We conducted this systematic review according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Appendix 1 Checklist 1) and Meta-Analyses of Observational Studies in Epidemiology (MOOSE) guidelines (Appendix 1 Checklist 2) in accordance with a published protocol (17). We identified publications describing social contact surveys conducted during January 1, 1997–August 5, 2018, through searches of PubMed, Embase, Global Health, and the Cochrane Database of Systematic Reviews (Appendix 1 Table 1). We searched reference lists from included publications by hand and contacted researchers with expertise in these surveys, particularly authors of a recent systematic review (18), to assist with identification of relevant publications.

Two authors (K.C.H. and A.L.H.) independently reviewed titles and then abstracts, in parallel, for relevance and included publications identified by either author for full-text review. These authors also reviewed full texts to determine which publications met inclusion criteria and then reviewed texts and supplemental materials to determine whether data on sex were recorded for participants and contacts. These authors contacted publication authors if it was unclear whether these data had been collected.

K.C.H. extracted data on methods from included surveys by using a piloted electronic form and gathered datasets from supplemental materials or a social contact data repository (https://www.socialcontactdata.org) if results were not reported in a format necessary for meta-analyses. When datasets were not publicly available, K.C.H contacted authors and asked them to share relevant results or data.

The review included cross-sectional surveys conducted to assess social contact patterns relevant to airborne disease transmission that recorded participant sex and contact sex. We included only surveys that recorded all contacts over the survey period; we excluded surveys that examined only a subset of participants’ contacts (e.g., only those within a workplace or with other participants). We also excluded surveys that included only participants or contacts of a single sex and, because of limited sources for translation, publications in languages other than English. When we identified >1 report for a single survey, we included the earliest source or most complete dataset and excluded other records.

We assessed each survey by using the Appraisal Tool for Cross-sectional Studies (AXIS tool). This tool evaluates survey design, reporting quality, and risk for bias (19).

We considered participation equitable by sex if each sex made up 45%–55% of the survey population. We adjusted numbers of participants for analyses of physical and location-based contacts to exclude participants who did not report this information.

We stratified participants and contacts by age as children (boys and girls) and adults (men and women). For most surveys, adults were defined as persons >15 years of age (1); in instances where aggregate age categories did not enable disaggregation at this cutoff point, we used the nearest possible value. We defined close contacts, including physical and nonphysical contacts, according to survey-specific definitions, typically by a conversation longer than a greeting or >3 words.

We defined sex-assortative mixing as like-with-like contacts according to sex (male or female), either within age groups (e.g., men-with-men) or between age groups (e.g., men-with-boys). We defined preferential mixing as more mixing with 1 sex/age group than another.

For each survey, we calculated the average number of contacts over a 24-hour period for each sex/age category of participants with each sex/age category of contacts. For surveys in which data were collected over a 48-hour period, we divided the number of contacts by 2. For surveys in which data were collected over a 72-hour period, we divided the number of contacts by 3. We compared the average number of contacts across sex and age groups by using the Mann–Whitney–Wilcoxon test.

We calculated the percentage of sex-assortative mixing with 95% Clopper-Pearson CIs as contacts with the same sex divided by total contacts. We assessed sex-assortative mixing in children’s contacts with children and adults and in adults’ contacts with children and adults. We also compared the proportion of sex-assortative mixing by contact location: contacts within the home and contacts outside the home and, among contacts outside the home, contacts at work (for adults), school (for children), and elsewhere. We assessed heterogeneity by using the I² statistic (20) and summarized findings across surveys by using the median and interquartile range (IQR).

We estimated the percentage of boys’, girls’, men’s and women’s adult contacts with men for subgroups based on survey setting characteristics (region, setting, and TB burden) and survey methods (sampling methods, reporting duration, age cutoff values for adults, and participation by sex). We excluded contact events for which the participant’s sex or age or the contact’s sex or age was missing. We made no adjustments for nonparticipation or nonsampling and used no weighting. We performed all analyses by using R version 3.2.2 (21).

Participation by children was considered equitable by sex in 15 (88%) of 17 surveys. In 2 (12%) surveys, participation by boys substantially exceeded that by girls; boys made up 56% and 57% of the population of each survey. Participation by adults was considered equitable by sex in 11 (55%) of 20 surveys. In 8 (40%) of 20 surveys, participation by women substantially exceeded that by men; women made up 56%–83% of the population of each survey. In 1 (5%) survey, participation by men substantially exceeded that by women; men made up 60% of the survey population.

The median number of contacts reported over a 24-hour period was 12.9 (IQR 9.3–15.9) for boys and 13.5 (IQR 9.5–15.9) for girls (Appendix 1 Table 5); the difference in numbers of contacts was not significant (p = 0.92). Approximately half of contacts reported by boys (median 53%, IQR 43%–55%) and girls (median 51%, IQR 45%–56%) were with other children.

Among contacts of children with other children, we found strong evidence of sex-assortative mixing reported by boys in 15 (88%) of 17 surveys and by girls in 15 (88%) of 17 surveys (Figure 2, panels A, C; Appendix 1 Table 6). The median percentage of sex-assortative mixing in contacts with children was 62% (IQR 59%–63%) for boys and 59% (IQR 59%–65%) for girls. Summary measures are not reported because of substantial heterogeneity between surveys (I² = 96.3% for boys, I² = 95.6% for girls).

Among contacts of children with adults, there was no evidence of sex-assortative mixing reported by boys and strong evidence reported by girls in 17 (100%) of 17 surveys (Figure 2, panel B, D, Appendix 1 Table 6). The median percentage of sex-assortative mixing was 42% (IQR 41%–43%) for boys and 61% (IQR 60%–63%) for girls. Boys reported preferential mixing with women in 15 (88%) of 17 surveys. Summary measures are not reported because of substantial heterogeneity between surveys (I² = 73.8% for boys, I² = 44.3% for girls).

Most contacts reported by children took place outside the home (median 65% [IQR 62%–72% for boys], median 67% [IQR 56%–73%] for girls) (Appendix 1 Table 7). The sex assortativity of children’s contacts outside the home was similar to that at home. Among contacts with children, boys and girls reported more sex-assortative mixing in contacts outside the home than at home in 6 (43%) of 14 surveys for boys and 5 (36%) of 14 surveys for girls (Figure 3, panels A, C; Appendix 1 Table 8). Among contacts with adults, boys reported no more sex-assortative mixing in adult contacts outside the home than at home in 14 (100%) of 14 (100%) surveys, and girls reported more sex-assortative mixing outside the home than at home in 6 (42%) of 14 surveys (Figure 3, panels B, D; Appendix 1 Table 8). Summary measures are not reported because of substantial heterogeneity between surveys (I² = 88.4% for boys, I² = 83.0% for girls).

Among contacts of children outside the home, ≈50% of contacts of boys and girls contacts (median 56% [IQR 39%–62%] for boys, median 55% [IQR 38%–63%] for girls) occurred at school (Appendix 1 Table 9). We found few differences in the sex assortativity of contacts at school compared with those at other locations outside the home (Appendix 1 Table 10, Figure 1). Summary measures are not reported because of substantial heterogeneity between surveys (I² = 84.7% for boys, I² = 74.1% for girls).

The median number of contacts reported over a 24-hour period was 11.1 (IQR 8.1–15.3) for men and 11.6 (IQR 7.8–14.3) for women (Appendix 1 Table 11); the differences were not significant (p = 0.88), and the total number of contacts reported by adults did not differ from the total number of contacts reported by children (p = 0.26). Most contacts reported by men (median 91% [IQR 88%–93%] and women (median 87% [IQR 83%–90%]) were with other adults, which was significantly more than the number of adult contacts reported by children (p = 0.01).

Among contacts of adults with children, there was strong evidence of sex-assortative mixing reported by men in 4 (20%) of 20 surveys and by women in 4 (20%) of 20 surveys (Figure 4, panels A, C; Appendix 1 Table 12). In 15 (75%) of 20 surveys, there was no major evidence of preferential mixing by sex reported by men or women in contacts with children. The median percentage of sex-assortative mixing was 53% (IQR 50%–57%) for men and 52% (IQR 50%–54%) for women. Summary measures are not reported because of substantial heterogeneity between surveys (I² = 76.3% for boys, I² = 81.6% for girls).

Among adult contacts with other adults, there was strong evidence of sex-assortative mixing reported by men in 16 (80%) of 20 surveys and by women in 19 (95%) of 20 surveys (Figure 4, panels B, D; Appendix 1 Table 12). The median percentage of sex-assortative mixing was 56% (IQR 54%–58%) for men and 59 (IQR 57%–63%) for women. Summary measures are not reported because of substantial heterogeneity between surveys (I² = 98.1% for men, I² = 97.0% for women).

Most contacts reported by adults took place outside the home (median 74%, IQR 62%–77% for men; median 70%, IQR 54%–76% for women) (Appendix 1 Table 13). Contacts of adults with children showed similar sex assortativity at home and outside the home (Figure 5, panels A, C; Appendix 1 Table 14). Among contacts of adults with adults, there was more sex-assortative mixing by men and women in contacts outside the home than in contacts within the home in 14 (93%) of 15 surveys (Figure 5, panel B, D; Appendix 1 Table 14). Summary measures are not reported because of substantial heterogeneity between surveys (I² = 63.1% for men, I² = 28.6% for women).

Among adult contacts outside the home, ≈33% of contacts of men and women (median 35% [IQR 28%–39%] for men, median 29% [IQR 26%–34%] for women) occurred at work (Appendix 1 Table 15). Because adults reported few contacts with children at work, CIs are wide for sex-assortative mixing estimates for men and women in most surveys (Appendix 1 Table 16, Figure 2, panels A, C). Men reported more sex-assortative mixing in contacts with other adults at work compared with contacts elsewhere outside the home in 12 (80%) of 15 surveys and elsewhere in 1 (7%) of 15 surveys (Appendix 1 Table 16, Figure 2, panels B, D). Women reported more sex-assortative mixing at work compared with contacts elsewhere outside the home in only 2 (13%) of 15 surveys and elsewhere in 1 (7%) of 15 surveys. Summary measures are not reported because of substantial heterogeneity between surveys (I² = 32.3% for men, I² = 87.0% for women).

Subgroup analyses did not show clear differences in the frequency of contact with men by survey setting or method. There was little variation in survey characteristics measured by the AXIS tool (Appendix 1 Table 17). Substantial heterogeneity remained in summary measures for subgroups examined (Appendix 1 Table 18).