Dynamics of a national Omicron SARS-CoV-2 epidemic during January 2022 in England

Rapid transmission of the SARS-CoV-2 Omicron variant has led to record-breaking case incidence rates around the world. Since May 2020, the REal-time Assessment of Community Transmission-1 (REACT-1) study tracked the spread of SARS-CoV-2 infection in England through RT-PCR of self-administered throat and nose swabs from randomly-selected participants aged 5 years and over. In January 2022, we found an overall weighted prevalence of 4.41% (n = 102,174), three-fold higher than in November to December 2021; we sequenced 2,374 (99.2%) Omicron infections (19 BA.2), and only 19 (0.79%) Delta, with a growth rate advantage for BA.2 compared to BA.1 or BA.1.1. Prevalence was decreasing overall (reproduction number R = 0.95, 95% credible interval [CrI], 0.93, 0.97), but increasing in children aged 5 to 17 years (R = 1.13, 95% CrI, 1.09, 1.18). In England during January 2022, we observed unprecedented levels of SARS-CoV-2 infection, especially among children, driven by almost complete replacement of Delta by Omicron.

healthcare systems struggled to cope with the associated increased healthcare demands 11,12 .
The REal-time Assessment of Community Transmission-1 (REACT-1) study tracked the spread of the SARS-CoV-2 virus among randomly-selected community samples in England, approximately monthly since May 2020, avoiding the biases associated with case incidence data and the delays inherent in hospitalisations and deaths 13 . We document the transmission dynamics of SARS-CoV-2 in England during January 2022 (round 17) following the emergence of Omicron as the dominant variant during December 2021 5,7 .
Ct values of swab-positive adults aged 18 to 54 years having received a third vaccine dose (compared to those having received two vaccine doses) were higher (lower viral load) in round 16 for N and E gene (when Delta predominated), and for E gene in round 15 (October 19 to November 5, 2021, all Delta), but not in round 17 (predominantly Omicron), suggesting less protection from infection by the booster dose for Omicron than for Delta. Irrespective of the round, we found lower Ct values (higher viral load) for N gene in swab-positives reporting symptoms compared to those not reporting any symptoms (p < 0.01) (Fig. 2B, C).

Temporal trends
We show a substantial increase in weighted prevalence between round 16 (November 23 to December 14, 2021) and round 17 (January 5 to 20, 2022) (Supplementary Table S1, Fig. 3A At all ages, weighted prevalence increased from two-(in those aged 5 to 11 years) to almost twelve-fold (in those aged 75 years and over) between round 16 and round 17 (Supplementary Table S3A, Supplementary Fig. S2). The highest weighted prevalence in round 17 was observed in those aged 5 to 11 years at 7.85% (95% CrI, 7.10%, 8.69%) and the lowest in those aged 75 years and over at 2.46% (95% CrI, 2.16%, 2.80%). Weighted prevalence also increased between rounds in all regions, with local areas of high prevalence detected (Supplementary Table S3A
Linking prevalence and national severe outcome incidence Matching the daily prevalence of swab-positivity in REACT-1 to publicly available data on hospitalisations we estimated lag time of 19 (95% CrI 18, 20) days for rounds 1 to 12 (May 1, 2020 to May 20, 2021) and 16 (95% CrI 14, 18) days for rounds 13 (June 24 to July 12, 2021) to 16. We observed high consistency between daily prevalence and the (lagged) hospitalisations data from 15 June to 30 November 2021, but from 1 December onwards (when Omicron variant began to emerge in REACT-1 data), the 16 days lag may appear too long. This may to some extent reflect incidental hospitalisations with Omicron due to its high population prevalence and support using an Omicronspecific lag time, but data were too sparse in the present study to be able to estimate such a lag with suitable precision.
Our results also showed: a close correspondence of prevalence of swab-positivity with hospitalisation (shifted by the estimated lag) through round 8 (January 6 to 22, 2021); an apparent reduced risk of hospitalisation vs. prevalence through rounds 9 to 11 (February 4 to May 3, 2021); the two coming together again in rounds 12 and 13 (May 20 to July 12, 2021) as Delta replaced Alpha; a further period of reduced risk of hospitalisation during rounds 14 to 15 (September 9 to November 5, 2021); and finally a further coming together in December 2021 as Omicron took off, followed by a rapid divergence once again (Fig. 4). Time lag estimates for death were 26 (95% CrI 25, 26) days for rounds 1-12 and 16 (95% CrI 15, 17) days for round 13-15 and trends showed a marked and consistent reduction in risk of death compared to prevalence of swab-positivity throughout rounds 9 to 16 (February 4 to December 14, 2021).

Discussion
The Omicron epidemic in January 2022 was further advanced in England than in the USA and most European countries. During December 2021 Omicron almost completely replaced Delta in England 5-7 , with the estimated peak in prevalence around six weeks after Omicron was first identified in England. Although the BA.1 sub-lineage still dominated, our data during January 2022 show an increase in the proportion of daily infections from BA.2 compared to BA.1 and its sub-lineage BA.1.1 with an R advantage of 0.46. Similar data showing a transmission advantage of BA.2 compared to BA.1 were reported from the national routine testing data in the UK 14 and in Denmark 15 .
As a result of the rapid rise in Omicron infections, during round 17 (January 2022), we saw the highest prevalence of SARS-CoV-2 infections ever observed in the REACT-1 study to that time. Overall prevalence was nearly three-fold higher than at the peak of the second wave in January 2021, with a near 12-fold increase in the oldest age group (75 years and over) since round 16 (November to December 2021). However, the dynamics underlying these population-level trends are complex with the prevalence decreasing in adults but increasing in children through January 2022. This is likely the consequence of the peak occurring during the end-of-year school break, causing a delay to school-based transmission among children. Also the high rates observed in children during round 16 were predominantly driven by Delta 5 , whereas round 17 trends were overwhelmingly due to Omicron infections. Essential/key workers other than healthcare/care home workers were found to be at increased risk of infection, even when other risk factors were adjusted for. Importantly, this group reported, on average, more contacts in the day before swabbing than other groups. How rates of contact have changed for different risk groups over the course of the pandemic warrants further investigation. An estimated 20 to 40 times greater antibody titre is required for neutralization of Omicron than for Delta 16 , although individuals who had received a third vaccine dose did show increased neutralisation of the Omicron variant 17,18 . Among infected individuals in round 17, the distribution of Ct values was similar in adults who had received three compared to two doses of vaccine, suggesting that the booster did not affect viral load for Omicron.
These findings notwithstanding, crucially, booster doses remain highly effective at reducing the risks of severe disease 3,19,20 . Our comparison of infection prevalence data from REACT-1 and public data on hospitalisations and deaths indicate that although trends in these serious outcomes continue to track infections (albeit with a time lag), this is at a lower level than in previous waves, before widespread rollout of vaccination in England (from January 2021) and greater access to effective therapeutics for severe disease (such as corticosteroid and other immunomodulatory therapy) 21 . Nonetheless, it will be important to continue to monitor hospitalisations and deaths closely in view of the high levels of infection, including among the older population, and as restrictions were lifted in England and elsewhere 22,23 .
Our study has limitations. In round 17, 12.2% of the invited participants returned swabs producing valid RT-PCR test results. Our descriptive analyses showed that those who agreed to participate were slightly older and from more affluent areas than the general population. We use weights, calculated for each participant in each round, to adjust for differential response rates in calculating prevalence estimates, but these corrections may not fully eliminate all biases. As implemented in subsequent rounds of REACT-1 incentives can improve the response rate in the younger and less affluent populations 24 . Changes in the way the swab samples were transported and tested may have introduced small changes in results across rounds, although these should not have affected within-round trends. Our results comparing trends in infections with those in hospitalisations and deaths suggest a divergence in the most recent data consistent with reduced severity of Omicron. However, we were unable to assess in our data to what extent this reflects the fact that the Omicron wave occurred at a time of high levels of immunity in the population as a result of both natural infection and the successful rollout of the vaccination programme in England 25 .
In conclusion, we have shown a substantial and rapid rise in infections from early December 2021 through January 2022 as the Omicron variant took hold and almost completely replaced Delta in England. Among school-aged children there was a rise in prevalence as they returned to school in January after the end-of-year break 14,19,20 . Many countries relaxed restrictions as the peak in Omicron cases passed, including in England where the requirement to self-isolate if testing positive for COVID-19 26 was removed despite the high prevalence of SARS-CoV-2 infection following contact with a known case. While vaccination (including the booster campaign) remains the mainstay of the defence against SARS-CoV-2 given the high levels of protection against severe disease 14,19,20 , further measures beyond vaccination may be required in the future in the event of the emergence of a new variant with greater potential than Omicron for hospitalisations and deaths.

Study design
The REACT-1 study involves a series of cross-sectional surveys of random samples of the population of England at ages 5 years and over 27 , conducted approximately monthly since May 2020. Round 17 (January 5 to 20, 2022) included 102,174 participants with a valid selfadministered throat and nose swab test result for SARS-CoV-2 by reverse transcription polymerase chain reaction (RT-PCR) (including 862 samples obtained between January 21 and 24, 2022). Up to round 13 (June 24 to July 12, 2021), we collected dry swabs sent by courier to the laboratory on a cold chain but from round 14 (September 9 to 27, 2021 including 509 samples from 28-30 September) we switched to 'wet' (saline) swabs which (round 14) were sent to the laboratory either by courier (no cold chain) or priority post, and from round 15 (October 19 to November 5, 2021) onwards by priority post only. From round 16, we included a multiplex for detection of influenza A and B as well as SARS-CoV-2 (only the SARS-CoV-2 results are presented here). Because of delays in the post for return of swabs, we include a small proportion of samples obtained after the nominated closing date for each round of the study from round 14.
The sampling frame was the general practitioner list of patients in England held by the National Health Service. Participants registered and completed a questionnaire, providing information on age, sex, residential postcode, ethnicity, household size, occupation, potential contact with a COVID-19 case, symptoms and other variables 28 . We used the residential postcode to link to an area-level Index of Multiple Deprivation 29 and urban/rural status 30 . A positive test result was recorded if both N gene and E gene targets were detected or if N gene was detected with Ct value below 37. Initially we aimed to obtain approximately equal numbers of participants in each lower-tier local authority (LTLA) in England (N = 315), but from round 12 (May 20 to June 7, 2021) we switched to obtaining a random sample in proportion to population size at LTLA level. We compare results for SARS-CoV-2 with those obtained during October to December 2021 in round 15 (N = 100,112) 31 and round 16 (N = 97,089) 5 .
Samples testing positive with Ct 34 or less in either the E or N gene were sent for viral genome sequencing to the Quadram Institute, Norwich, UK. We used the ARTIC protocol 32 (version 4) for viral RNA amplification, CoronaHiT for preparation of sequencing libraries 33 , the ARTIC bioinformatics pipeline 32 and assigned lineages using Pango-LEARN (version 2022-01-20) 34 .

Data analyses
We estimated round-specific weighted prevalence using random iterative method (rim) weighting 35 to provide prevalence estimates for the population of England as a whole, and by region, adjusting for age, sex, deciles of the Index of Multiple Deprivation 29 , LTLA counts, and ethnic group and 95% credible intervals overall and for sub-groups. For round 17 we fit a Bayesian logistic regression model to the proportion of BA.2 lineage compared to the BA.1 lineage (or its sub-lineage BA.1.1) to estimate daily growth rate advantage for the odds of BA.2 versus BA.1. We then estimated the additive reproduction number (R) advantage as the product of the daily growth rate advantage and the Omicron-specific mean generation time 36 .
We used an exponential model of growth/decay to examine temporal trends in swab-positivity assuming a binomial distribution for the proportion of positives by day of swabbing where reported (or day of first Post Office scan if available) using a bivariate No-U-Turn Sampler and a uniform prior distribution for the probability of swabpositivity 37 . R was estimated assuming a gamma-distributed generation time with Omicron-specific mean 3.3 days and standard deviation 3.5 days (shape n = 0.89 and rate β = 0.27) as 36 : where r is the daily exponential growth/decay rate.
To visualise temporal trends in swab-positivity, we used a No-U-Turn Sampler in logit space to fit a Bayesian penalised-spline (P-spline) model 38 to the daily data, split into approximately 5-day sections by regularly spaced knots. Edge effects were minimised by adding further knots beyond the study period. We used fourth-order basis splines (bsplines) over the knots, including a second-order random-walk prior distribution on the coefficients of the b-splines to guard against overfitting; the prior penalised against changes in growth rate unless supported by the data 39 . We also fit age-group-specific P-splines separately with the smoothing parameter obtained from the model fit to all the data.  We used a neighbourhood spatial smoothing method to examine geographical variation in SARS-CoV-2 prevalence at the LTLA level. For each of 15 randomly selected participants within an LTLA, we calculated the prevalence of infection among the nearest M people, where M was the median number of study participants within 30 km, and then estimated the smoothed neighbourhood prevalence in that area.
We compared (using a Kruskal-Wallis test) Ct values as a proxy for viral load among test-positive swabs (N gene and E gene where Ct>0), by vaccination status (lagged by a 14-day period from date of vaccination) and symptom status across rounds 15 to 17, where information on vaccination history and dates of vaccination was obtained (with consent) by linking to data from the national COVID-19 vaccination programme.
We used logistic regression to estimate the odds of testing positive by employment, ethnicity, household size, children in household, urban area, and deprivation, adjusting for age, region and the other variables examined. We estimated the average number of contacts in the day before swabbing (excluding contacts with household members) by employment type to aid interpretation of the logistic regression results.
We compared daily swab-positivity prevalence in REACT-1 with daily hospitalisations and (separately) deaths from (external) national data. We first fit P-spline models as described above to the daily hospital admissions and to the daily death data. We then fit a model to match daily weighted prevalence, and smoothed daily hospitalisation/death. This model included (i) a time lag parameter corresponding to the time between the P-spline estimate for hospitalisations or deaths and daily prevalence, and (ii) a scaling parameter, corresponding to the percentage of swab-positive participants with (lagged) hospitalisations or deaths. Estimation was done using in-house R scripts maximising the binomial likelihood based on daily weighted prevalence. The scaling parameter was estimated using REACT-1 data up to and including round 7 (November 13 to December 3, 2020), before the vaccination programme in England began and was kept constant thereafter. To account for variant-specific lags, we considered the original lag estimate (based on rounds 1 to 7) for round 1 (May 1 to June 1, 2020) to round 12 (May 20 to June 7, 2021) when Alpha dominated in England and a second lag parameter thereafter, estimated from round 13 (June 24 to July 12, 2021) to round 16 (November 23 to December 14, 2021), when Delta dominated. The second lag parameter was estimated fixing the scaling parameter to the estimate obtained based on analysis of data from rounds 1 to 7.
Analyses were performed with R software, version 4.0.5.

Ethics
We obtained research ethics approval from the South Central-Berkshire B Research Ethics Committee (IRAS ID: 283787). Participants in the study (or their parents or guardians for children) provided informed consent.

Public involvement
A Public Advisory Panel provides input into the design, conduct, and dissemination of the REACT research program.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Access to REACT-1 individual-level data is restricted to protect participants' anonymity. Summary statistics, descriptive tables, from the current REACT-1 study are available at https://doi.org/10.5281/zenodo. 6819880. REACT-1 study materials are available for each round at https://www.imperial.ac.uk/medicine/research-and-impact/groups/ react-study/react-1-study-materials/. Sequence read data are available without restriction from the European Nucleotide Archive at https:// www.ebi.ac.uk/ena/browser/view/PRJEB37886, and consensus genome sequences are available from the Global initiative on sharing all influenza data (GISAID). Accession numbers for these data are available in Supplementary Data 1.