Ethics statement
Participation in this opt-in study was voluntary, and all analyses were carried out on anonymised data. The study was approved by the ethics committee of the London School of Hygiene & Tropical Medicine reference number 21795.
Survey methodology
We commissioned the market research company Ipsos to conduct a survey of UK adults (referred to here as the CoMix survey). Adults (≥ 18 years) were recruited into the survey by sending email invitations to existing members of their online panel. Representativeness of the general UK population was ensured by setting quotas on age, gender, geographical location, and socioeconomic status. This cohort of individuals will be requested to answer the survey every 2 weeks for a total of 16 weeks to track changes in their self-reported behaviour. The first surveys were sent on Tuesday, 24 March, 1 day after a lockdown was announced for the UK.
Participants were asked about their attitudes towards COVID-19 and the effect of physical distancing interventions, whether they or any of their household members experienced any recent symptoms, whether they were tested for COVID-19, whether they had had any contact with known COVID-19 cases, and whether they were affected by physical distancing measures.
Participants reported (i) if any person in their household were advised to quarantine, isolate, or limit time in their workplace or educational facility in the preceding 7 days due to COVID-19 and (ii) if they heeded the advice and isolated, quarantined, or stayed away from their workplace or educational facility. In the survey, we defined quarantine as limiting contacts and staying at home, with restricted allowance for movement outside the home after a potential exposure with a COVID-19 case. We defined isolation as completely separating from uninfected contacts, including household members, either in the home or in a health facility. To assess the impact of advice and policy changes regarding physical distancing, we asked participants to indicate if they had planned to participate in a set of events in the preceding week. For each event type, they reported (i) whether they proceeded with their plan, or (ii) if it was cancelled or they decided not to go, and (iii) the frequency of the event type in the previous 7 days. Additional questions were asked about preventive behaviours, such as handwashing or wearing masks, and about the use of public transport in the previous 7 days.
In addition, we asked participants to record all direct contacts made between 5 am the day preceding the survey and 5 am the day of the survey. A direct contact was defined as anyone who was met in person and with whom at least a few words were exchanged, or anyone with whom the participants had any sort of skin-to-skin contact. We were unable to ask parents to provide contact information for their children due to lack of ethical approval; however, participants were able to list contacts who were under 18.
For every recorded contact, participants documented the age and gender of the contact, relationship to the contact, the frequency with which they usually contact this person, whether contact was physical (skin to skin) or not, and the setting where the contact occurred (e.g. at home, work, school, or whilst undertaking leisure activities), including whether contact occurred in- or outside an enclosed building. Questions on social contacts were consistent with those from the UK arm of the POLYMOD survey [14], which was used as the baseline pre-pandemic comparison dataset. Details on survey methodology, the study protocol, and a copy of the questionnaire used are provided in Additional files 1 and 2.
Statistical analysis
R version 3.6.3 was used for all analyses; the code and data are available on github (see the “Availability of data and materials” section) [15,16,17].
We grouped study participants and contacts into the following age bands: 18–29, 30–39, 40–49, 50–59, 60–69, and 70+. Age, gender, and locations of participants were compared to the 2018 mid-year estimates provided by the UK Office of National Statistics (ONS) to assess the representativeness of the study sample [18]. We descriptively analysed answers related to symptoms, attitudes, exposure to physical distancing measures, and individual preventative measures. We present the number and percentage or mean and standard deviation where appropriate (Table 3).
We calculated the average number of social contacts per person per day overall, and stratified by age category, sex, household size, location of contact, type of contact, and day of the week. We then compared the mean total number of daily contacts by age group to POLYMOD stratified by contact location.
We calculated social contact matrices for the age-specific daily frequency of direct social contacts, adjusting for the age distribution in the study population and reciprocity of contacts, using the socialmixr package in R [19].
As children (< 18 years) were not included as survey participants, we imputed contacts for younger age groups (child-child and child-adult contacts) using the POLYMOD UK data. Specifically, for those child contact groups that were missing, we used a scaled version of the POLYMOD social contact matrix. Following previous methods developed by Klepac et al. [20], as the scaling factor, we took the ratio of the dominant eigenvalues of the POLYMOD and CoMix matrices, for all age groups present in both studies, stratified by setting. Furthermore, to reflect school closures during the collection of our survey, we removed school contacts from the POLYMOD data from our analysis.
The basic reproduction number, or R0, is the average number of secondary infections arising from a typical single infection in a completely susceptible population and can be estimated as the dominant eigenvalue of the next generation matrix [21]. The exact form of the next generation matrix is model dependent. For respiratory infections, such as SARS-CoV-2 (the pathogen causing COVID-19), this is usually a function of the age-specific number of daily contacts, the probability that a single contact leads to transmission, and the total duration of infectiousness. Therefore, R0 is proportional to the dominant eigenvalue of the contact matrix [19].
We assumed that contact patterns prior to physical distancing were similar to those observed in the POLYMOD data and that the duration of infectiousness and the probability that a single contact leads to transmission did not change during the study period. We also assume that all age groups contribute equally to transmission. Under these assumptions, the relative reduction in R0 is equivalent to the reduction in the dominant eigenvalue of the contact matrices. By multiplying the value of R0 prior to the interventions by the ratio of the dominant eigenvalues from the POLYMOD and CoMix contact matrices, we were able to calculate R0 under the physical distancing interventions. Prior to interventions, we assumed R0 followed a normal distribution with mean 2.6 and standard deviation of 0.54 based on a meta-analysis of the literature presented in Additional file 3 [22,23,24,25,26,27,28,29,30,31,32,33,34].
To assess uncertainty, we repeated the age imputation process by taking 10,000 bootstrapped samples from both POLYMOD and CoMix matrices. For every bootstrap sample, we calculated the ratio between the dominant eigenvalues for the sampled POLYMOD and CoMix matrices. This sampling provided a distribution of relative change in R0 from the contact patterns observed in POLYMOD and CoMix. Subsequently, we scaled the initial distribution of R0 with the distribution of bootstrap samples to estimate R0 under physical distancing interventions.
Recent results of the BBC Pandemic study [20] suggested a decrease of nearly 50% in the average number of contacts made by teenagers (13–18 years) compared with the POLYMOD data. We assessed the sensitivity of our results to a potential reduction in contacts over time by taking a conservative reduction of 50% between 5 and 18 year olds in the POLYMOD study and repeating our approach to estimate the reduction in R0.