Quantifying pupil-to-pupil SARS-CoV-2 transmission and the impact of lateral flow testing in English secondary schools

Abstract

A range of measures have been implemented to control within-school SARS-CoV-2 transmission in England, including the self-isolation of close contacts and twice weekly mass testing of secondary school pupils using lateral flow device tests (LFTs). Despite reducing transmission, isolating close contacts can lead to high levels of absences, negatively impacting pupils. To quantify pupil-to-pupil SARS-CoV-2 transmission and the impact of implemented control measures, we fit a stochastic individual-based model of secondary school infection to both swab testing data and secondary school absences data from England, and then simulate outbreaks from 31st August 2020 until 23rd May 2021. We find that the pupil-to-pupil reproduction number, R_school, has remained below 1 on average across the study period, and that twice weekly mass testing using LFTs has helped to control pupil-to-pupil transmission. We also explore the potential benefits of alternative containment strategies, finding that a strategy of repeat testing of close contacts rather than isolation, alongside mass testing, substantially reduces absences with only a marginal increase in pupil-to-pupil transmission.

Fig. 1. Fitting the model to testing and secondary school absences data.

The stochastic individual-based model is fitted to (a) the percentage of 11–16 year olds who test PCR positive (excluding confirmatory PCR tests) each day from 1st September 2020 to 23rd May 2021, (b) the percentage of 11–16 year olds who test LFT positive each day from 8th March 2021 to 23rd May 2021. Circles correspond to the data, with shaded intervals around mean model traces (solid lines) representing 95% prediction intervals in all plots. Shaded vertical grey regions represent time periods when schools were not fully open (either due to closures or school holidays). The model is also fitted to (c) the distribution of peak number of confirmed COVID-19 absences in secondary schools from 1st September 2020 to 18th December 2020, and (d) the distribution of peak number of confirmed COVID-19 absences in secondary schools from 8th March 2021 to 23rd May 2021. Circles denote the data and shaded blocks the 95% prediction interval estimated from the model. The plots above show the mean values obtained from 100 simulations in 2979 secondary schools, each with a distinct parameter set sampled from the posterior distribution.

Fig. 2. Incidence and R_school from the fitted model.

We display time-series of (a) total incidence among pupils (blue) alongside incidence occurring through external (non-school) infections (red) (b) R_school through time (thin line) alongside its seven-day moving average (thick line). Results obtained from 100 simulations in 2,979 secondary schools, each with a distinct parameter set sampled from the posterior distribution. In all panels, solid lines correspond to mean temporal profiles, shaded ribbons represent 95% prediction intervals and the shaded vertical grey regions represent time periods when schools were not fully reopen (either due to closures or school holidays).

Fig. 3. Quantifying the impact of LFTs on transmission and absences and the potential impact of alternative strategies.

Time-series under different intervention strategies of (a) incidence among pupils, (b) R_school within secondary schools, and (c) the percentage of pupils absent; and (d) the average R_school from 19th April 2021 to 9th May 2021 realised by different levels of participation in mass testing. We compare a policy of twice weekly mass testing and isolating close contacts (purple) to a strategy of isolating close contacts only (blue), twice weekly mass testing only (red), and twice weekly mass testing alongside serial contact testing (green). Results obtained from 100 simulations of 2979 secondary schools, each with a distinct parameter set sample from the posterior distribution. In all panels, solid lines correspond to the mean estimate, shaded intervals represent 95% prediction intervals and the shaded vertical grey regions represent time periods when schools were not fully reopen (either due to closures or school holidays). The data in panel (c) consists of the number of absences due to a confirmed case or a suspected case of COVID-19, and absences arising as a result of students told to isolate due to potential contact with a case of COVID-19 from inside their educational setting.

Fig. 4. Overview of the individual-based model components.

a A schematic of the modelled within-school mixing structure. Within a school, pupils interact with close contacts in their year at baseline rate α₀ = 1, with other pupils within their year at a relative rate α₁, and interact with other pupils in other years at a relative rate α₂, where 0 ≤ α₁, α₂ ≤ 1. b England divided by LTLA, with an example LTLA highlighted in purple. Each school is situated within an LTLA, which determines its probability of infection from the community, its relative frequency of the B.1.1.7 (Alpha) variant, and LFT uptake. Each LTLA contains multiple secondary schools, shown as blue dots (the number of blue dots shown is illustrative rather than an accurate depiction of the number of secondary schools in the highlighted LTLA). c A time-series of the percentage of that LTLA’s population testing positive on a PCR test on that day. A pupil’s probability of external infection on day t depends upon prevalence in the community, which we assume to be proportional to the proportion of the population in that LTLA testing PCR positive on day t + 5. d A time-series of the fitted estimate of the relative frequency of the B.1.1.7 variant in the example LTLA. The expected number of secondary infections from infected pupils depends upon the proportion of cases that are of the (more transmissible) B.1.1.7 variant, which varies through time and is dependent on the LTLA the school is situated within. Cross markers indicate the percentage of PCR tests from an LTLA that return an S-gene negative result out of those that return an S-gene status. Our model does not consider the impact of the B.1.617.2 (Delta) variant, which became the dominant variant in circulation during late May 2021, i.e. occurring beyond the time horizon of our analyses.