A self-controlled case series study to measure the risk of SARS-CoV-2 infection associated with attendance at sporting and cultural events: the UK Events Research Programme events

Phase 3 ERP event organisers provided data, where available, on attendees, to the UKHSA. The study population was then formed from the subset of these attendees, who also had a result for any SARS-CoV-2 test recorded in the National Health Service (NHS) COVID-19 Test and Trace system in days 1–18 following event attendance. One of the remits of the Test and Trace system was to record SARS-CoV-2 test results in England, whether positive or negative. Data were available on attendee age, sex, the event attended, date of the event, and self-reported COVID-19 vaccine status. The proportion of attendees for whom attendance data were available varied between events for two reasons. Some events asked attendees to actively opt in to making data available, in which case-only data for those providing consent were made available. For other events, availability was determined by how organisers arranged and recorded bookings.

The exposure was attendance at a phase 3 ERP event, as recorded by the event organisers. Events for which data were available are listed in Table 1. Events varied in nature substantially and have been broadly grouped according to predominant features: indoor seated, outdoor seated, outdoor partially structured, and outdoor unstructured. Categorisation was for convenience, and we recognise that, in reality, all events have a more complex nature. For multi-day events, where recorded, first and last possible attendance dates were obtained. Where exact attendance dates for multi-day events were unavailable (i.e. Latitude and Tramlines festivals), attendance on all dates was assumed.

The main outcome of interest was a positive COVID-19 test, as recorded in the NHS Test and Trace Second Generation Surveillance System. At the time these events occurred, national guidance on testing recommended that individuals who had new onset of the cardinal COIVD-19 symptoms test by polymerase chain reaction (PCR) as soon after onset as possible. Whilst awaiting a result, they were required to self-isolate. The general public over the age of 12 years were advised to undertake twice weekly lateral flow device (LFD) tests, though the uptake was low [4]. Additionally, there were other national testing programmes, e.g. PCR testing of individuals prior to elective care, and more intensive asymptomatic testing regimes of care home staff. Results from private providers were not included in SGSS data. Positive results, whether obtained via a lateral flow device (LFD) or a PCR test, were used for the primary analysis. Any recorded negative result via either LFD or PCR was considered a control outcome.

Covariates of interest were attendee age (in 5-year bands), sex, self-reported COVID-19 vaccine status, and weekly estimated regional prevalence of COVID-19 using figures local to the attendee, and irrespective of age, available from the Office for National Statistics.

The nature of such a large-scale events programme meant conventional study designs, such as a cohort approach, would be challenging, both in collecting comprehensive data on all participants, and in identifying an appropriate control group. We therefore used a self-controlled case series (SCCS) design, in which each participant acts as their own control [5]. Inclusion in a SCCS is conditional on having the outcome of interest during a predefined observation period. The observation period is divided into high and baseline risk periods, and an incident rate ratio is calculated comparing the rate of the outcome during these two periods. The method relies on two key assumptions. Firstly, the occurrence of the outcome should not affect the likelihood of subsequent exposure. In this instance, a positive COVID-19 test would preclude attendance at an event, and so we applied an extension of SCCS methodology where observation begins at the point of exposure [6]. This extension is valid for exposures that are unique within the observation time of interest, which is reasonable during the 18-day period after each event (i.e. that unique event could not be attended again). The second assumption is that the outcome of interest should not substantially curtail observation. This assumption tends to be invalid for outcomes that are frequently fatal. COVID-19 in this context is only fatal in rare instances, though in any case, the extension we applied also negates this assumption.

The key advantage of an SCCS design is that each participant acts as their own control, which means that characteristics that do not vary over the observation period cannot confound any associations. It is also a statistically efficient study design, not requiring a high absolute risk of the outcome to be able to detect an increased risk. Risk factors for the outcome that do vary over the observation period can be adjusted for in the analysis. Figure 1 outlines the key features of the SCCS design we used.

Using conditional Poisson regression, a within-person incidence rate ratio was calculated comparing the rate of positive testing for SARS-CoV-2 during high risk and baseline periods of time. The high risk period was days 3 to 9 following event attendance, and the baseline period was days 1 and 2 and 10–16 following the event when infection detection was assumed to be unaffected by attendance. For events where data indicated multi-day attendance (Goodwood, Tramlines, Latitude, Silverstone, Open Golf), the first day of the high risk period was defined as either 3 days after the first attendance date or as the final attendance date if this was > 3 days after the first attendance date. Rate ratios were adjusted for estimates of underlying regional SARS-CoV-2 prevalence to account for population level fluctuations in infection risk. Prevalence was categorised as < 200, 200–< 300, 300–< 400, 400–< 500, 500–< 600, > 600 per 10,000, with data shown in Additional File 1 Table S1. Each day of the observation period was then categorised according to the prevalence in the region of the attendee on that day, and prevalence was included in the regression model as a categorical variable. A rate ratio of 1 suggests that SARS-CoV-2 infection risk was unaffected by attendance at the event, whilst a rate ratio and 95% confidence interval all greater than 1 suggests transmission risk may have been increased by attendance.

We anticipated that the probability of attendees testing themselves for SARS-CoV-2 would be highest immediately after the event and would likely fall over the observation period. This would mean that positive cases were more likely to be detected in the high risk period than in the baseline. To account for this, we also calculated the rate ratio for negative test results, using the same risk and baseline periods as in the primary analysis. We used this as a proxy for testing behaviour and subsequently calculated a ratio of incidence rate ratios (IRR) (IRR for positive tests/IRR for negative tests), to account for any declining testing probability over the study period.

The following sensitivity analyses were conducted: (1) the timing of high risk periods was extended to days 3–11 after the event, with days 1 and 2 and 12–18 acting as the baseline; (2) the definition of a positive test was restricted to first PCR positive test; and (3) at the request of referees, we stratified by individual outdoor unstructured event (Goodwood, Latitude and Tramlines).

Other UK ERP studies investigated both air quality and behavioural risk factors at events [8, 9]. Air quality was measured through carbon dioxide levels, with high levels indicating poor ventilation. The majority of venues had good air quality throughout, but peaks were noted in areas of high congestion, allowing organisers to focus on areas where ventilation improvements were needed. Behavioural studies found wide variation in the ability of attendees to physically distance at an event and in face covering usage. These were largely driven by venue size, event capacity, and whether mitigations such as wearing a face covering were mandatory rather than discretionary.

Over the last 12 months, others have evaluated transmission risks associated with a variety of events, using a range of methodologies. Their findings largely coincide with our results.

Indoor live concerts were studied through randomised trials in both Barcelona and Paris, with all those undergoing randomisation required to test negative shortly before the event. Neither found any increased risk of infection associated with attendance [10, 11]. However, pre-event testing and mask wearing were mandatory, and the total number of cases detected was very small. By contrast, a non-randomised cohort analysis found that attendees of festivals in Catalonia had a twofold increased risk of infection compared with non-attendees [12]. Shukla et al. [13] also found a strong corelation between attendance at the Kumbh Mela and local infection rates. A non-randomised study in March 2021 amongst 4584 indoor full-capacity concert attendees in Barcelona found that after pre-concert screening, only six attendees were diagnosed with COVID-19 within 2 weeks of attending the concert [14].

Smith et al. studied the UK ERP using data from contact tracing to see how ERP event attendance and subsequent infection risk varied [15]. Some events were found to have higher rates of positive case attendance and onwards transmission; Euros attendees were more likely to be infected on entry and to be associated with onwards transmission than attendees at other events such as Wimbledon. Sociocultural and mitigating factors were highlighted as being potentially very different between events.

These findings are largely reassuring as there was little evidence of an increased risk of SARS-CoV-2 transmission after attendance at most event types, in the context of a carefully managed event programme where pre-attendance negative testing was required and where venues had taken some mitigating steps to reduce the likelihood of transmission. Attendance at the three festivals we studied was associated with a 70% increased risk of transmission. The majority of infected festival attendees were under 50 years and unvaccinated at the time of the event. It is unlikely that the additional cases acquired at these events would have led to substantial morbidity for those directly infected, and they would be unlikely to have a large impact on wider community incidence compared with the number of cases occurring outside the festival setting. This study also highlights a novel use of the self-controlled case series design which could be used in future for monitoring the impact of single event attendance on infection risk, requiring only data on cases who attended the event, along with information about general testing trends amongst all attendees. Whilst the events we studied were all cultural, the method would be equally effective at a broader range of events, e.g. international conferences or summits. What is essential is the collection of sufficient data to permit the relevant analysis, namely attendance at the event and subsequent testing results amongst all attendees, regardless of result. We recommend this be factored into privacy notices for event attendance in the context of any future important epidemic situation where evidence of risk from event attendance is needed.