key: cord-0326492-zmmcxcyj
authors: Hoang, T. V.; Willem, L.; Coletti, P.; Van Kerckhove, K.; Minnen, J.; Beutels, P.; Hens, N.
title: Exploring human mixing patterns based on time use and social contact data and their implications for infectious disease transmission models
date: 2022-01-28
journal: nan
DOI: 10.1101/2022.01.25.22269385
sha: aa1b70de67fa18a140420c17326b775693c3fde2
doc_id: 326492
cord_uid: zmmcxcyj

Background: The increasing availability of data on social contact patterns and time use provides invaluable information for studying the transmission dynamics of infectious diseases. Social contact data provide information on the interaction of people in a population whereas the value of time use data lies in the quantification of exposure patterns. Both have been used as proxies for transmission risks within a population and the combination of both sources has led to investigating which kind of social encounters are most relevant to describe transmission risk. Methods: We used social contact and time use data from 1707 participants from a survey conducted in Flanders, Belgium in 2010-2011. We calculated weighted exposure time and social contact matrices to analyze age- and gender-specific mixing patterns and to quantify behavioral changes by distance from home. We compared the value of both data sources, individually and combined, for explaining seroprevalence and incidence data on parvovirus-B19, Varicella-Zoster virus (VZV), and influenza-like illnesses (ILI), respectively. Results: Assortative mixing and inter-generational interaction are more pronounced in the exposure matrix due to the high proportion of time spent at home. This pattern is less pronounced in the social contact matrix, which is more impacted by the reported contacts at school and work. The average number of contacts declined with distance, however on the individual-level, we observed an increase in the number of contacts and the transmission potential by distance when travelling. We found that both social contact data and time use data provide a good match with the seroprevalence and incidence data at hand. When comparing the use of different combinations of both data sources, we found that the social contact matrix based on close contacts of at least 4 hours appeared to be the best proxy for parvovirus-B19 transmission. Social contacts and exposure time were both on their own able to explain VZV seroprevalence data though combining both scored best. Compared with the contact approach, the time use approach provided a better fit to the ILI incidence data. Conclusions: Our work emphasizes the common and complementary value of time use and social contact data for analyzing mixing behavior and infectious disease transmission. We derived spatial, temporal, age-, gender- and distance-specific mixing patterns, which are informative for future modeling studies.

We linked social contact and time use data to obtain for each social contact the time spent at the reported 142 location and the distance from home. Participants were able to report multiple distances for one location type 143 over different time-slots in the time use part. For example, one could report "other" at 2 km and 10 km from 144 home, for shopping in the morning and sport activities in the evening. As such, we applied a probabilistic 145 link procedure between a social contact and the distance from home, based on the relative time spent at 146 each distance per location type. The interaction between travel and social contact patterns has been studied 147 unconditionally and conditionally upon presence at each distance, i.e., the former presents the population 148 average and the latter is in line with the individual-level perspective. To elaborate on social contact dispersal, 149 we classified participants by age using a cutoff of 18 years (child and adult), gender and type of day (regular 150 weekday, weekend, holiday). We excluded the contacts with missing distance (±7%) to calculate the weighted 151 average number of contacts by distance and by age, gender and day type. The matrix M dt , representing the mean number of contacts at distance d during one day of type t, can be estimated by the following expression:

where P i is the number of participants in age class i, w p the weight for participant p and y dt ijp the reported number of contacts made by participant p of age class i with someone of age class j at distance d during one day of type t. The social contact matrix c ij , representing the per capita daily contact rate between age classes, was calculated as

with N j the population size in age class j, obtained from census data.

The next generation matrix G with elements g ij indicates the average number of secondary infections in age class i through the introduction of a single infectious individual of age class j into a fully susceptible population [31] . Assuming a rectangular population age-distribution, the next generation matrix at distance d during one day of type t is defined by:

with N the population size, D the mean duration of infectiousness, L the life expectancy, C dt the contact 156 matrix at distance d during one day of type t and q the proportionality factor. The basic reproduction number 157 R 0 can be calculated as the dominant eigenvalue of the next generation matrix. To enable the comparison of 158 transmission dynamics by distance, we needed to specify the disease-specific q parameter of equation (3). As 159 such, we constrained the average R 0 for regular weekdays using all reported contacts from the bootstrapped 160 and imputed data sets to 2. 161 162 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 January 28, 2022. ;

We calculated the age-specific exposure times based on time use data following the Proportionate Mixing 164 Assumption (PTM) as previously used by [10] . As such, for one single location and time slot of the survey 165 day, we calculated the exposure time of a participant to the other participants proportionally to their relative 166 participation in that location. We used 17 categories; 0-2 years, 3-5 years, 6-11 years, 12-17 years, 18-25 years, 167 5-year age categories between 25 and 80 years of age, and a closing category of 80-90 years of age. Participants 168 had to report their household members in terms of age, gender and whether they were present at home during 169 the survey day. This allowed us to compute the time of exposure between members of the same household, 170 which is formalized in the matrix H.

For locations other than home, the exposure time between people in age group i and j at specific location l and time slot s, t ls ij , can be computed under the PTM assumption as follows:

where k ls i and k ls j are the number of people present at location l during time slot s in age group i and j, respectively, k ls . is the sum over all age classes at location l during time slot s and d s is the duration of each time slot s in hours. From (4), we can compute the time of exposure between people in age group i and in age group j, referred to as matrix T, as follows:

The sum of T and H determines the overall exposure time matrix E, with elements e ij :

The response matrix E contains non-negative quantities that are considered to follow a mixed discrete-173 continuous distribution, comprising value Y = 0 with probability p 0 and value Y = Y 1 ∈ (0, ∞) with probability 174

(1-p 0 The "suitable contact" approach assumes that not all social contacts are informative for disease transmission and that long duration and more intimate contacts are more likely to be relevant. To construct suitable contact matrices, we followed the procedure of De Cao et al. [11] , which considered contacts and exposure between age classes i and j (c ij and e ij , respectively). Let u ij be a random variable representing the number 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. (which was not certified by peer review)

The copyright holder for this preprint this version posted January 28, 2022. ; of suitable contacts between age group i and j. Then the expected number of suitable contacts is given by the product of the average number of contacts c ij and the proportion of these contacts that are relevant for transmission 1 − exp(−e ij /c ij ) where exp(−e ij /c ij ) is the Poisson probability that a contact is not suitable.

Therefore the probability of infection β i,j is given by

where q 1 is a constant disease-specific transmission coefficient and

with q 2 the fraction of total exposure time between age groups that is relevant for transmission.

Fitting mixing matrices to serological data for parvovirus-B19 and VZV

The social contact and the time use approach rely on the assumption that infectious disease transmission 185 between people in different age categories is proportional to their number and duration of their physical 186 encounters, respectively. To complement the age-specific social contact and exposure time matrices within a 187 transmission model framework, we estimated q 1 and q 2 from Equation 7 We estimated the model parameters by minimizing the sum of squared differences between the observed ILI 201 incidence rate and the predicted incidence rates. The following parameters were estimated: age-specific and 202

proportionality factors q i based on the social contact hypothesis [2] and a scaling factor. The scaling factor is 203 used to align the predicted incidence rates to the observed incidence rates, which accounts for those individuals 204 with influenza who do not seek medical attention (the consultation rate) [34, 35, 36 ].

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. (which was not certified by peer review)

The copyright holder for this preprint this version posted January 28, 2022. ;

Time use patterns 207 The final sample size for the analysis is 1707 participants, as reported in [19] , though 49 participants left the 208 time use part entirely blank. People older than 45 years of age accounted for the highest proportion of missing 209 data for at least one time slot (Table 1) . People spent on average about two thirds of their day at home.

The elderly over 65 years of age and children less than 6 years old reported the highest proportion of time 211 spent at home. For the active population, we observed that time at work is higher for males than for females, 212 hence the opposite is observed for time spent at home. Temporal factors such as day of the week and holiday 213 periods had a large impact on time use.

Health status was also linked to substantial changes in participants' time use. Participants did report an 215 increase of 5 hours at home when feeling ill and almost no time at work. However, the time spent at other 216 locations is still substantial when people reported feeling ill. Adults between 25 to 65 years of age living with 217 children under 13 years of age reported more time spent at home compared to those without children.

We analysed the reported time at work more in detail and observed that the average time is similar for males 220 and females until the age of 30 years (Additional file 2 Figure S1 ). Differences emerge from the age of 30 221 years, in which females reported on average less time at work compared to males in the same age group. We 222 observed the highest differences in age groups [40, 45) , [50, 55) and [55, 60) in which males reported 10% more 223 time at work compared to females. The reported time at work declined after 65 years of age for both genders. 224

The time spent at "other" places was similar between males and females in general, except for the age cat-225 egories [6,12) and [18, 25) in which males reported 6% more compared to females. The reported time for 226 transport was similar for males and females.

We analysed the time use patterns with a divergence-based regression model with multinomial logit link 229 function including gender, age, day type and period. Table 2 shows all parameter estimates and 95% confidence 230 intervals. After controlling for age and temporal factors, time at work reported by males is significantly higher 231 compared to females. In reverse, females reported more time at home than males. The gender-specific 232 difference in time spent at school was not statistically significant. Age had a significant effect on presence at 233 home, school and work but not on the reported time at location category "other". As expected, temporal 234 factors played a crucial role in the time use patterns, particularly on the presence at work and school. We also 235 observed a clear increase in the reported time in the "other" category during weekends vs. regular weekdays. 236

This denotes compensation behavior of people not at work or at school, which needs to be considered when 237 modeling weekend days.

Social contacts patterns taking into account time use 239 We analysed the number of reported social contacts with a zero-inflated negative binomial model and observed 240 that the overall number of contacts was inversely associated with the time spent at home, and positively 241 associated with the time spent at school or work (Table 3 ). The number of contacts at school or work tends to 242 increase with the time spent in these settings. We observed a gender effect, implying that males tend to have 243 on average fewer contacts compared to females per time-unit. However, there was no statistical difference in 244 8 . 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 January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269385 doi: medRxiv preprint the number of contacts at different locations between males and females. With respect to age, the highest 245 overall number of contacts were observed among children up to 18 years of age and among people in the 246 working age (25-65 years of age). The age effect was not observed for the social contacts at home and school, 247 but observed for the social contacts at other places.

Household size was positively associated with the total number of contacts, contacts at home and other places, 249 however household size showed no significant effect on the social contacts at school and at work. The reported 250 class size did not have a clear effect on the number of school contacts. As expected, we observed an explicit 251 link between temporal factors and the number of school and work contacts. To explore social contact dispersal, we analyzed the social contact data by distance in combination with age, 255 gender and type of day (Figure 1 ). At population level, i.e. unconditional upon whether or not people travel, 256 we observed most contacts for children (0-18 years of age) in the category 0-9 km from home during week-257 days and very few contacts beyond 10 km from home. This pattern was the same for boys or girls. During 258 weekends, children reported a decline in the number of contacts by distance, but less so for girls. For adults 259 (18+ years of age), we observed an increase by distance on weekdays up to 10-75 km from home. On average, 260 people reported almost no contacts 75+ km from home. Females reported more contacts up to 9 km from 261 home compared to males, though the opposite holds for the category 10-74 km. For weekends, we observed a 262 decline in the number of contacts beyond 10 km from home, both for males and females. 10-74 km from home equals the number of contacts close to home. During weekends, the reported number of 268 contacts for girls increased with distance though the social contact behavior for boys seemed indifferent with 269 distance up to 74 km. For adults, the number of social contacts during weekdays at 2-9 km and 10-74 km 270 from home was two and three times the number of reported contacts at home, respectively. Also for the last 271 distance category (75+ km from home), we observed an increase compared to the contacts at home but with 272 large uncertainty. If adults leave home during weekends, the reported number of contacts by adults seems 273 indifferent to the distance up to 74 km. Only at 75+ km from home, females reported fewer social contacts, 274 compared to males. In conclusion, the number of contacts decreased by distance at population level, though 275 if people made the effort to go somewhere, they made it count in terms of social encounters. The social contact dispersal as illustrated in Figure 1 induces distinct transmission potential by distance. 279 We calculated social contact matrices by distance for child/adult interactions for each bootstrapped data set 280 (e.g., Additional file 4 Figure S4 and S5) and corresponding R 0 values. We calibrated the disease-specific 281 proportionally factor q for a flu-like disease with median R 0 = 2 with the full set of social contact data from 282 regular weekdays. The successive R 0 values using distance-specific contact data should be interpreted as 283 9 . 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 January 28, 2022. ; the relative transmission potential and are presented in Figure 2 . On the population level, the transmission 284 potential slightly increased until 74 km from home during regular weekdays. This can be explained by the 285 strong assortative mixing 2-9 km from home and the work-related mixing between adults 10-74 km from home 286 ( Additional file 4 Figure S4 ). The overall reduction in the transmission potential during weekends, can be 287 explained by the fewer number of contacts for the distance category 10-74 km from home. Social contacts at 288 75+ km from home had almost no impact on the unconditional transmission potential.

The transmission potential conditional upon presence showed a clear increase with distance from home during 291 regular weekdays ( Figure 2 ). This effect was similar for weekends, however with a decrease of the estimated 292 R 0 for the last distance category. In general, the estimated transmission potential increased by distance from 293 an individual-based perspective.

Exposure time and social contact matrices 295 We used a zero-adjusted log-normal model on the time use data to estimate the age-specific exposure matrices. 296

The resulting exposure matrix reflected contributions from exposure at home, school, work, and other locations 297 ( Figure 3 ). The main diagonal indicates that people tend to spend time with people of similar age. The two 298 sub-diagonals represent the mixing pattern between generations. We also calculated the corresponding social 299 contact matrix representing the weighted average number of contacts by age. Both matrices showed strong 300 assortative mixing by age, especially for young children and teenagers. Especially the exposure matrix showed 301 strong interaction among family members, as a result of the reported time at home. This pattern is very clear 302 in the location-specific exposure matrix at home ( Additional file 3 Figure S1 ).

Time reported at home constituted on average up to ±66% of the total time per day, while contacts at home 304 only accounted for ±18% of the total number of contacts. In contrast, the impact of employment is higher 305 for the contact matrix compared to the exposure matrix (±40% of the total number of contacts though only 306 ±9% of total time per day). We observed more pronounced assortative mixing pattern in the same-gender 307 matrices for children and teenagers (Figure 4 ), though more different-gender exposure time for adults older 308 than 50 years of age. In addition, the interaction between mothers and daughters seemed more pronounced 309 compared to other child-adult interactions.

. 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 January 28, 2022. ;

We tested the value of exposure and social contact matrices as a proxy for effective contacts governing trans-312 mission by their value in MSIRWb and MSIR models for parvovirus-B19 and VZV, respectively. Model 313 estimations were compared to serological data and model selection was based on the AIC criterion. The over-314 all time use matrix implied a better fit for both parvovirus-B19 and VZV compared to total social contact 315 matrix. When exploring different levels of intimacy, location and duration, the social contact matrix based on 316 close contacts lasting more than 4 hours, provided the best proxy for the transmission dynamics of parvovirus-317 B19 (Additional file 5 Table S2 and S3). For VZV, the social contact matrix based on close contacts of at 318 least 1 hour did improve the contact approach, though the AIC was still higher compared to the time use 319 approach.

The suitable contact approach, which accounts for both the number of social contacts and exposure time, 322 scored slightly better for parvovirus-B19 compared to the contact approach, but not better in comparison 323 to the time use approach. For VZV, the application of suitable contact estimates provided the overall best 324 fit with the lowest AIC. The estimated proportionality factor q 2 of exposure data with respect to the overall 325 model prediction is much higher for VZV (0.94) than for parvovirus-B19 (0.13), which implied a higher number 326 of suitable contacts for transmission of VZV. Our best-fitting models estimated a reproduction number of 1.9 327 [1.7; 2.1] and 7.8 [6.8; 8.5] for parvovirus-B19 and VZV, respectively. The estimated prevalence and force 328 of infection by the 3 approaches were quite similar ( Figure 5 ). We did observe differences in the predicted 329 age-specific relative incidence by the time use and social contact approaches ( Figure 6 ). The highest relative 330 incidence obtained by the social contact approach was among children in the age group [12,18), while the 331 highest relative incidence obtained by the time use approach is among children of [6-12) years. The latter also 332 predicts a relatively higher incidence in adults between 35 and 50 years of age.

Fitting social contact and exposure time matrices to ILI incidence data 334 Social contact and exposure time matrices were used to compute transmission rates in the dynamic, differential-335 equation SEIR model for ILI incidence data. The model comparison was based on the least square score, a 336 direct measure of goodness of fit with smaller values indicating a better fit to the ILI incidence data. The 337 intimacy and duration of contacts seemed to matter for the ILI transmission modelling, such that physical 338 contacts and contacts lasting at least 1 hour provided most information regarding transmission dynamics. 339

The exposure time matrices provided a better fit for ILI compared to the total contact matrix (Additional 340 file 5 Table S4 ). R 0 estimated from the Based on the best scoring model, we estimated the R 0 to be 1.43 341 with scaling factor of 0.266. The combination of contact and exposure time matrices did not improve the fit. 342 Figure S1 (Additional file 5) shows the fit of different matrices to ILI incidence rate. The estimated incidence 343 rates from models using suitable contact matrices and exposure time matrices are almost overlapping, with 344 only a slight difference at the seasonal peaks of transmission. 345 Figure 7 shows the fit of the models based on the contact and exposure time matrices for the total population 346 and different age groups. For each (sub)model, we present the least square value, the observed number and 347 modeled based estimates of ILI cases, their ratio, and the mean absolute error (MAE). For the total population, 348 the estimated ILI cases captures quite well the weekly observed number of ILI cases. The contact matrices and 349 exposure time matrices provided quite similar results for the total population and most of age groups, except 350 11 . 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 January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269385 doi: medRxiv preprint for the age group of 65 years and older. More precisely, the use of exposure time captured the observed ILI 351 curve for this age group better compared to models informed by all social contact data. In general, the quality 352 of the fit does not differ substantially between age groups. The models tend to overestimate the number of 353 ILI cases for children and teenagers aged 0-19 years and adults aged 20-64.

Discussion and conclusions 355 We analysed time use and social contact data and compared their use as proxy for effective contacts governing 356 disease transmission for disease transmitted through the respiratory route. Our data set is unique since it 357 provides both time use and social contact data from the same participants, avoiding possible differences due 358 to sample biases. In our analysis we identified the main drivers in shaping everyday time use and linked this 359 info to social contact patterns.

The reported time use patterns in Belgium are quite similar to the published patterns for Italy [37] , but dif-362 ferent from the results from Zimbabwe [38] . In Zimbabwe, the working-age participants and children less than 363 6 years old reported much less time at work and school, respectively, compared with participants of the same 364 age group in Belgium and Italy. We found that males spent on average ±2 hours more at work than females, 365 which is in line with previous work [39, 40] . We also found that both males and females living with children 366 spend more time at home than people living without children, which is consistent with what was found in [39] . 367

The expected temporal patterns were observed, with more time spent at "other" locations during weekends 368 and holiday periods. The time spent at home seemed not to be affected by the type of day. Our gender-specific 369 analysis of the time use data indicated that participants were prone to spend more time with the same gender 370 when they are young, and more time with the other gender when they are older. This result differs from the 371 observed gender-specific contact rates as reported in [19] , where the assortativity is reported to be higher for 372 same-gender contacts.

Power law dispersal has been useful to model disease counts [13, 14, 15] , though questions remained whether 375 this also holds for social contact behavior, the driver of transmission dynamics [2] . Danon et al. [16] showed 376 a relation between clustering and distance from home, with high clustering within two miles, dominated by 377 home contacts, but the highest value of clustering occurring at a distance 50 miles or more from home. The 378 authors hypothesize that this might be due to differences in the purpose behind contacts made at various 379 distances.

In our study, we observed an increase in the number of contacts by distance during weekdays until an age-381 specific distance-threshold. A large-scale study in Taiwan [41] reported that 52.7% of contacts took place at 382 a distance less than 1 km from home, 29.2% at a distance 1-9 km from home, 14.6% at a distance 10-49 km 383 from home. In our study, this pattern was clearly age-specific. Half of our reported contacts during regular 384 weekdays for children between [0-18) years of age took place less than 1 km from home. During weekends, we 385 observed more contacts at +10 km from home. In general, we observed that the average number of contacts 386 decreased by distance, while the individual-level pattern is reverse. As such, social contact patterns condi-387 tional upon presence at each distance provided useful info to inform spatial transmission models. A study in 388

China quantified the distances from home based on the latitude and longitude of each reported contact and 389 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 January 28, 2022.

observed an increase in assortative mixing when contacts were made further from home [42] . We observed 390 similar patterns with the construction of conditional social contact matrices by distance. A study in the 391 United Kingdom [43] requested for infants to report the maximum distance travelled from home, but, to date, 392 did not report results thereof. Other survey designs (e.g., [38] ) included the distance between home and work 393 but did not report results related to this information, yet.

The age-specific social contact and time use patterns followed a similar trend of assortativeness, which stresses 396 once more the tendency for people to interact with people of a similar age. In addition to that, strong mixing 397 among generations (parent-child) was present in both the social contact and time use matrices. The inter-398 generation mixing is mostly observed at home, and was more pronounced in the exposure time matrix than 399 in the contact matrix. With our unified survey, we can confirm the contrasting effect of the relative small 400 number of social contacts at home compared to the large amount of time spent at home, as observed in [10, 38] . 401 402 Social contact matrices provide useful data to estimate disease transmission dynamics in terms of the transmis-403 sion dynamics and relative incidence [1, 2] . As such, we estimated the R 0 for each distance conditional upon 404 presence, and observed a clear increase by distance. This reflects an increasing transmission potential by dis-405 tance from the individual-level perspective. The latter is of interest for meta-population and individual-based 406 models, were individuals join other sub-populations at distance. We found that a constant (or decreased) 407 social mixing behavior by distance conditional upon presence is likely to underestimate the transmission po-408 tential. Some individual-based models handled this by the use of location-specific mixing patterns irrespective 409 of distance from home [44] .

We compared the value of different social contact features (duration, physical/non-physical, etc.) to inform 412 transmission models for parvovirus-B19 and VZV. By scoring the model-based prevalence with Belgian sero-413 logical data, we found that physical contacts provided the best proxy for both parvovirus-B19 and VZV. In 414 terms of contact duration, the best model fit was obtained with physical contacts of long duration (more than 415 4 hours) for parvovirus-B19 and (more than 1 hour) for VZV. Goeyvaerts et al. [3] reported the best fit to 416 VZV with physical contacts of at least 15 minutes; this result is also in line with the study of [4] . However, 417 the previous studies did not analysed all combinations in terms of contact duration and physical/non-physical 418 contacts, which explains the new "best estimate" in our study.

We also compared the results of the contact approach, the time use approach and the suitable contact approach 421 in fitting serological data of VZV and parvovirus-B19. In the case of VZV, the suitable contact approach pro-422 vided the best fit, while for parvovirus-B19, the time use approach gave the best fit. Our results are consistent 423 with the findings of De Cao et al. [11] , although we observed much higher parameter estimates for q 2 (0.13 vs 424 0.001 for parvovirus-B19 and 0.94 vs 0.37 for VZV), but the confidence intervals of these parameter estimates 425 were overlapping.

We also tested the value of contact matrices and exposure time matrices in fitting the dynamic transmission 428 model to weekly ILI incidence data in the season 2010-2011 in Belgium. Exposure time matrices provided 429 a better fit to ILI than overall contact matrices and the integration of these two types of matrices did not 430 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 January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269385 doi: medRxiv preprint improve the fit to the data. Within the social contact approach, physical contacts provided a better proxy for 431 the risks of influenza transmission than non-physical contacts. We found that R 0 of the best model is 1.43, 432 this result is consistent with a systematic review of estimates of R 0 for different types of influenza [45] and 433 a study in UK [35] that also used contact matrices to fit to ILI incidence data. However, this value is lower 434 than the estimates in [34], in which physical contact matrices from the Belgian POLYMOD data were used 435 to fit to ILI data over multiple influenza season from 2003 to 2009. The difference in R 0 can be explained by 436 the relatively low number of ILI cases reported in the 2010-2011 season we used in this study. In addition, 437 note that small differences in model parametrization entail substantial differences between the estimates of 438 R 0 , which was also mentioned in [34] .

In our study, we combined information from time use and social contact data to gather information on human 441 mixing patterns. One of the main advantages with respect to previous work is that both sources of informa-442 tion came from the same survey. To keep participants' burden as low as possible, time use information was 443 collected with rather large time slots and participants were asked to fill in only one location for each time 444 slot. However, the comparison with more refined time use surveys performed in Flanders [39, 40] confirms 445 that we were able to well characterize time use patterns at an aggregated level. Therefore, we expect that this 446 limitation did not substantially affect our results. Furthermore, our analysis based on data from the same survey is in line with studies that merged information 453 from different surveys [10, 11] . This indicates that complementing social contact with independent time use 454 data is a viable choice for the analyses presented here. 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 January 28, 2022. ; 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 January 28, 2022. ; https://doi.org/10.1101 https://doi.org/10. /2022 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 January 28, 2022. ; https://doi.org/10. 1101 /2022 [33] Goeyvaerts, N., Hens, N., Aerts, M., Beutels, P.: Model structure analysis to estimate basic immunological 568 processes and maternal risk for parvovirus B19. Biostatistics 12(2), 283-302 (2010)

[34] Goeyvaerts, N., Willem, L., Van Kerckhove, K., Vandendijck, Y., Hanquet, G., Beutels, P., Hens, N.: 570 Estimating dynamic transmission model parameters for seasonal influenza by fitting to age and season-571 specific influenza-like illness incidence. Epidemics 13, 1-9 (2015) 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 January 28, 2022. . 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 January 28, 2022. ; q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q 0.0 0.5 1.0 1.5 2.0 2.5 3.0

Distance from home Estimated R 0 Total 0−1km 2−9km 10−74km +75km

Regular weekday Weekend q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Figure 2 : Estimated basic reproduction number R 0 by distance, calculated as the leading eigenvector of the distance-specific social contact matrix and calibrated so the median R 0 for regular weekdays equals 2 based on population-based (unconditional) contact matrices. For each distance category, we used the unconditional and conditional social contact patterns. The median is represented by the horizontal line in the box (75% interval), the whiskers denote the 95% confidence intervals and the dots are outliers.

. 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 January 28, 2022. 

. 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 January 28, 2022. age seroprevalence q q q q q q q q q q q q q q q q q q q q q q q q qq q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q Combined approach TU approach Contact approach 

. 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 January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269385 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 January 28, 2022. . 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 January 28, 2022. ; Table 2 : Parameter estimates and 95% confidence intervals of the divergence-based regression analyses for location-specific time use patterns. The asterisks (*) denote the confidence intervals which do not include zero. 

. 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 January 28, 2022. ; https://doi.org/10. 1101 /2022 Additional Files 607 Additional file 1 Missing data exploration, descriptive analysis results and the list of variables 608 included in the imputation model.

Additional file 2 Time use at different locations by gender and age.

Additional file 3 The mean daily exposure time among age groups across locations.

Additional file 4 Number of contacts and temporal social contact matrices by distance.

Additional file 5 Fitting results of social contact and exposure matrices to parvovirus-B19 and 613 VZV serological data and ILI incidence data.

. 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 January 28, 2022. ; https://doi.org/10. 1101 /2022 

18,25) [25,30) [30,35) [35,40) [40,45) [45,50) [50,55) [55,60) [60,65) [65,70) [70,75) [75,80) [80,90) 3) [3,6) [6,12) [12,18) [18,25) [25,30) [30,35)

We greatly thank the Scientific Institute of Public health for their permission to use the ILI data.

References

Fitting social contact and exposure matrices to parvovirus-B19 and VZV serological data