key: cord-0986848-x807fpa7
authors: Backer, J. A.; Mollema, L.; Klinkenberg, D.; van der Klis, F. R. M.; de Melker, H. E.; van den Hof, S.; Wallinga, J.
title: The impact of physical distancing measures against COVID-19 transmission on contacts and mixing patterns in the Netherlands: repeated cross-sectional surveys
date: 2020-05-20
journal: nan
DOI: 10.1101/2020.05.18.20101501
sha: d901641b87428ce8a3eead24357cfe9764970178
doc_id: 986848
cord_uid: x807fpa7

Background During the current pandemic of coronavirus (COVID-19) many countries have taken drastic measures to reduce transmission of SARS-CoV2. The measures often include physical distancing that aims to reduce the number of contacts in the population. Little is known about the actual reduction in number of contacts as a consequence of physical distancing measures. Methods In the Netherlands, a cross-sectional survey was carried out in 2016/2017 in which 8179 participants retrospectively reported the number, age and gender of different persons they had contacted (spoken to in person or touched) during the previous day. The survey was repeated among 2830 of the original participants, using the same questionnaire, in March and April 2020 after physical distancing measures had been implemented. Results The average number of contacts in the community was reduced from on average 12.5 (interquartile range: 2-17) to 3.7 (interquartile range: 0-4) different persons per participant, a reduction of 71% (95% confidence interval: 71-71). The reduction in the number of community contacts was highest for children and adolescents (between 5 and 20 years) and smallest for elderly persons of 80 years and older. The reduction in the effective number of total contacts, measured as the largest eigenvalue of the matrix with community and household contacts, was 62% (95% confidence interval: 48 - 72). Conclusion The substantial reduction in contacts has contributed greatly in halting the COVID-19 epidemic. This reduction was unevenly distributed over age groups, household sizes and occupations. These findings offer guidance for the lifting of age-group targeted measures.

Since the beginning of 2020, the novel coronavirus SARS-CoV2 that causes COVID-19 disease has rapidly spread around the world. Most patients only have mild symptoms, but mainly elderly and persons with comorbidities can develop severe acute respiratory disease (China CDC, 2020) . Hospitals have been confronted with a fast growing number of patients, often exceeding the intensive care (IC) capacity. Many countries have implemented control measures that include a combination of increased hygiene, travel restrictions, case finding, contact tracing and physical distancing measures. The specific physical distancing measures differ between countries and between regions; their overall aim is to reduce the number of contacts in the population thus preventing the transmission of infection. The impact of these measures on the reduction of contacts in the population, and the reduction of contacts made by different groups within that population, are poorly quantified.

Different approaches exist to measure behavioural changes. Mobile telephone data provided by telecom companies are used to measure the change in mobility patterns (Oliver et al., 2020; Pepe et al., 2020) . Similarly, the location history of smart phones can be tracked with apps (Google 2020). The anonymized and aggregated mobility patterns can suggest changes in contact patterns in the population at large. To obtain direct and detailed information on contacts, cross-sectional studies are conducted in which participants report their age and gender as well as the age and gender of all persons they have contacted during a single day (Mossong et al., 2008; Hoang et al., 2019) . Contact surveys have been successfully used to quantify the reduction in the number of contacts associated with physical distancing measures in Shanghai and Wuhan, China, estimated at 88% and 86%, respectively (Zhang et al., 2020) , and in the UK (Jarvis et al., 2020) among the adult population, at 74%. One of the challenges with the contact survey approach is to obtain a reliable baseline measurement before physical distancing measures were implemented. In the Wuhan study participants needed to recall the number of contacts during a regular weekday at the end of 2019. In the UK study the baseline was provided by a similar study conducted 13 years ago among a different representative UK study population (Mossong et al., 2008) .

We present a large contact survey conducted in the Netherlands in March and April 2020.

The participants were recruited from a large sample of the Dutch population who had . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 20, 2020 . . https://doi.org/10.1101 participated in an earlier large cross-sectional survey in 2016/2017 (Verberk et al., 2019) .

The contact questionnaire was nearly identical in both surveys, which allows us to use the first survey as a baseline measurement, and the second survey as a measurement of contacts during the implementation of physical distancing measures. By 16 March, the Netherlands had imposed physical distancing measures including closing daycare centers, schools, universities; working from home whenever possible; closing of cafes, pubs, restaurants, theaters, cinemas, and sport clubs; cancelling events with more than 10 persons attending; maintaining 1.5 m distance from others outside one's household.

By comparing the survey results before and after the implementation of these physical distancing measures, we could determine the impact on the number of contacts made in the community (i.e. outside the household), distinguishing between different age groups, genders, household sizes, days of the week and occupations. We also assessed how the physical distancing measures have affected the total number of contacts, including contacts with household members, and the age-specific mixing patterns.

Between February 2016 and October 2017, a cross-sectional study was conducted in a sample of the Dutch population from 0 to 89 years of age (Verberk et al., 2019) , henceforth referred to as the baseline survey. Participants were randomly selected from the Dutch population registry using a two-stage cluster design. Infants under 1 year of age, persons living in areas with low vaccination coverage, and persons with a migration background were oversampled in this survey. The study consisted of an extensive questionnaire, and included questions regarding the participants' age, gender and occupation, the age and gender of their household members, and the total number of unique persons they had contacted outside their household the previous day and which day of the week (i.e., Monday through Sunday) that was. Participants reported the age of the contacted persons in 11 age groups (0-4, 5-9, 10-19, 20-29, 30-39, 40-49, 50-59, 60-69, 70-79, 80-89, 90+) . A contact was defined as a conversation in person, or a physical contact.

Of the 8179 participants that had filled out the questionnaire in 2016/2017, 6712 had indicated to be willing to participate in future research. Of these, 6102 were invited on 26 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 20, 2020 . . https://doi.org/10.1101 March 2020 to participate in the follow-up study, referred to as 'the physical distancing survey'. As of 26 April 2020, 2830 had returned a questionnaire. Participants not reporting their household composition or reporting more than 100 contacts were excluded from the analysis.

The questionnaire in the physical distancing survey was identical to the baseline survey questionnaire, apart from two questions. A question asking whether participants had had any contacts outside their household was added (before asking how many), and the question about occupation was changed to cover occupations that involve many contacts and that are part of vital processes.

We analysed the baseline and physical distancing contact surveys by comparing the number of contacts in the community per participant stratified by several characteristics: age, gender, household size, day of the week, and occupation (as reported in the physical distancing survey, under the assumption that participants did not change occupation between the two surveys).

We obtain the total number of contacts in the population by adding contacts with reported household members to the reported contacts in the community. We estimate age-stratified contact matrices that contain the numbers of contacts made between and within age groups, using an approach that accounts for reciprocity of contacts between different age groups ( Van de Kassteele et al., 2017) , and using age-specific population size data for the Netherlands on 1 January 2017 and on 1 January 2019 (Statistics Netherlands, 2019a). To check the effect of enforcing reciprocity between contacts, we compare the estimated and observed mean number of contacts per participant.

We characterize the mixing pattern of the age-specific contacts by the disassortativeness index (Farrington et al., 2009) , and we characterize the effective number of age-specific contacts by the largest eigenvalue of the contact matrix (cf. Diekmann et al., 2010) .

All analyses were done using R version 3.6.0 (R Core Team, 2019).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 20, 2020. . https://doi.org/10.1101/2020.05.18.20101501 doi: medRxiv preprint

In total 7548 baseline survey participants and 2698 physical distancing survey participants were included for analysis. The composition of the survey population by age and gender reflected the Dutch population ( Figure 1 ). The mean age was 35 years (range 0 -88) in the baseline survey, and 39 years (range 3 -87) in the physical distancing survey. Each age group consisted of more than 100 participants in both surveys, except for the 80+ age group in the physical distancing survey ( Table 1 ). The household size distribution for the baseline survey population reasonably reflects the household size distribution in the Netherlands (Statistics Netherlands, 2019b) , whereas the household size distribution for the physical distancing survey population contained relatively few single-person households. The reported average household size of 2.8 in the baseline survey is smaller than the reported average household size of 3.3 in the physical distancing survey. There were more than 100 participants reporting their contacts for each day of the week, for both surveys.

The percentage of participants who did not report any community contacts increased from 17% in the baseline survey to 41% in the physical distancing survey. On average, a participant had 12.5 (2 -17, interquartile range or IQR) community contacts per day in the baseline survey, and 3.7 (0 -4) community contacts in the physical distancing survey (Tab.

2).

In the baseline survey, participants aged 10 to 19 years had the highest number of contacts, and this number gradually declined with increasing age. In contrast, the number of contacts was similar across the different age categories in the physical distancing survey (Fig. 2) . The reduction in the number of contacts was greatest for participants aged 5 to 9 years, and lowest for participants aged 80 to 89 (Tab. 2).

The reduction in the mean number of contacts was similar for male and female participants.

The number of community contacts increased with household size in the 2017 survey, whereas this number was similar for the different household sizes in 2020. In the baseline . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 20, 2020. . https://doi.org/10.1101/2020.05.18.20101501 doi: medRxiv preprint survey, most contacts were made during weekdays, slightly fewer on Saturdays and fewest on Sundays. The reduction of contacts between the two surveys was similar for all days of the week, except for Saturdays, when participants decreased their contacts to a lesser extent.

The reduction in contacts varied between participants depending on their occupation.

Among the occupations represented by more than 40 participants, the greatest reduction was observed for schoolchildren and students, and the lowest reduction for those working in food industry, retail and health care.

In the estimation of the contact rates between age groups, the reciprocity between contacts is explicitly taken into account. The observed and estimated mean numbers of community contacts per participant are in agreement (Fig. 2) , showing that participants accurately reported their contacts.

The contact matrices for both contact surveys and for the different types of contact (community, household) are shown in Fig. 3 . The contact matrices for household members illustrate how participants live with persons in their own age category and with their children or parents (i.e., 30 years younger and older); this is apparent in both baseline and the physical distancing surveys. The matrices for contacts in the community indicate fewer contacts within the younger age groups in the physical distancing survey compared with the baseline survey. In the physical distancing survey, persons of 80 years and older had a relatively large number of community contacts with age groups from 50 to 79 years old. In this survey, 85% of the contacts with persons of 80 years and older were reported by health care workers. Taken together, the household and community contact patterns reveal that age-specific mixing does not change much between the baseline and the physical distancing surveys (disassortativeness index is 0.53 for the baseline and 0.54 for physical distancing survey). However, the effective number of contacts per person (i.e. the largest eigenvalue of the contact matrix) decreased from 16 (14 -21, 95% credible interval) in the baseline survey to 6.1 (5.4 -7.7) in the physical distancing survey, an average reduction of 62% (48 -72).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 20, 2020. . https://doi.org/10.1101/2020.05.18.20101501 doi: medRxiv preprint

In the Netherlands, physical distancing measures came into effect on 16 March 2020. Three weeks after implementation, the numbers of occupied hospital and IC beds reached a peak, after which they gradually declined. Through comparison with the baseline survey, physical distancing measures were estimated to have drastically reduced the numbers of contacts in the community, by 71%. The reduction in the mean number of community contacts varied with occupation, and tended to be larger for younger age groups, larger household sizes, and weekdays as compared to weekends. The effective number of the total contacts per person decreased by 62%.

This effective number of contacts per person is proportional to the reproduction number (i.e. the number of secondary infections caused by a single infectious person in the population) under two conditions. First, the definition of contact (having a conversation in person or physical contact) is a good proxy measure for at-risk contact events where SARS-CoV-2 can be transmitted. Second, all age groups are equally susceptible and infectious. As evidence accrues that children are less infectious or less susceptible (Zhang et al., 2020; Gudbjartsson et al., 2020) , and children report a larger reduction in contacts compared with other age-groups after imposition of physical distancing, a reduction in the effective number of contacts represents an upper bound for the reduction of the reproduction number. This, in turn, implies that the physical distancing measures that reduce the effective number of contacts by 62%, are sufficient to reduce a reproduction number with values up to 2.6, to 1 or lower, and halt the COVID-19 epidemic.

Elderly persons have a relatively small reduction in their number of contacts. Most of their reported contacts after the implementation of physical distancing measures were with health care workers. These remaining contacts are likely essential, and there is little prospect for further reducing the number of contacts for this age group. Measures other than physical distancing are required to protect the elderly who are the most susceptible to infection (Zhang et al, 2020) and have the highest risk of complications (China CDC 2020).

The study participants were randomly sampled from the Dutch population in 2016, with oversampling of zero-year olds, persons living in a low vaccination coverage area, and person with a migration background. We have identified the following limitations. Not all . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 20, 2020. . https://doi.org/10.1101/2020.05.18.20101501 doi: medRxiv preprint participants of the baseline survey indicated that they would participate in a follow up, and not all of those who were invited participated. For instance, 90% of participants in the physical distancing survey were indigenous Dutch, compared to 78% in the baseline survey (Verberk et al., 2019) . Apart from the potential for selection bias, there are differences in participant characteristics between the baseline study in 2016/2017 and the physical distancing survey in 2020, and these differences may have affected the results. First, the physical distancing survey participants are older, because they have aged in the intervening 3.5 years, with the consequence that the physical distancing survey does not contain any 0

to 2 year olds. To assess the effect on the mean number of contacts, we created a synthetic population with the size and age distribution of the physical distancing population, sampled from the baseline population. This synthetic population has on average 13.2 (12.7 -13.8, 95% bootstrap interval) community contacts per participant, which is higher than the baseline population with a mean of 12.5 community contacts per participant. This means that the contact reduction may have been underestimated. Second, the physical distancing survey participants reported a larger household size on average, which results in a greaterer number of household contacts. Third, the physical distancing participants reported their contacts more often on weekdays (85% versus 64% in the baseline survey), which would result in an underestimation of the reduction in the number of contacts. Fourth, the physical distancing survey was carried out in two months in spring, whereas the baseline survey was conducted over a period of almost two years. Because contact patterns change little throughout the course of a year (Béraud et al, 2015) , we do not expect this to substantially affect the estimated reduction. Fifth, the additional question in the physical distancing survey questionnaire that asked beforehand whether the participants had had any contacts allowed us to check the reliability of their reported number of contacts. The reported number of contacts in the baseline survey, without such a check, may have been larger. All these differences together may have led to a slight underestimation in the reduction in the number of contacts.

The reduction in the number of contacts associated with physical distancing measures in the Netherlands is smaller than the reductions of 88% and 86% observed in Shanghai and Wuhan, China (Zhang et al., 2020) , most probably because of the complete lockdown in both cities. In the UK a 74% reduction in the number of all contacts among adults (18+) has . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 20, 2020. . https://doi.org/10.1101/2020.05.18.20101501 doi: medRxiv preprint been reported (Jarvis et al. 2020) . To compare this with our results, we calculate the total number of contacts including community and household contacts for the participants in our surveys who are 18 years and older, and find a reduction of 56%. At the time of the studies the control measures in the UK and the Netherlands rank at similar values of the stringency index (Hale et al., 2020) . The effect on contact reduction differs, possibly due to study population differences or compliance.

The results of this study have immediate application for a number of important issues of concern to the public health response and management of the COVID-19 pandemic. The estimated contact reduction is applicable to other countries and regions with similar control measures. The estimated age-specific contact matrices are useful for conducting scenario analyses with age-structured transmission models of COVID-19, to project the future course of the epidemic under physical distancing measures, and to explore the possible effects of lifting them Salje et al., 2020) . As we plan to repeat the contact survey at regular intervals, we will be able to monitor the number of contacts made over time, and detect the impact of future changes in these measures such as the reopening of schools, as well as changes in compliance with the measures that remain. We believe that contact surveys such as these can help to inform and guide infection control measures in this time of unprecedented physical distancing, and in the next phase of the epidemic, when the most drastic control measures might be lifted.

We gratefully acknowledge the participants of the contact surveys in 2016/2017 and 2020, as well as Dr. S. A. McDonald for critically assessing the manuscript.

The MEC-U (Medical Research Ethics Committees United; R20.022) has approved the research protocol "Third population-based immune surveillance study used for the evaluation of immunity against SARS-CoV-2' (PIENTER Corona)" and written informed . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 20, 2020 . . https://doi.org/10.1101 consent was obtained from all adult participants, and parents or legal guardians of minors included in the study.

Data of contact matrices shown in Fig. 3 are available as tab separated file (S1_contact_matrices.tsv).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 20, 2020 . . https://doi.org/10.1101 1554 (20.6) 0 (0) * Households in the Netherlands consist of 1 (38%), 2 (33%), 3 (12%), 4 (12%) and 5+ (5%) persons (Statistics Netherlands, 2019b) , leading to an expected distribution per participant to live in a 1 (17%), 2 (31%), 3 (17%), 4 (23%) or 5+ (12%) person household.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 20, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 20, 2020. . https://doi.org/10.1101/2020.05.18.20101501 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 20, 2020. . https://doi.org/10.1101/2020.05.18.20101501 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 20, 2020 . . https://doi.org/10.1101 

The French Connection: The First Large Population-Based Contact Survey in France Relevant for the Spread of Infectious Diseases

China CDC The Novel Coronavirus Pneumonia Emergency Response Epidemiology Team. The epidemiological characteristics of an outbreak of

2020, Vital Surveillances -China CDC Weekly, 2020 available from cdn.onb.it

The construction of next-generation matrices for compartmental epidemic models

Measures of disassortativeness and their application to directly transmitted infections

Spread of SARS-CoV-2 in the Icelandic Population

Oxford COVID-19 Government Response Tracker, Blavatnik School of Government

A Systematic Review of Social Contact Surveys to Inform Transmission Models of Closecontact Infections

Quantifying the impact of physical distance measures on the transmission of COVID-19 in the UK

Social contacts and mixing patterns relevant to the spread of infectious diseases

Mobile phone data for informing public health actions across the COVID-19 pandemic lifecycle

Ciro Cattut o, Michele Tizzoni COVID-19 outbreak response: a first assessment of mobility changes in Italy following national lockdown

The effect of control strategies to reduce social mixing on outcomes of the COVID-19 epidemic in Wuhan, China: a modelling study

R: A language and environment for statistical computing. R Foundation for Statistical Computing

Estimating the burden of SARS-CoV-2 in France. 2020. pasteur-02548181, available from halpasteur

Third national biobank for population-based seroprevalence studies in the

including the Caribbean Netherlands

Efficient estimation of age-specific social contact rates between men and women

Using data on social contacts to estimate age-specific transmission parameters for respiratory-spread infectious agents

Changes in contact patterns shape the dynamics of the COVID-19 outbreak in China