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

Abstract

Background: Until recently, mathematical models of person to person infectious diseases transmission had to make assumptions on transmissions enabled by personal contacts by estimating the so-called WAIFW-matrix. In order to better inform such estimates, a population based contact survey has been carried out in Belgium over the period March-May 2006. In contrast to other European surveys conducted simultaneously, each respondent recorded contacts over two days. Special attention was given to holiday periods, and respondents with large numbers of professional contacts.

Methods: Participants kept a paper diary with information on their contacts over two different days. A contact was defined as a two-way conversation of at least three words in each others proximity. The contact information included the age of the contact, gender, location, duration, frequency, and whether or not touching was involved. For data analysis, we used association rules and classification trees. Weighted generalized estimating equations were used to analyze contact frequency while accounting for the correlation between contacts reported on the two different days. A contact surface, expressing the average number of contacts between persons of different ages was obtained by a bivariate smoothing approach and the relation to the so-called next-generation matrix was established.

Results: People mostly mixed with people of similar age, or with their offspring, their parents and their grandparents. By imputing professional contacts, the average number of daily contacts increased from 11.84 to 15.70. The number of reported contacts depended heavily on the household size, class size for children and number of professional contacts for adults. Adults living with children had on average 2 daily contacts more than adults living without children. In the holiday period, the daily contact frequency for children and adolescents decreased with about 19% while a similar observation is made for adults in the weekend. These findings can be used to estimate the impact of school closure.

Conclusion: We conducted a diary based contact survey in Belgium to gain insights in social interactions relevant to the spread of infectious diseases. The resulting contact patterns are useful to improve estimating crucial parameters for infectious disease transmission models.

Figure 1

Contact number densities per weekday. Histograms of log(number of contacts+1) per day in the week, distinguishing non-holiday (top row) from holiday (bottom row) periods.

Figure 2

Contact intensity distributions in- and outside households. Contact intensity distributions in- and outside households: close or non-close (left figure); duration (middle panel); frequency (right panel). Darker colors correspond to touching, longer duration and more frequent contacts.

Figure 3

Contact number densities for different contact characteristics. Boxplots of the number of (close) contacts in and outside households (left upper panel); the number of contacts per location (right upper panel); the number of contacts for the different frequencies (left lower panel) and duration (right lower panel).

Figure 4

Close or non-close contact classification tree. Classification tree for contacts, involving skin to skin touching. Variable codes can be found in Table 1.

Figure 5

Contact location classification tree. Classification tree for contact location. Variable codes can be found in Table 1.

Figure 6

Contact frequency classification tree. Classification tree for contact frequency. Variable codes can be found in Table 1.

Figure 7

Contact duration classification tree. Classification tree for contact duration. Variable codes can be found in Table 1.

Figure 8

Contact patterns and newly emerging infections. Contact Patterns (left column) and leading eigenvector (right column) for a newly emerging infection in Belgium based on close contact patterns in a non-holiday period (first row), holiday period (middle row) and during weekends (excluding weekends during holidays) (last row). Contact patterns overlaid with contours are plotted at the population level.

Figure 9

Living with or without children and average daily number of close contacts. Distribution of daily number of close contacts for adults living without and with children (left and right panel, respectively.).