41]. Although a national identifier is overall a great strength, we need to conduct further studies to explore whether there is any denominator inflation in the PCSC database.

Past work codifying COVID-19 vaccination records in primary care settings has shown a high level of completeness of vaccine brand recording for this vaccination program. The National Immunization Management Service has recorded 99.6% of vaccine administration details and electronically transmits brand information to every individual’s GP record on a daily basis [42]. This compares with between 43.8% and 81.9% of influenza vaccine records having vaccine type and brand information accessible in the PCSC, depending on age.

Influenza vaccine effectiveness assessment is increasingly stratified by vaccine type or even vaccine brand to enable comparisons of different formulations that help to inform public health decision-making [43,44].

The United Kingdom is a possible candidate country to source comparative data for the newer influenza vaccine formulations, as it has historically had high annual influenza vaccine coverage for those aged 65 years or older. Between the 2004-2005 and 2019-2020 seasons, average coverage ranged from 71% to 75%; these figures are relatively high compared with the rest of Europe [45,46]. The World Health Organization has set a vaccination coverage target of at least 75% in the older adult population and among risk groups [47]. Influenza vaccine coverage among at-risk groups <65 years of age, excluding pregnant women, in England has been lower—approximately 49% between the 2007-2008 and 2016-2017 seasons [48,49].

The English PCSC was broadly representative of the national population. It included high-quality routine data, in terms of sociodemographic risk groups, recording of risk groups, and influenza uptake compared with national data sources. However, although data quality was reported to be good, there were significant gaps in vaccine type and brand. Methods implemented to capture COVID-19 vaccine data and encode it within general practice CMRs should now be implemented for influenza vaccines. The PCSC of the RSC, and the UK health care ecosystem itself, provide favorable environments for conducting influenza vaccine benefit-risk studies.