key: cord-0319067-3s2exavc
authors: Hoang, T. V.; Coletti, P.; Kiffle, Y. W.; Van Kerckhove, K.; Vercruysse, S.; Willem, L.; Beutels, P.; Hens, N.
title: Close contact infection dynamics over time: insights from a second large-scale social contact survey in Flanders, Belgium, in 2010-2011
date: 2020-10-02
journal: nan
DOI: 10.1101/2020.09.30.20204891
sha: 6684a8f6f9c222dc608ea45faf782ac2cb4f2eea
doc_id: 319067
cord_uid: 3s2exavc

Background: In 2010-2011, we conducted a social contact survey in Flanders, Belgium, aimed at improving and extending the design of the first social contact survey conducted in Belgium in 2006. This second social contact survey aimed to enable, for the first time, the estimation of social mixing patterns for an age range of 0 to 99 years and the investigation of whether contact rates remain stable over this 5-year time period. Methods: Different data mining techniques are used to explore the data, and the age-specific number of social contacts and the age-specific contact rates are modelled using a GAMLSS model. We compare different matrices using assortativeness measures. The relative change in the basic reproduction number (R0) and the ratio of relative incidences with 95% bootstrap confidence intervals (BCI) are employed to investigate and quantify the impact on epidemic spread due to differences in gender, day of the week, holiday vs. regular periods and changes in mixing patterns over the 5-year time gap between the 2006 and 2010-2011 surveys. Finally, we compare the fit of the contact matrices in 2006 and 2010-2011 to Varicella serological data. Results: All estimated contact patterns featured strong homophily in age and gender, especially for small children and adolescents. A 30% (95% BCI [17%; 37%]) and 29% (95% BCI [14%; 40%]) reduction in R0 was observed for weekend versus weekdays and for holiday versus regular periods, respectively. Significantly more interactions between people aged 60+ years and their grandchildren were observed on holiday and weekend days than on regular weekdays. Comparing contact patterns using different methods did not show any substantial differences over the 5-year time period under study. Conclusions: The second social contact survey in Flanders, Belgium, endorses the findings of its 2006 predecessor and adds important information on the social mixing patterns of people older than 60 years of age. Based on this analysis, the mixing patterns of people older than 60 years exhibit considerable heterogeneity, and overall, the comparison of the two surveys shows that social contact rates can be assumed stable in Flanders over a time span of 5 years.

Infectious diseases and, more specifically, airborne infections can be transmitted between hosts via close contact 28 interactions; therefore, quantifying such interactions provides important information for properly modelling 29 infectious disease transmission. In recent years, we have witnessed a paradigm shift with respect to this: 30 whereas at the start of this century, mathematical models relied on simplifying assumptions such as homoge-31 neous mixing or on using mathematically convenient "Who Acquires Infection From Whom" constructs [1] , a 32 vast number of studies now rely on the use of social contact data [2, 3, 4, 5, 6, 7, 8] . 33 The literature on social contact surveys has shown how human interactions are heterogeneous in nature and 34 present a large degree of homophily in terms of age [9, 10] and gender [11] . The information coming from 35 social contact surveys is therefore usually summarized in what is called the social contact matrix, quantifying 36 the average number of contacts made between individuals within and between given age classes. Using the so-37 cial contact hypothesis [2] , i.e. assuming that transmission rates are proportional to social contact rates, these 38 data-driven mixing patterns have been implemented into models of infectious disease transmission showing 39 good correspondence to (sero)prevalence data; see, e.g., [4, 9, 12] . 40 Social contact survey data allow for an exploration of contact rate patterns stratified by age, gender, and loca-41 tion, which helps to better describe the structure of the transmission network [13, 14] . However, a systematic 42 review by Hoang et al. (2019) [10] showed that half of the social contact surveys before 2019 used convenience 43 sampling, while quite a few surveys were conducted in specific settings, e.g., schools or universities, and/or 44 focus on specific target groups; thus, it is impossible to extrapolate the results to an entire population. Even 45 in population-based social contact surveys with representative samples, two problems might still exist: the 46 sample does not cover all age ranges of the population, or the number of elderly participants is insufficient for 47 investigating mixing patterns of these people. Indeed, no study reported the contact rates of people up to 99 48 years old. 49 Particular attention has been devoted to behavioural changes with respect to individual health status (e.g., 50 being ill [6, 15, 16, 17] ), weather conditions [5] or day of the week (weekday or weekend in holiday/non-holiday 51 or regular periods [9, 18, 19, 20, 21, 22] ) -hereafter referred to as microscopic time settings, and how these 52 affect disease dynamics [6, 8, 19, 23, 24] . 53 The use of social contact data to inform modelling has become so prominent in recent works that it has also 54 been applied to settings for which social contact studies are not available, leading to the question of how social 55 contact matrices should be projected onto other geographical areas and in time [7, 25, 26, 27] . However, to the 56 best of our knowledge, there has been no empirical assessment of whether mixing patterns change over longer 57 time periods (e.g., years) within a particular population and how this should be taken into account when 58 projecting social contact matrices. We will refer to these as macroscopic time changes to mark the difference 59 with microscopic time changes. 60 A first population-based social contact survey in Belgium was conducted in 2006, and its results were reported 61 in [9, 19, 28] , in which the impact of microscopic time changes on the contact mixing pattern was investigated, 62 although this study was not designed for doing so. A second population-based survey in Belgium was con-63 ducted 5 years later in 2010-2011. This survey was conceived as an improvement over the 2006 survey, with a 64 larger sample size covering a wider age range of participants and a better distribution of surveyed participants 65 over four different time settings (weekday/weekend days in regular/holiday periods). 66 2 . 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 October 2, 2020. . In this work, we aim to describe and analyse the Flemish social contact survey from 2010-2011 by accounting 67 for the mixing patterns of people 0-99 years of age, with a special focus on elderly people. We study both the 68 impact of microscopic and macroscopic time changes on contact patterns, and we assess whether the contact 69 rates remain stable over 5 years timespan. with quota sampling by age, gender, and region, making it representative for the whole Belgian population. 76

Each participant was asked to fill in a background questionnaire and a paper diary in which they record their 77 contacts over 2 days: one randomly assigned weekday and one randomly assigned weekend day. Two types 78 of contacts were defined: (1) two-way conversations during which at least three words were spoken and (2) 79 contacts that involved skin-to-skin touching. Information recorded in the diary included the gender and the 80 exact age or presumed age interval of each contacted person over the entire day. Contact features included 81 frequency, location and duration. If participants established more than 20 professional contacts per day, then 82 they only had to provide an estimated number of professional contacts and the age interval(s) with whom 83 they interacted most. Contact information (e.g., contact age or contact duration) was then imputed for such 84 contacts. More details can be found in [28] . We will refer to those contacts as additional professional contacts. (less than 13 years old), which was completed by a proxy, e.g., parents or school teachers; one for people aged 90 13-60 years and one for people aged 60+ years, which could also be filled out by a proxy. A total of 1,774 91 participants were recruited by random digit dialing on mobile phones and landlines, with quota sampling by 92 age, gender and geographical location. The contact definitions were the same as those used in the 2006 survey. 93

Participants were asked to complete a background survey and record their social contacts in a paper diary 94 during one randomly assigned day. Information on additional professional contacts was imputed the same way 95 as done for 2006 data. Compared with the 2006 survey, the 2010-2011 survey explored more features that 96 might influence the number of contacts recorded: the health conditions of participants, time use, distance 97 from home, animal ownership and touching. To date, the impact of animal ownership and touching on social 98 contacts has been investigated [29] ,so has the impact of weather on social contacts [5] . Of particular focus 99 were people aged 60 years and above; i.e., participants up to 99 years of age were recruited, and information 100 about contact frequency with children and grandchildren and residence size for elderly people living in nurs-101 ing/elderly homes was recorded.

Since participants have been shown to be influenced by fatigue in reporting on multiple days [10, 22] were unreliable (many answers left blank, incoherent answers, etc.). We also excluded 46 people living in an 110 elderly/nursing home and explored the contact patterns of these people separately; in addition, 6 people aged 111 more than 90 years were removed to avoid problems related to data sparsity. As a result, the final sample for 112 the analysis of the 2010-2011 survey is 1,707 participants. We defined four microscopic time settings: regular 113 weekdays, regular weekends, holiday weekends, and holiday weekdays. Holiday periods include both public 114 holidays and weekends inside or adjacent to these holidays. More details on the number of participants by 115 age and microscopic time in both surveys can be found in SA1 Table S1 . The datasets of both surveys are 116 available online within the social contact data sharing initiative [30] and the SOCRATES platform [14] .

We start with a descriptive analysis to explore the socio-demographic characteristics of survey participants and 119 features of their reported contacts for the contact survey in 2010-2011. Subsequently, data mining techniques 120 are used to explore associations among variables of interest and contact profiles of survey participants. We 121 then investigate the factors associated with the number of contacts, differences in gender in mixing patterns 122 and the impact of holidays and weekends as a proxy for the impact of school closure on disease transmission. 123 We end with the comparison between the contact surveys from 2006 and 2010-2011 using different measures. 124

Data mining techniques 125 We use two unsupervised learning methods: association rules and clustering. Association rules are used 126 to assess the possible associations pertaining to contact features, e.g., type of contact (close or non-close), 127 duration and frequency of contacts, . . . , using support, confidence and lift values as measures of interestingness 128 [19, 31] (see SA2 for additional information). Rules are considered of interest only when the support value 129 exceeds 1%, equivalent to at least 3142 contacts involved in constructing the rules. The threshold for the 130 confidence is 70%, and rules with greater lift indicate stronger association. In addition to association rules, we 131 investigate contact profiles using a clustering method. The contact profiles are defined by (1) the number of 132 contacts per survey participant in six different locations (home, work, school, leisure, transport and other), (2) 133 characteristics of participants (age and gender) and (3) time indicators (weekday/weekend and regular/holiday 134 period). Clustering is implemented using the daisy function in the R package "Cluster" [32] and using the 135 Gower distance, which allows for mixed types of variables. We visualize the clusters by projecting them into 136 a low-dimensional space using a dimension reduction technique known as the t-distributed stochastic neighbor 137 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 October 2, 2020. We consider both physical and non-physical contacts, including additional professional contacts reported by 143 participants. We model the number of contacts using a weighted negative binomial regression model to account 144 for over-dispersion. Socio-demographic characteristics, the health status of participants and microscopic time 145 settings (weekdays/weekends and regular/holiday period) are included as possible determinants (descriptive 146 statistics see SA1). In addition, diary weights computed from age and household size are used to account 147 for under-/over-sampling over participant features [9, 28] . We perform variable selection using a random 148 forest analysis [31] and the likelihood ratio test (LRT). Interactions between age and microscopic time settings 149 are retained, as they are the two most significant determinants of the number of contacts reported in the 150 literature [10].

Estimating age-specific contact rates 152 We define the age-specific number of contacts y ijr as the number of contacts made by the r th participant in 153 age class i with people in age class j per day (i, j = 1, · · · , J; r = 1, · · · , n i ), where J is the number of age 154 classes, and n i is the number of participants in age class i.

The age-specific number of contacts y ijr is assumed to follow a negative binomial distribution to account 157 for over-dispersion [9] . This distribution is defined as y ijr |x ∼ N B(m ij , κ ij ) for a vector of covariates x, 158 in our case the age of the participant x 1i and the age of the contact x 2j . The mean and variance of this 159 distribution are defined as m ij and m ij + κ ij * (m 2 ij ), respectively, where κ ij is the over-dispersion parameter. 160 To model the age-specific number of contacts, we apply generalized additive models for location, scale and 161 shape (GAMLSS). This allows for modelling both the mean and variance (over-dispersion) parameters of the 162 negative binomial distribution over participants' age x 1i and contacts' age x 2j . We refer to SA3 and [34] for 163 details about the GAMLSS. When estimating the social contact matrix C, the reciprocal nature of making 164 contact needs to be taken into account, as m ij N i = m ji N j , where N i is the population size in age class i 165 (obtained from demographic data) [35] . Based on m ij and N i , the reciprocal contact rates c ij can be obtained 166

.

For all quantities of interest, introduced in this and following subsections, we use a non-parametric bootstrap 168 of participants, to obtain 95% percentile bootstrap confidence intervals (BCIs) [36] .

Measures of comparison between different mixing patterns 170 We use four different measures of comparison: two for measuring assortativeness, the relative change in R 0 and 171 the relative incidence (RI). We measure the assortativeness of contacts by age using 2 different indices. The first 172 index is Gupta's Q [37] , which ranges from 0 (= homogeneous mixing) to 1 (= completely assortative mixing). 173

The second index is I 2 s , as proposed in [38] ranging from 0 (= perfect assortativity) to 1 (= homogeneous 174 mixing).

The third measure is based on the basic reproduction number R 0 . R 0 is given by the dominant eigenvalue 176 of the next generation matrix G. Assuming the age-specific transmission rates β(i, j) are proportional to 177 the age-specific social contact rates c(i, j) (also known as the social contact hypothesis [2] ), the ratio of 178 5 . 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 October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204891 doi: medRxiv preprint dominant eigenvalues of the next generation matrix yields the relative change in basic reproduction number 179 using different mixing patterns. Lastly, the ratio of relative incidences (RRI) is used for comparison. The 180 expected age-specific RI in the population during the exponential phase is given by the leading right eigenvector 181 of the next-generation matrix [39] . 182 For more details, we refer to SA3.

Investigating gender differences in mixing patterns 184 To gain insights into possibly different mixing behaviour between males and females, we estimate the age-185 specific average number of contacts using the GAMLSS approach (as previously introduced) for all four 186 combinations of gender interactions (male-male, female-female, male-female and female-male). We use assor-187 tativeness measures (I 2 s and Q indices), and the RRI to study differences between matrices.

Investigating the impact of school closure on disease transmission 189 We estimate the impact of school closure based on social contact data using contact rates from holidays and 190 weekend days as a proxy and compare them with contact rates from regular weekdays. We use the changes in 191 R 0 and the RRI to quantify these differences. (see SA1 Table S1 ). Therefore, for the comparison of contact matrices between 2006 and 2010-2011, we only 196 use participants less than 65 years old and merged weekend-regular and weekend-holiday into one "weekend" 197 category to overcome the data sparsity problem. Both assortativeness indices, the change in R 0 and the RRI are used to compare mixing patterns. Furthermore, 207

we use the ratio of transmission rates that allows for the direct comparison of contact rates between contact 208 matrices (a cell-wise comparison). Lastly, we applied the methods outlined in [4] to use social contact matrices 209 to fit VZV serological data from Belgium based on the social contact hypothesis (i.e., constant proportionality, 210

[2]), with both contact data sets separately and compare the results. 211 6 . 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 October 2, 2020. during public holidays and 151 on weekend days during or adjacent to public holidays (2 cases did not indicate 216 the date). The average participant age was 38 years, and participants younger than 18 years of age accounted 217 for 22% of the sample. The average household size was 3, ranging from 1 to 11; participants with household 218 sizes of 2-4 accounted for nearly 75% of the sample size, while only 13% of participants lived in households 219 with more than 4 residents. Approximately 21% of the participants were still students, 48% had a job, 13% 220 were retired and approximately 11% were at home or unemployed. Nearly two-thirds of working participants 221 were office clerks, 19% were manual workers, and only 6% were self-employed. Nearly half the number of contacts involved touching (with missing information in 345 cases). More than 231 10% of all contacts were with household members. Daily contacts accounted for nearly one-third of the total 232 number of contacts, while only 10% were first-time contacts. Short contacts (less than 5 minutes) made up 233 approximately 15% of the total number of contacts; long contacts (longer than 1 hour) constituted nearly half 234 of the total number of contacts. Nearly two-thirds of all reported contacts were made at home, work and 235 school, while contacts at multiple locations accounted for only 6% of all contacts .

Data mining techniques 237 The association rules with the highest lift value are presented in Table S1 in SA2. Seventy-four percent of 238 daily contacts lasting longer than 4 hours involved skin-to-skin touching. In contrast, 81% of the contacts 239 lasting less than 5 minutes with non-household members were usually non-physical contacts. Contacts with 240 household members are the most influential factor in determining whether contacts occur on a daily basis. 241

Contacts lasting longer than 4 hours, occurring on weekdays in a regular period, tend to occur on a daily basis 242 (71%).

In the clustering analysis, the largest silhouette width is obtained for six clusters ( Figure S1 in SA2). The 244 cluster sizes ranged from 151 participants (cluster 5) to 443 participants (cluster 1). All clusters present a 245 strong connection with the microscopic time settings, including participants from only weekdays/weekends or 246 regular days/holidays (Table S2 in SA2). Some clusters are easy to interpret when looking at the cluster 247 members' features. Cluster 2, for example, is composed of participants whose average age is 9 years, with a 248 7 . 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 October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204891 doi: medRxiv preprint large number of contacts at school, i.e., school-aged children. Cluster 4 includes participants with an average 249 age of 39 years and a large number of contacts at work, i.e., working-age adults. Other clusters present less 250 specific contact patterns but still exhibit a strong connection with microscopic time settings. Specifically, 251 cluster 1 includes participants who have a low number of contacts in all locations on regular weekdays. 252

Participants in this cluster have the highest average age (51 years) and can be interpreted as being socially 253 non-active. Cluster 3 includes participants surveyed on the weekend and regular period, with few contacts at 254 work and school but the highest number of contacts during leisure activities and at "other" locations. Cluster 5 255 contains participants surveyed in the weekend and holiday period, with no contacts at school, few contacts 256 at work and most contacts at home and in "other" locations. Cluster 6 consists of participants surveyed in 257 the weekday and holiday period, with an average of 6 contacts at work, a very low number of contacts during 258 leisure activities and transportation (see SA2 Table S2 and Figure S2 ). the model, as they strongly correlate with age, and that for participants younger than 13 years, the mother's 264 educational level is used instead of that of the participant. Other variables (gender, animal ownership, health 265 states regarding self-care and pain) were also excluded after model selection was performed (see Table S1 and 266 Figure S1 ) . The interactions between age and microscopic time indicators are highly significant. In Figure 1 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 October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204891 doi: medRxiv preprint younger than 18 years old, and this figure increases to 60% on weekdays and during regular periods. People 287 with a job have a much higher number of contacts than those who are retired or currently unemployed/job 288 seeking. Working people have most contacts at work (63% on a random day and 71% on regular weekdays). 289

When considering occupation, the number of contacts of office clerks is observed to be significant higher than 290 those of people with other occupations .

Modelling the number of contacts for participants older than 60 years who are not living in an elderly/nursing 292 home (SA3 Table S2 The effect of microscopic time settings is significant: more contacts were observed in weekend-regular periods 297 than in weekday-regular periods. The majority of elderly people who have children and/or grandchildren 298 reported having contacts with their children and grandchildren a few times per week or month. Figure 2 299 describes the social interaction of people aged 60+ years with other age groups. People aged 61-79 years have 300 the highest number of contacts with age group [40, 60) , which may describe the mixing pattern of people from 301 2 generations. Interaction between people aged 60+ years and young children/teenagers is significantly higher 302 on holiday-weekdays and weekend days than on regular-weekdays.

People living in an elderly/nursing home 304 Forty-six people reported living in a nursing/elderly home, with ages ranging from 79 to 99 years. Most of 305 them have health problems: some problems or not being able to perform their daily activities (96%), some 306 problems walking (67%) or staying in bed all the time (15%); some problems with self-care (46%) or not 307 being able to care for themselves (43%); and experiencing mild to serious pain (85%) and anxiety (48%). 308

These people reported 13.7 contacts on average, significantly higher than those aged 60+ years and living at 309 home (P<0.0.001, Mann-Whitney test). No statistically significant difference in the number of contacts for 310 people in elderly/nursing homes was found with respect to the residence size (P=0.47, Kruskal-Wallis test for 311 3 groups of residence sizes: <50, 50-100 and 100+). We compared people living in an elderly/nursing home 312 with people aged 60+ years living at home with respect to their social interaction with other age groups (see 313 SA3 Figure S2 ): almost no interaction with young children and teenagers is observed for people living in an 314 elderly/nursing home, while this interaction is more observed for people aged 60+ years living at home.

Overall contact patterns 316 The contact patterns by age group were summarized in a contact matrix displaying ages from 0 to 90 years, 317 whose elements represent the contact rate between an individual in a given age group and an individual in 318 another age group in the Flemish population. The resulting contact matrix shown in Figure 3 is described by 319 the pronounced main diagonal indicating contacts with individuals in the same age group, e.g., at home, at 320 school and at work, and the 2 less-pronounced sub-diagonals representing contacts between generations, e.g., 321 children and their parents. 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 October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204891 doi: medRxiv preprint Gender differences in mixing patterns 323 Figure 4 shows the age-and gender-specific average number of contacts. The assortative mixing pattern 324 characterized by the main diagonal is still observed for all interactions. male-female and female-male, respectively). Focusing on people aged less than 30 years, we noticed that the 328 assortative mixing pattern is more pronounced in male-to-male and female-to-female contacts. Specifically, 329 for contacts made between people in the same age groups, the average number of male-to-male contacts and 330 female-to-female contacts are 3.9 (BCI The inter-generational mixing pattern (mostly parent-child), marked by the two sub-diagonals, is similar for 333 females and males. The relative incidences for both genders also follow a similar pattern, with peaks at 334 approximately 15 years of age and between 40 and 45 years of age, and no difference is found in the overall 335 RI of males compared to that of females (see Figure S3 in SA3).

School closure impact 337 We observed a significant difference in R 0 between holiday and regular periods: the relative change in R 0 338 equals 0.71 (BCI [0.60; 0.85]), or equivalently a 29% reduction in R 0 for the holiday vs. the regular period. 339

When comparing the relative change in R 0 from a weekday to the weekend, a slightly higher reduction of 340 30% was observed. The difference in RI by age group is shown in Figure 5 . The comparison of weekdays to 341 weekends shows that the RI decreases significantly in the age group 0-15 years, while it is higher in the age 342 group [60,65) and [70,75) on the weekend compared with the weekday. The RI also decreases from regular to 343 holiday periods for the 3 age groups from 5 to 20 years, with the highest reduction observed in the age group 344 10-15 years. When comparing regular to holiday periods, we observe an increase in the RI for participants 345 aged 65 to 80 years, though the RI variability for this age group is considerably high. The result of the random forest analysis is shown in Figure S5 SA3: gender yields the lowest mean decrease 356 in accuracy, so it is removed. Significant predictors for both mean and over-dispersion parameters are further 357 selected using the likelihood ratio test. Accordingly, the mean and over-dispersion regressions in the final 358 10 . 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 October 2, 2020. 

The relative change in R 0 , RRI and the ratio of transmission rates are used to further compare the epidemio-382 logical differences between contact matrices from the two surveys. The relative changes in R 0 are presented in 383 

. 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 October 2, 2020. The association rules revealed that contacts of less than 15 minutes with non-household members usually do 407 not involve skin-to-skin touching. This finding is in line with the results of the 2006 survey [28] . To investigate 408 the contact profiles of participants, we performed clustering analysis. Our clustering results are comparable 409 to the results in [3] , in which a two-step clustering approach was applied to contact data from eight European 410 countries. Specifically, we endorsed the "school profile", "professional profile", and "leisure profile" from [3] , 411 with more contacts during leisure activities during weekends.

Demographic factors, including age, household and province of residence, have significant effects on the num-413 ber of contacts, as do the temporal factors, e.g., weekdays vs weekend days or regular terms vs holiday periods 414 [9, 10, 16, 22, 40] . It is noted that the interaction between people aged 60+ years and young children/teenagers 415 is significantly higher during holidays and weekends compared to regular-weekdays. For people living in an 416 elderly/nursing home, however, almost no contacts with young children/teenagers are reported. Using public 417 transportation is associated with a higher number of contacts in total. Our analysis also showed that those 418 who reported to feel ill had fewer contacts than those who reported to be healthy [6, 10, 15, 16, 17] . This 419 also holds for participants reporting health problems such as anxiety or those experiencing problems in daily 420 activities.

There is evidence, at least among school-aged children, that contact patterns are assortative with respect to 422 both age and gender. While an assortative mixing pattern with respect to age is still observed in adults, 423 albeit with lower contact rates, an assortative mixing pattern with respect to gender disappears in people 424 aged 30+ years. This analysis was also performed in [41] , where a hierarchical Bayesian model was used to 425 infer age-specific contact rates between genders. In contrast to [41] , we did not find significant differences in 426 infection risk between males and females. There are some reasons that may explain this difference. First, 427 we aggregated the age of participants in 20 age classes instead of using continuous age, which can incur an 428 inevitable loss of detail. Second, the dispersion parameter in our model was assumed to be age-dependent, 429 while it was treated as a nuisance parameter in [41] to avoid computation challenges. In addition, we used 430 diary weights in contact modelling to account for under-/over-sampling over the age of participants, while 431 weights were not taken into account in the model of [41] . 432 We found that the number of contacts was lower on weekends than on weekdays and during holidays com-433 pared to regular periods. We find a 30% (BCI:[17; 37%) reduction in R 0 for weekends versus weekdays or a 434 29% (BCI:[14; 40%]) reduction in R 0 for holidays versus regular periods. This result is consistent with the 435 results of other studies [8, 19, 40, 42, 43] . However, computing the age-specific relative incidence showed that 436 this reduction is due to the younger age classes, both during weekends and during holidays. Additionally, 437 12 . 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 October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204891 doi: medRxiv preprint the age-specific relative incidence showed that during holidays, there is a more complex change than during 438 regular weekends: while younger people have a lower relative incidence, people older than 60 years have an 439 increased relative incidence during holidays. 

In this study, we found that contact patterns remained fairly constant over 4-5 years. Additionally, within 455 each microscopic time period, no substantial changes in the spread of infection, measured by the relative basic 456 reproduction number and age-specific incidences, were observed ( Figure 7) . After taking into account multiple 457 testings, the pair-wise comparison of contact rates over time present only few significant differences during 458 holiday weekdays, mostly for people aged 50+ years. While the comparison of only two observational periods 459 about five years apart can be considered a limitation, to the best of our knowledge, this is the first study, that 460 investigates empirically whether contact rates remain stable, in the absence of major shocks to risk perception 461 (as we expect to observe in the SARS-CoV-2 pandemic emergence year 2020) and demography. Hence our 462 results suggest that stable social mixing patterns can be assumed over a time span of 5 years when no major 463 shocks to risk perceptions or demography occur. The datasets analysed during the current study are available on Zenodo [45, 46] . 468 Competing interests 469 The authors declare that they have no competing interests.

Author's contributions 471 NH and PB designed and coordinated the survey. NH conceived the study and laid out a paper structure. LW 472 and KVK performed data cleaning, TVH and YWK conducted the data analyses and drafted the manuscript 473 13 . 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 October 2, 2020. . https://doi.org/10.1101/2020.09. 30.20204891 doi: medRxiv preprint in consultation with all the other authors. NH, PC and KVK made substantial revisions to the manuscript. 474

All authors contributed to the final version of the manuscript. All authors approved the final manuscript and 475 agreed with its submission to the journal. 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 October 2, 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.

The copyright holder for this preprint this version posted October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204891 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.

The copyright holder for this preprint this version posted October 2, 2020. 19 . 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 October 2, 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.

The copyright holder for this preprint this version posted October 2, 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.

The copyright holder for this preprint this version posted October 2, 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 October 2, 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 October 2, 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 October 2, 2020. Table 1 : List of the common determinants selected from the two social contact surveys. Variable Categories Age

[0,5), [5, 10) , [10,15), [15, 20) , [20, 25) , [25, 30) , [30, 35) , [35, 40) , [40, 45) , [45, 50) 

. 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 October 2, 2020. . https://doi.org/10.1101/2020.09.30.20204891 doi: medRxiv preprint

Infectious Diseases of Humans: Dynamics and Control

Using data on social contacts to estimate age-specific trans-490 mission parameters for respiratory-spread infectious agents

Contact profiles in eight European countries and implications for 493 modelling the spread of airborne infectious diseases

Estimating 495 infectious disease parameters from data on social contacts and serological status

A nice day for an infection? Weather 498 conditions and social contact patterns relevant to influenza transmission

The impact of illness on social networks: 500 implications for transmission and control of influenza

Contact, travel, and transmission: The impact of winter 503 holidays on Influenza dynamics in the United States

Social contact patterns in Vietnam and implications for the control 541 of infectious diseases

Social mixing patterns within a south african township community: implications for res-544 piratory disease transmission and control

The French connection: The first large population-based contact 547 survey in France relevant for the spread of infectious diseases

Age-and sex-specific social contact patterns and incidence of 550 mycobacterium tuberculosis infection

Struc-552 tural differences in mixing behavior informing the role of asymptomatic infection and testing symptom 553 heritability

Projecting social contact matrices in 152 countries using contact surveys 555 and demographic data

Data-driven 557 model for the assessment of mycobacterium tuberculosis transmission in evolving demographic structures

Projecting social contact matrices to different demographic 560 structures

Mining social mixing 562 patterns for infectious disease models based on a two-day population survey in Belgium

Animal Ownership and Touching Enrich the Context of Social Contacts Relevant to the 566 Spread of Human Infectious Diseases

Random forests

Finding groups in Data: Cluster analysis extended Rousseeuw et. R Package. version 570 2

Instructions on how to use the gamlss package in r

Accompanying documentation in the current GAMLSS help files

Design and analysis of social contact surveys relevant for the spread of infectious 577 diseases

An Introduction to the Bootstrap

Networks of sexual contacts: implications for the pattern of 580 spread of HIV

Measures of disassortativeness and their ap-582 plication to directly transmitted infections

On the definition and the computation of the basic repro-585 duction ratio R 0 in models for infectious diseases in heterogeneous populations

Social contact patterns of school-age children in taiwan: comparison of the term 588 time and holiday periods

Efficient estimation of age-specific social contact rates 590 between men and women

School closures and student contact patterns

A modeling study of school closure to 594 reduce influenza transmission: A case study of an influenza A (H1N1) outbreak in a private thai school

Household members do not contact each other at random: implications for infectious 598 disease modelling

Social Contact Data for 600 Belgium in

RWD [25,30):RWD [20,25):RWD [15,20):RWD [10,15):RWD [5,10):RWD [0,5):RWD*

(n = 1,705, excluding 2 cases with missing information on the survey date). RWD, HWD, RWK and HWK stand for regular weekdays, holiday weekdays, regular weekends and holiday weekends, respectively. Stars (*) indicate reference groups for covariates with more than 2 categories