Skip to main content
Nature Communications logoLink to Nature Communications
. 2025 Oct 30;16:9586. doi: 10.1038/s41467-025-64850-9

Characterizing social behavior relevant for infectious disease transmission in four low- and middle-income countries, 2021-2023

Kristin N Nelson 1,✉,#, Moses C Kiti 1,#, Machi Shiiba 1, Charfudin Sacoor 2, Azucena Bardají 2,3, Ivalda Macicame 4, Edgar Jamisse 2, Corssino Tchavana 2, Americo José 4, Nilzio Cavele 2, Herberth Maldonado 5, Claudia Jarquin 5, H María Ajsivinac 5, Venkata Raghava 6, Prasanna Samuel 6, Rajan Srinivasan 6, Momin Kazi 7, Raheel Allana 7, Sara S Kim 1, Pragati V Prasad 1, Dehao Chen 1, Carol Liu 1, Samuel M Jenness 1, Noureen Ahmed 8, Obianuju Aguolu 9, Maria A Sundaram 10, Inci Yildirim 11, Fauzia Malik 8, Alessia Melegaro 12, Benjamin A Lopman 1, Saad B Omer 8
PMCID: PMC12575605  PMID: 41168147

Abstract

Infectious diseases account for nearly half of all child mortality worldwide, with most of the burden concentrated in low and middle-income countries (LMIC). Person-to-person interactions, or ‘contacts’, facilitate the spread of respiratory and enteric pathogens. The number and nature of contacts likely vary across countries along with social and cultural norms, but few studies have compared behaviors across countries and none have done so with a focus on children. Here we present data from a population-based study conducted from 2021 to 2023 in Guatemala, India, Mozambique, and Pakistan. Across four countries, 5085 participants reported a total of 84,829 contacts across two days. Mean contact rates were highest among 10- to 19-year-olds except in Pakistan, where contacts were highest among 5- to 9-year-olds. Non-home locations which presented high risk for transmission were schools in India, workplaces in Pakistan, and ‘other’ social / leisure locations in Mozambique and Guatemala. Among children under 5 years of age, the proportion of contacts with non-household members was highest in Mozambique and lowest in India; most of these were reported at home. Contact patterns by age diverge from prior projections that are extrapolated from contact data from high-income countries, underscoring the value of local data collection.

Subject terms: Infectious diseases, Epidemiology, Computational models


Social contact data are important for modelling epidemic transmission dynamics but limited data are available for lower- and middle-income countries. Here, the authors present social contact data in Guatemala, India, Mozambique and Pakistan collected in 2021-2023.

Introduction

Infectious diseases account for nearly a quarter of all mortality and half of all child mortality worldwide1,2. Half of deaths among children and adolescents caused by infectious diseases are due to enteric and lower respiratory infections3.

Person-to-person interactions, or ‘contacts’, facilitate the spread of pathogens which are transmitted through the air and by direct deposition onto surfaces. Patterns of contact are likely to vary across countries along with social and cultural norms that influence the composition and size of households, and the interactions that occur in community settings such as schools, workplaces, and on public transport. Within countries, variation in the structure and importance of social institutions, employment opportunities, and population density in rural compared to urban communities may lead to differences in individual behaviors which may influence the how quickly and to whom infections are likely to spread4,5. Children under five years of age play important roles in the transmission of many pathogens6–8 and are more vulnerable to severe disease from enteric and respiratory infections. Differences in caregiving practices and school attendance for young children that vary by setting may lead to different patterns of exposure during the vulnerable period of early childhood, shaping the epidemiology of enteric and respiratory infections in this age group9.

Past efforts to collect data on person-to-person interactions across countries have focused primarily on high-income countries. The POLYMOD study collected data on contact patterns in eight European countries in the early 2000s9; more recently, the CoMix study documented changing behaviors in Europe during the COVID-19 pandemic10. In contrast, relatively little data has been available directly from low- and middle-income countries (LMIC), which make up 85% of the global population11 and in which 89% of mortality from infectious disease among children occurs12. Social contact data collected from young children is available from some high-income countries but sparse in LMIC13,14. Among the data available from LMIC, key study design elements including the mode of data collection and sampling strategy have varied15,16, making comparisons across studies difficult. A study conducted during the COVID-19 pandemic collected data on social contacts in nineteen African countries, but recruited and interviewed participants by phone, so its generalizability to the general population and to non-pandemic time periods is unclear17.

The paucity of data from LMIC presents a challenge to disease modeling efforts, since estimating community-level impact of an intervention relies upon assumptions about the extent to which those directly receiving the intervention may interact with and therefore indirectly protect those who do not. Specifically, information on contact rates disaggregated by age, sex, and location of contact can be used to investigate the potential impact of targeted interventions, such as vaccination of a specific age group, or implemented in specific locations, such as social distancing or masking policies enacted in schools or workplaces. Efforts to generalize the POLYMOD data to other countries18,19 have been cited over 3000 times since they were published, demonstrating the persistent use of these data in modeling efforts despite known differences between contact patterns between high-income and LMIC20. Models supported by robust data on human behavior may facilitate more accurate predictions of the intensity and length of an outbreak, a clearer understanding of entrenched patterns of transmission of endemic infections, and more precise forecasts of intervention effectiveness.

To fill these gaps, we conducted a study across eight sites in four countries: Guatemala, India, Mozambique, and Pakistan. We compared person-to-person interactions relevant for the spread of enteric and respiratory infections within and across sites, with a special focus on children under 5 years of age.

Results

Across four countries, 5085 participants (n = 1141 in Guatemala, n = 1244 in India, n = 1363 in Mozambique, and n = 1337 in Pakistan) reported a total of 84,829 contacts across both days (Table 1 and Supplementary Table 1). Of those participants, 29% (n = 1480) were under five years of age. The daily number of contacts was lowest (mean: 6.5 contacts, SD: 1.9) in Guatemala and highest (12.5 contacts, SD: 6.2) in Pakistan. In Mozambique and Pakistan, more contacts were reported in the rural than in the urban site (8.3 contacts compared to 5.4 contacts, p < 1e−5 in Mozambique and 15 contacts compared to 10.2 contacts, p < 1e−5, in Pakistan). In India and Guatemala, mean daily contacts were more similar in the urban and rural sites (p = 0.040 and 0.043, respectively). (Table 2) Comparisons of medians, rather than means, gave qualitatively similar results (Supplementary Table 2).

Table 1.

Characteristics of study participants by site

Guatemala India Mozambique Pakistan
Urban Rural Urban Rural Urban Rural Urban Rural
N 566 575 622 622 687 676 679 658
n (%, col) n (%, col) n (%, col) n (%, col) n (%, col) n (%, col) n (%, col) n (%, col)
Age
<6mo 52 9.2% 59 10.3% 55 8.8% 58 9.3% 66 9.6% 62 9.2% 69 10.2% 64 9.7%
6–11mo 58 10.2% 45 7.8% 55 8.8% 54 8.7% 64 9.3% 82 12.1% 68 10.0% 62 9.4%
1–4 y 63 11.3% 55 9.6% 65 10.5% 61 9.8% 72 10.5% 63 9.3% 66 9.7% 62 9.4%
5–9 y 52 9.2% 65 11.3% 63 10.1% 63 10.1% 58 8.4% 64 9.5% 63 9.3% 66 10.0%
10–19 y 102 18.0% 100 17.4% 132 21.2% 127 20.4% 124 18.0% 125 18.5% 130 19.1% 128 19.5%
20–29 y 61 10.8% 64 11.1% 63 10.1% 62 10.0% 61 8.9% 64 9.5% 66 9.7% 63 9.6%
30–39 y 47 8.3% 62 10.8% 63 10.1% 63 10.1% 61 8.9% 64 9.5% 74 10.9% 63 9.6%
40–59 y 64 11.3% 61 10.6% 64 10.3% 71 11.4% 120 17.5% 89 13.2% 67 9.9% 79 12.0%
60+ y 67 11.8% 64 11.1% 62 10.0% 63 10.1% 61 8.9% 63 9.3% 70 10.3% 63 9.6%
Missing 0 0% 0 0% 0 0% 0 0% 0 0% 0 0% 6 0.9% 8 1.2%
Sex
Female 363 64.0% 372 64.7% 326 52.4% 317 51.0% 334 48.6% 332 49.1% 334 49.2% 322 48.9%
Male 202 35.8% 203 35.3% 296 47.6% 305 49.0% 353 51.4% 343 50.7% 343 50.5% 331 50.3%
Missing 1 0.2% 0 0.0% 0 0% 0 0% 0 0.0% 1 0.1% 2 0.3% 5 0.8%
Highest education completeda
Currently in school 128 32.6% 138 33.2% 163 36.5% 176 39.2% 195 40.2% 173 36.9% 83 17.4% 107 22.8%
No formal education 113 28.8% 133 32.0% 46 10.3% 42 9.4% 14 2.9% 65 13.9% 188 39.5% 260 55.3%
Primary 73 18.6% 90 21.6% 41 9.2% 41 9.1% 152 31.3% 141 30.1% 50 10.5% 71 15.1%
Secondary 39 9.9% 28 6.7% 155 34.7% 137 30.5% 99 20.4% 64 13.6% 152 31.9% 29 6.2%
Some college or higher 19 4.8% 3 0.7% 33 7.4% 45 10.0% 4 0.8% 4 0.9% 2 0.4% 3 0.6%
Missing 21 5.3% 24 5.8% 9 2.0% 8 1.8% 21 4.3% 22 4.7% 1 0.2% 0 0.0%
Occupationb
Student 98 28.7% 92 26.2% 112 29.2% 121 31.3% 132 19.2% 115 28.4% 67 16.2% 48 11.9%
Semiprofessional/professional 24 7.0% 4 1.1% 20 5.2% 12 3.1% 56 8.2% 27 6.7% 48 11.6% 6 1.5%
Semiskilled/skilled laborer 81 23.8% 53 15.1% 70 18.2% 85 22.0% 121 17.6% 95 23.5% 65 15.7% 157 38.9%
Other 1 0.3% 7 2.0% 34 8.9% 42 10.9% 14 2.0% 58 14.3% 29 7.0% 5 1.2%
Unemployed outside home 137 40.2% 195 55.6% 148 38.5% 126 32.6% 104 15.1% 110 27.2% 204 49.4% 188 46.5%
Household size (categories)
0–2 148 26.3% 145 25.2% 82 13.2% 78 12.5% 56 8.2% 70 10.3% 27 4.0% 8 1.2%
3–5 356 62.8% 348 60.5% 393 63.1% 383 61.6% 273 39.7% 277 41.0% 386 56.8% 184 28.0%
6–10 58 10.2% 81 14.1% 142 22.8% 155 24.9% 261 38.0% 249 36.8% 259 38.1% 347 52.7%
11+ 3 0.5% 1 0.2% 5 0.8% 6 1.0% 32 4.7% 31 4.6% 7 1.0% 119 18.1%
Missing 1 0.2% 0 0.0% 0 0.0% 0 0.0% 65 9.5% 49 7.2% 0 0.0% 0 0.0%
Median household size (IQR) 3 (2, 4) 3 (2, 5) 4 (4, 5) 5 (4, 6) 5 (4, 7) 5 (4, 7) 5 (4, 6) 7 (5, 10)
Generations in household
1 126 22.2% 121 21.0% 75 12.1% 73 11.7% 158 23.0% 145 21.4% 18 2.7% 21 3.2%
2 405 71.4% 422 73.4% 363 58.4% 310 49.8% 308 44.8% 322 47.6% 610 89.8% 518 78.7%
3 35 6.2% 32 5.6% 184 29.6% 239 38.4% 211 30.7% 208 30.8% 50 7.4% 118 17.9%
Missing 0 0.2% 0 0.0% 0 0.0% 0 0.0% 10 1.5% 1 0.1% 1 0.1% 1 0.2%
Multi-family household
No 503 88.7% 521 90.6% 523 84.1% 495 79.6% 515 75.0% 478 70.7% 660 97.2% 557 84.7%
Yes 63 11.1% 54 9.4% 99 15.9% 127 26.0% 162 23.6% 197 29.1% 18 2.7% 100 15.2%
Missing 0 0.2% 0 0.0% 0 0.0% 0 0.0% 10 1.5% 1 0.1% 1 0.1% 1 0.2%

aAmong participants over 5 years of age; in Mozambique, missing category includes some 5- to 9-year-olds who have not yet entered school.

bAmong participants 10 years of age and older.

Table 2.

Mean daily contacts by participant characteristics by country and site

Guatemala India Mozambique Pakistan
Overall, mean (SD) 6.5 (1.9) 7.9 (3.1) 6.8 (4.2) 12.5 (6.2)
Urban Rural Urban Rural Urban Rural Urban Rural
Mean (SD) Mean (SD) Mean (SD) Mean (SD) Mean (SD) Mean (SD) Mean (SD) Mean (SD)
6.4 (1.7) 6.6 (2.0) 8.1 (2.9) 7.7 (3.2) 5.4 (3.0) 8.3 (4.7) 10.2 (5.4) 15.0 (6.0)
Age
<6mo 6 (1.4) 6.1 (1.4) 6.7 (1.7) 6.2 (1.8) 3 (2.0) 6.3 (3.1) 7 (2.1) 11.6 (5.7)
6–11mo 6.3 (1.2) 6 (0.9) 7.4 (2.0) 6.6 (2.0) 4.4 (2.4) 7.2 (3.5) 6.7 (2.2) 11.8 (5.2)
1–4 y 6.5 (1.8) 6.7 (1.3) 8 (2.3) 8.3 (3.0) 5.4 (3.7) 7.3 (3.1) 6.6 (2.0) 15 (6.7)
5–9 y 6.5 (1.7) 7.1 (1.7) 9.8 (3.3) 10.1 (3.6) 5.1 (3.4) 8.4 (4.9) 9.5 (3.7) 18.8 (5.7)
10–19 y 6.8 (1.5) 7.2 (1.9) 9.7 (2.5) 10.7 (3.7) 7.2 (3.4) 9.6 (4.7) 10.8 (4.6) 16 (5.9)
20–29 y 6.8 (2.1) 6.6 (2.6) 7.9 (2.7) 7.8 (3.1) 5 (2.0) 8.8 (4.7) 10.2 (5.6) 13.3 (4.9)
30–39 y 6.1 (1.6) 6.4 (1.8) 7.8 (2.9) 7.7 (2.8) 5.3 (2.8) 7.8 (4.7) 13.2 (7.4) 15.2 (6.2)
40–59 y 6.2 (1.7) 6.7 (1.9) 7.8 (2.8) 6.9 (2.3) 4.7 (2.7) 8.4 (5.3) 9.8 (6.0) 14 (5.8)
60+ y 5.3 (1.5) 5.8 (1.7) 6.7 (3.0) 5.9 (2.3) 4 (2.5) 5.7 (4.1) 10.6 (4.3) 14.8 (5.4)
Sex
Female 6.5 (1.7) 6.7 (2.1) 8.0 (3.1) 7.5 (3.3) 5.3 (2.9) 8.2 (4.8) 6.9 (2.3) 16.2 (6.4)
Male 6.3 (1.7) 6.6 (1.7) 8.3 (2.8) 7.8 (3.0) 5.5 (3.1) 8.4 (4.7) 13.3 (5.6) 13.7 (5.3)
Household size
0–2 5.5 (1.6) 5.6 (1.5) 6.2 (2.5) 5.9 (2.8) 4.1 (2.5) 6.9 (3.9) 8.5 (6.1) 12.2 (4.9)
3–5 6.4 (1.4) 6.5 (1.5) 8.0 (2.7) 7.5 (3.2) 5.1 (2.7) 8.0 (4.8) 9.6 (5.7) 13.5 (6.0)
6–10 8.3 (1.8) 8.7 (2.4) 9.4 (2.9) 9.0 (2.7) 5.7 (3.1) 9.0 (5.1) 11.1 (4.6) 15.0 (5.9)
11+ 10.7 (5.2) –a 14.6 (2.8) 13.4 (0.6) 7.2 (3.5) 9.1 (3.5) 13.6 (2.3) 17.4 (5.6)
Highest education level attainedb
Currently enrolled in school 6.8 (1.7) 7.2 (2.0) 9.9 (2.7) 10.7 (3.7) 6.1 (3.3) 8.8 (4.7) 11.6 (4.5) 17.3 (5.8)
No formal education 6.2 (1.5) 6.5 (1.8) 7.0 (3.1) 6.4 (2.5) 4.3 (2.9) 6.9 (3.8) 8.8 (4.4) 14.4 (6.1)
Primary school 6.3 (1.6) 6.2 (1.6) 7.7 (2.4) 5.6 (2.0) 4.9 (2.6) 7.9 (4.9) 8.7 (5.0) 15.1 (4.9)
Secondary school 6.1 (1.7) 6.2 (3.0) 7.8 (2.9) 7.5 (2.6) 5.1 (2.5) 9.5 (5.0) 11.8 (6.3) 12.0 (4.9)
Some college/higher education 7.6 (3.0) 8.4 (3.5) 7.9 (2.7) 7.4 (2.4) 8.0 (4.1) 7.3 (4.0) 19.0 (14.5) 22.0 (5.8)
Occupationc
Student 6.7 (1.6) 7.2 (1.9) 9.8 (2.7) 10.7 (3.8) 6.2 (3.3) 9.2 (4.7) 10.7 (4.2) 17.7 (5.9)
Semiprofessional/professional 5.9 (1.6) 8.3 (3.8) 8.6 (3.2) 8.7 (3.1) 5.3 (2.3) 8.7 (4.7) 16.4 (4.9) 14.4 (4.6)
Skilled/semiskilled 6.9 (2.2) 6.5 (1.8) 8.4 (2.8) 7.6 (2.5) 4.6 (2.2) 8.5 (5.4) 13.5 (6.8) 13.5 (5.1)
Unemployed outside home 6.0 (1.5) 6.4 (2.1) 7.1 (2.8) 6.2 (2.3) 5.0 (3.1) 6.9 (4.4) 8.1 (4.4) 15.5 (6.2)
Other –a 5.9 (2.3) 7.9 (2.9) 6.9 (2.5) 9.1 (3.4) 9.0 (4.8) 8.4 (2.3) 12.2 (3.2)
Day of week
Weekday 6.5 (1.9) 6.8 (2.1) 8.3 (3.3) 7.8 (3.5) 6.0 (3.8) 8.6 (5.7) 10.1 (5.6) 14.7 (6.1)
Weekend 6.3 (1.7) 6.3 (2.0) 7.6 (2.6) 7.5 (3.2) 5.6 (3.4) 8.5 (5.7) 10.5 (5.4) 15.7 (6.6)

an = 1.

bAmong participants over 5 years of age; ‘Missing’ category excluded.

cAmong participants 10 years of age and older.

Mean daily contact rates were highest among 10- to 19-year-olds (7.0 contacts in Guatemala, 10.2 contacts in India, and 8.4 contacts in Mozambique) except in Pakistan, where mean contacts were highest among 5- to 9-year-olds (14.3 contacts). Qualitatively, patterns of contact by age were similar across countries and sites. In every site, assortative contacts, or contacts within the same age group, were highest among 10- to 19-year-olds (Fig. 1, Supplementary Fig. 1).

Fig. 1. Daily contact rates by participant age and site, contact age, and location of contact.

Fig. 1

Patterns of contact by country and age group. Panel A shows the number of contacts reported on day 1 by age group of participants in rural and urban sites. The center line of the box shows the median, and edges of the box are the 25th percentile (Q1) and 75th percentile (Q3). The minima of the whisker extends to Q1–1.5 *IQR, and the maxima of the whisker extends to Q3 + 1.5*IQR. Dots indicate outliers that exceeded minima or maxima of the whiskers. The number of participants is n = 1143 in Mozambique (Rural: n = 576, Urban: n = 567), n = 1244 in India (Rural: n = 622, Urban: n = 622), n = 1363 in Guatemala (Rural: n = 676, Urban: n = 687), n = 1323 in Pakistan (Rural: n = 650, Urban: n = 673). Panel B shows contact rates by participant age and contact age. Contact rates are made symmetric to adjust for differences in the size of age groups included and lack of reciprocity in contact reporting. Colors in each cell reflect the mean number of contacts reported between age groups, with higher numbers in dark blue and lower numbers in yellow. (The color scale is country-specific to visually demonstrate patterns.) Panel C shows the proportion of contact occurring in each location by age group. Contacts for which the location was not reported are not shown.

Household members accounted for one-third to one-half (36–44%) of contacts in all countries. In Pakistan, 86% of contacts were physical, compared to 66–73% of contacts in India, Mozambique, and Guatemala. Participants from the urban site in Pakistan reported an especially high proportion of physical contacts (93%). Outdoor contacts made up 18–22% of contacts in India, Guatemala, and Pakistan, but nearly half of all contacts (48%) in Mozambique (Supplementary Table 1).

In India, school contacts made up a larger proportion of children’s contacts than in other countries, peaking at 64% among 10- to 14-year-olds. Work contacts made up the highest proportion of contacts in India (6%) and the lowest proportion in Guatemala (2%). Work contacts were higher among older adults in India (10% among 60+-year-olds) than in other countries. Contacts on transit were more common in Guatemala (11% of all contacts) and India (11%) than in Mozambique (2%) or Pakistan (6%) (Fig. 1). Overall, the location at which contacts were reported were not substantially different by site within a country (Supplementary Fig. 2). Contacts greatest in duration were reported at home (mean of 2.5 h) and typically were shortest on transit (mean 1.2 h) and in markets or when performing other essential activities (mean 52 min) (Supplementary Fig. 3). In all sites and age groups, exposure-hours of contact were greatest at home (Supplementary Fig. 4).

Among children under 5 years of age, daily contact rates were highest in Pakistan (daily mean of 9.2, 9.2 and 10.7 contacts for children <6 months, 6–11 months, and 1–4-year-olds, respectively) and lowest in Guatemala (mean of 6, 6.2, and 6.5 contacts, respectively). The proportion of contacts with non-household members was highest in Mozambique (69% among 6- to 11-month-olds) and lowest in India (45% among 6- to 11-month-olds). To identify locations posing the highest risk to young children, we calculated contact exposure-hours (summing the length of each contact) by location with non-household members. Contact exposure-hours with non-household members were highest in Mozambique and occurred primarily at home in all countries except Guatemala, where the most contacts among <6-month-olds and 6- to 11-month-olds were reported in Other locations (Fig. 2).

Fig. 2. Mean daily exposure-hours of contact with non-household members by age group and location among children under 5 years of age.

Fig. 2

Mean daily exposure-hours were calculated by summing the product of the duration of contact and the number of contacts reported in each location. Contacts for which the location was not reported are not shown. Error bars show 95% confidence intervals of the mean. The number of participants in each age group in each country is as follows; n = 111, 103, 118 in Guatemala, n = 113, 109, 126 in India, n = 128, 146, 135 in Mozambique, n = 133, 130, 128 in Pakistan for <6mo, 6–11mo, and 1–4 y groups, respectively.

When we compared contact rates estimated from our study to those projected for each country using the POLYMOD data18, we found that projections tended to overestimate absolute contact rates, especially among children. However, we did not find substantial differences in age-specific contact patterns overall. In all countries, our data showed more contacts at home and fewer contacts at work than in the projections (Fig. 3, Supplementary Fig. 5).

Fig. 3. Comparison of contact rate estimates by age and location of contact with POLYMOD-based projections.

Fig. 3

A Mean daily contact rates by age estimated by Prem et al.2 based on the POLYMOD data and the mean daily contact rates estimated in our study, GlobalMix. 95% confidence intervals are shown with vertical error bars for GlobalMix data. The number of participants is n = 1141 in Guatemala (Rural: n = 575, Urban: n = 566), n = 1244 in India (Rural: n = 622, Urban: n = 622), n = 1352 in Mozambique (Rural: n = 675, Urban: n = 677, n = 1321 in Pakistan (Rural: n = 649, Urban: n = 672). B Proportions of contacts reported in each location, categorized using the POLYMOD scheme. Contacts in GlobalMix for which the location was unreported are shown in brown.

In a multivariable model including sex, household size, education level, and occupation, age remained a strong predictor of total daily contacts: school-age children and adolescents (5–9- or 10–19-year-olds) had the highest number of contacts in most sites. (Supplementary Tables 3–6) In the same model, sex was weakly associated with contact rates, except in Pakistan, where men reported a higher number of contacts than women in the urban site (OR: 1.52, 95% CI: 1.43–1.62) and a lower number in the rural site (OR: 0.82, 95% CI: 0.77, 0.88). In urban sites, participants that reported at least some college had more contacts than participants with primary school education (ORs ranging from 1.08 to 1.75). In a model with the same variables evaluating the relationship between number of non-household contacts reported and demographic characteristics, results were similar, but associations with age were generally weaker and differences by sex in Pakistan were more pronounced. In most sites, fewer non-household contacts were reported as household size increased.

We used a transmission model to simulate the introduction and spread of an infection under the age-structured contact patterns in each site. Cumulative incidence of infection by age group varied by country, with the highest overall incidence in rural Pakistan (96%) and the lowest in urban Mozambique (41%). Cumulative incidence was higher in the rural compared to the urban sites in Mozambique and Pakistan, and more similar across sites in Guatemala and India. Cumulative incidence was highest among adolescents (10- to 19-year-olds) in Mozambique but highest among adults in all other sites (Fig. 4, Supplementary Fig. 6). Cumulative incidence under projections from Prem et al.18 was higher than under GlobalMix contact patterns across all sites and age groups except in rural Pakistan. Likewise, Prem et al. contact patterns resulted in earlier epidemic peaks and more volatile effective reproduction numbers, indicating more rapid spread and resolution of the outbreak than under GlobalMix contact patterns. (Supplementary Figs. 7, 8)

Fig. 4. Effect of contact rate assumptions on epidemic dynamics by country and site.

Fig. 4

Brown bars show cumulative incidence of infection estimates under contact patterns based on those reported in Prem et al.2 which are based on projections using the POLYMOD data. Green and blue bars show cumulative incidence estimates under contact patterns estimates from our study, GlobalMix, in the rural and urban site in each country, respectively.

Discussion

We describe social behaviors relevant for transmission of enteric and respiratory infections in four low- and middle-income countries across three continents, with an emphasis on children under 5 years of age. Mean daily contact rates were highest among school-aged children in every country and tended to be higher in rural sites. Important non-home locations for contact varied across countries. Among children under 5 years of age, contacts with non-household members occurred primarily at home and ‘other’ locations. Overall, we show that patterns of social interaction are broadly similar across countries, but key differences, including within-country rural-urban variation and the relative importance of non-home locations, could have implications for transmission dynamics of enteric and respiratory infections and the effectiveness of interventions aiming to limit their spread.

Daily contact rates reported in this study are within range of some reported in the UK21 and Zimbabwe22, but overall lower than previous estimates. In a recent review and meta-analysis, the mean number of contacts per day reported across studies conducted from 2005 to 2018 in 22 countries, was 14.5 contacts20. It is notable that our estimates are low despite using prospective data collection and use of paper questionnaires, which have been shown to more fully ascertain contacts than other survey methods23. The timing of our study may partially account for fewer reported contacts, given the protracted impact of the COVID-19 pandemic on social behavior. Mozambique and Guatemala were likely most affected given the overlap of pandemic mitigation measures with data collection periods in 2021. However, despite official changes in pandemic mitigation policies that occurred during these periods, we show that the number of contacts reported was relatively stable through multiple months of data collection in each site. (Supplementary Figs. 9–12) This is consistent with anecdotal reports from field staff that while some policies to limit dense public gatherings were still in place through 2021, these restrictions were no longer meaningfully impacting daily interactions of most community members. Lower rates of contact could also stem from our household-based method of recruitment, which could have led us to recruit individuals who were more likely to be unemployed or otherwise less engaged in communal activities. Finally, it is also possible that contact rates have truly declined in recent years, given the widespread use of mobile smartphones and other technologies that have weakened social connections and reduced the frequency of in-person interactions.

Variations in the nature and number of contacts by country may reflect social, cultural, economic and even climactic differences. While the proportion of physical contacts reported in our study overall was similar to that reported in other low- and middle-income countries20, we report a particularly high proportion of physical contacts in Pakistan. This likely reflects customs of greeting friends and acquaintances that often involve touch, especially among men. (Men in Pakistan also reported substantially more contacts than women, which may reflect gendered religious and social norms that govern common interaction.) In Mozambique, more interactions were reported outdoors than in other countries. This may reflect the influence of climate, which in southern Mozambique features mild temperatures that fluctuate little through the year, and differences in the built environment that encourage congregating outside. All of these differences may influence the speed and patterns of spread of infections and can be used to better understand behaviors to target for intervention that may vary across settings.

In two of four countries, rural contact rates were substantially higher than in urban areas, a counterintuitive result that may reflect the relative closeness of community structure in rural areas despite lower population density. Our findings are generally consistent with other studies, some of which show higher contact rates in rural vs peri-urban participants16,24, and others4,25 which do not show differences. Understanding variation in contact patterns in rural and urban settings is particularly important when modeling the spread of new infections, which are often initially introduced into well-connected urban centers and diffuse more slowly to rural areas.

While we found similar rates of contact as reported previously for infants and young children26,27, our study provides additional detail on changes of children’s contact patterns from birth to age 5, a critical period of vulnerability to infectious disease. Across countries, the majority of children’s contacts are reported at home. However, the proportion of contacts occurring at home falls steadily from birth to age 5 in Guatemala and India and remains more stable in Mozambique and Pakistan. Patterns in biological susceptibility to infection change during this period, as maternally derived immunity wanes and the innate immune system matures28. Our data demonstrate the social changes that occur in parallel and shape exposure to pathogens in early childhood, and thereby risk of infection and illness in this age group28. Importantly, our study also differentiates contacts that occur at home and those that occur with household members. For children under 5 years of age, the two most important locations of exposure to non-household contacts were in their own homes and in Other locations. These patterns likely reflect sociocultural differences in caregiving practices for young children across countries, which can shape transmission patterns among children or between children and caregivers.

Similar to previous surveys18, we find the highest rates of contact among school-aged children and the highest rates of assortative mixing among adolescents (10–19-year-olds). The relative importance of contacts in schools across settings can inform context-specific predictions of the effectiveness of school closings and other policies that may limit school attendance to prevent spread of infection, especially given the great social costs of such interventions on children and communities.

While POLYMOD-based projections show broadly similar patterns of contact by age across the four countries; our estimates show differences in age patterns by country. POLYMOD-based projections tend to overestimate contacts at work and underestimate contacts at home, consistent with previous literature that has shown that people in LMIC have more contacts at home and fewer contacts at work than in high-income countries20. When we used our empirical data to model the spread of an emerging pathogen in each site, we found that cumulative incidence varied substantially, with sites reporting lower contact rates displaying lower incidence. In Pakistan, qualitative differences in contact patterns by age led to higher incidence among young children in the rural site as compared to the urban site. Compared to the epidemic simulated using POLYMOD-based projections of contact patterns, we saw qualitatively different patterns by age in cumulative incidence of infection, with substantially more variation by age in simulations using the GlobalMix data and a generally slower progression of the epidemic. Though this modeling exercise was based on simple assumptions which would need to be adjusted to better capture the more complex natural history of most enteric and respiratory pathogens, it illustrates that using context-specific data can lead to different inferences about how pathogens will spread and affect demographic sub-populations.

There were some limitations to this study. Information on children’s contacts were recorded using ‘shadows’, or individuals who recorded contacts on behalf of the participant. This may have led to inaccuracies since we do not know how completely shadows observed children’s contacts. Second, we defined contact as an interaction involving three or more words with another person at distance of an arm’s length. More transient interactions, which include being near a person in a crowded public place without speaking to them, are not captured by this definition. However, closer contacts are likely more important for transmission and allowed for comparison with previous work18. Third, these data should not be considered representative of each country. They can be considered representative of each site, since we adjusted overall contact rates for age and sex population distributions at each site. We recruited at multiple sites per country to capture at least some variation in contact patterns associated with varying population density within each country. Finally, we recruited participants over a limited time frame at each site, so could not capture seasonal variations in contact patterns that may be important for transmission (i.e., holidays, school start and end). One previous study in India has documented a higher number of contacts in the winter season relative to the monsoon or summer seasons; however, these differences were slight32.

This study represents the first effort to collect and compare social contact data using standardized tools across multiple LMIC, with a special focus on contact patterns among children under 5 years of age. These data can inform future modeling efforts aiming to accurately capture behavioral patterns that drive the speed and patterns of spread of infections in different contexts, and to estimate the impact of interventions targeting a broad range of pathogens.

Methods

Data collection procedures

The GlobalMix study aimed to characterize social contact patterns across eight sites – one rural and one urban per country – in four countries. We chose countries spanning three continents, three World Bank income level classifications, and a range of social, cultural, and economic features. (Supplementary Table 7) The study design is described in detail elsewhere and involved a qualitative data collection phase, which included focus group discussions and cognitive interviews to adapt the data collection instruments and study procedures to each site29. Households, defined as a group of people who shared space for cooking, were enumerated and sampling was guided by existing demographic surveillance systems (India, Mozambique, Pakistan) and satellite imaging of households (Guatemala). In India, Mozambique and Pakistan, we used demographic surveillance data to generate age-stratified lists of community members and randomly select individuals within each age group for participation using a random number generator. Selected individuals were approached by field staff in the order they were selected until age-specific quotas were reached. In Guatemala, where demographic surveillance system data were not available, households within communities were randomly numbered and visited sequentially. Field staff randomly selected one member from each household and offered participation in the study until age-specific quotas were reached. Informed consent was obtained from adults over 18 years old and from parents/guardians for minors under 18 years old. Participants completed an enrollment questionnaire which included information on their age, gender, school enrollment/occupation, and characteristics of their household.

Field workers trained participants to complete a paper-based contact diary. Participants were randomly assigned a day of the week to begin two consecutive days of reporting contacts. Field workers returned to the household after this period and entered the contacts recorded by the participant into an electronic survey platform. Participants reported characteristics of the person with whom they had contact, such as their age, gender, and relationship to the participant, as well as the location (home, store/market, school, work, place of worship, and other country-specific options), duration (<5 min, 5–15 min, 16–30 min, 31 min–1 h, 1–4 h, and 4+ hours), environment (indoors/outdoors), and proximity (physical/non-physical) of the contact. For adults or children unable to complete the diary themselves, study staff assisted in selecting another person (a ‘shadow’) to record the participant’s contacts. Both participants and shadows received training in how to record contacts and field staff periodically checked in with participants over the two-day contact reporting period to ensure complete and accurate reporting by shadows.

We defined contact as a two-way conversation with three or more words exchanged at a physical distance of an arm’s length or less from another person. A subset of contacts were classified as physical, which we defined as a contact that involved directly touching someone’s skin or the clothes they were wearing.

Statistical analysis

We applied post-stratification weights representative of the population age distribution at each site. Using these weights, we calculated the mean, standard deviation, median, and interquartile range of daily contacts, averaged across two days of reporting, for each site and country overall. We generated symmetric age-stratified contact matrices by age. Symmetric contact matrices adjust for differences in the size of age groups included in the sample and lack of reciprocity in contact reporting (i.e., 10–14-year-olds reported more contacts with 60+ year-olds than 60+ year-olds reported with 10–14-year-olds). We calculated contact exposure-hours by summing the length of each contact, using the midpoint of each duration category (e.g., 2.5 min for <5 min), and using 4 h for contacts reported as longer than 4 h.

We defined physical contacts that lasted longer than one hour as high risk for transmission of most infections. For infections that transmit via deposition onto surfaces (hands, faces, fomites), physical contact is more likely to lead to transmission; for infections that transmit through the air, exposures longer than one hour are usually deemed sufficient for spread30–33.

To assess the robustness of relationships between demographic characteristics with contact rates, we evaluated associations between participant characteristics and (1) overall contacts and (2) non-household contacts using negative binomial multivariable regression models. We report site-specific adjusted odds ratios and 95% confidence intervals for each model.

Transmission modeling

To explore the impact of incorporating context-specific, empirical data on contact patterns on transmission dynamics, we built a simple compartmental model representing the transmission of an acute respiratory or enteric infection. We simulated an epidemic of an infection to which everyone is initially susceptible, with a mean duration of 14 days and which confers lifelong immunity. (Supplementary Table 8) We created a synthetic population representative of the age distribution in each site and separately modeled the spread of the infection in rural and urban areas based on (1) the GlobalMix contact patterns and (2) projected contact patterns for each country using POLYMOD data from Prem et al.18. We calculated and compared age-specific cumulative incidence, epidemic peak timing, and the effective reproduction number under different contact patterns. We simulated from the compartmental model using the ‘EpiModel’ R package version 2.4.0. and effective reproduction numbers were estimated using the parametric serial interval method, assuming a mean serial interval of 2.6 and standard deviation of 1.5, in the ‘EpiEstim’ R package version 2.2.4.

All data analysis was conducted in R version 4.1.2.

Ethical approval

This study was approved by Emory University Institutional Review Board (approval number 00105630), Yale University Institutional Review Board (reliance agreement approval number 2000026911), and review boards at Christian Medical College Vellore (Approval #12065), Universidad del Valle de Guatemala (Protocol #223-12-2020) and the National Ministry for Publicparticipants 10 years of age Health and Social Assistance in Guatemala (Protocol #31-2020), Manhiça Health Research Institute Internal Scientific Committee (CCI/03/2020) and Internal Ethical Review Board (initial approval CIBS-CISM/011/2020), and The Aga Khan University (Approval #2022-7912-23648).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Reporting Summary (75.3KB, pdf)

Acknowledgements

This study was funded by NIH/NICHD R01HD097175-01 and the lead author was supported by NIH/NIAID K01AI166093-01A1. The Manhiça Health Research Center (CISM) is supported by the Government of Mozambique and the Spanish Agency for International Development Cooperation (AECID). Polana Caniço Health Research Center (CISPOC) is supported by the Government of Mozambique. The funders of this study had no role in its design and conduct, nor the analysis and presentation of results. This study would not have been possible without the generous time and support of the individuals who participated. We also thank those who provided essential support of this project at the sites, including the social sciences teams in Mozambique and Guatemala (Clara Secaira, Liza Muñoz, María Fernanda González, among others), Consejos Comunitarios de Desarrollo Urbano y Rural (COCODES) and Ministry of Health Centers and Posts from San Juan Ostuncalco and Concepción Chiquirichapa, Quetzaltenango, Guatemala. We are grateful for the tireless work of field teams and data managers that were involved in data collection and database management at the sites. Finally, we are indebted to Gayathri S., a friend and colleague, who made valuable contributions to the study in India and to whom this manuscript is dedicated.

Author contributions

B.A.L., S.B.O., F.M., A.M., S.M.J., V.R., A.B., C.S., H.M., and M.K. conceived of and designed the study, M.C.K. led the cross-country data collection efforts, K.N.N., I.M., E.J., C.T., A.J., N.C., C.J., H.M.A., P.S., R.S., R.A., C.L., N.A., O.A., M.A.S., and I.Y. supported specific aspects of the data collection and/or led data collection at the sites, K.N.N., M.C.K., M.S., S.S.K., P.V.P., and D.C. conducted data analysis, K.N.N. wrote the initial draft of the manuscript, and all authors critically reviewed the manuscript for intellectual content.

Peer review

Peer review information

Nature Communications thanks Juanjuan Zhang and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The questionnaires and data for this analysis are available at https://github.com/lopmanlab/GlobalMix and can also be found on Zenodo. (10.5281/zenodo.16754436)34.

Code availability

The code to reproduce this analysis is available at https://github.com/lopmanlab/GlobalMix and can also be found on Zenodo. (10.5281/zenodo.16754436)34.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

These authors contributed equally: Kristin N. Nelson, Moses C. Kiti.

These authors jointly supervised this work. Benjamin A Lopman, Saad B Omer.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-64850-9.

References

  • 1.Global health estimates: Leading causes of death. https://www.who.int/data/gho/data/themes/mortality-and-global-health-estimates/ghe-leading-causes-of-death (2000–2021).
  • 2.Perin, J. et al. Global, regional, and national causes of under-5 mortality in 2000–19: an updated systematic analysis with implications for the Sustainable Development Goals. Lancet Child Adolesc. Health6, 106–115 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.The unfinished agenda of communicable diseases among children and adolescents before the COVID-19 pandemic, 1990–2019: a systematic analysis of the Global Burden of Disease Study 2019. Lancet402, 313–335 (2023). [DOI] [PMC free article] [PubMed]
  • 4.Read, J. M. et al. Social mixing patterns in rural and urban areas of southern China. Proc. R. Soc. B Biol. Sci.281, 20140268 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Kiti, M. C., Melegaro, A., Cattuto, C. & Nokes, D. J. Study design and protocol for investigating social network patterns in rural and urban schools and households in a coastal setting in Kenya using wearable proximity sensors. Wellcome Open Res4, 84 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Bennett, A. et al. Community transmission of rotavirus infection in a vaccinated population in Blantyre, Malawi: a prospective household cohort study. Lancet Infect. Dis.21, 731–740 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Worby, C. J. et al. On the relative role of different age groups in influenza epidemics. Epidemics13, 10–16 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Cohen, C. et al. Incidence and transmission of respiratory syncytial virus in urban and rural South Africa, 2017-2018. Nat. Commun.15, 116 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Mossong, J. et al. Social contacts and mixing patterns relevant to the spread of infectious diseases. PLOS Med.5, e74 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Wong, K. L. M. et al. Social contact patterns during the COVID-19 pandemic in 21 European countries – evidence from a two-year study. BMC Infect. Dis.23, 268 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.World Bank Open Data. World Bank Open Datahttps://data.worldbank.org.
  • 12.Jamison, D. T. et al. Global health 2035: a world converging within a generation. Lancet382, 1898–1955 (2013). [DOI] [PubMed] [Google Scholar]
  • 13.Campbell, P. T., McVernon, J., Shrestha, N., Nathan, P. M. & Geard, N. Who’s holding the baby? A prospective diary study of the contact patterns of mothers with an infant. BMC Infect. Dis.17, 634 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Hoek et al. The social life of infants in the context of infectious disease transmission; social contacts and mixing patterns of the very young. PLOS One8, e76180 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Nagpal, S. et al. Seasonal variations in social contact patterns in a rural population in north India: Implications for pandemic control. PLOS One19, e0296483 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Leung, W. T. M. et al. Social contact patterns relevant for infectious disease transmission in Cambodia. Sci. Rep.13, 5542 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Dobreva, Z. et al. Characterising social contacts under COVID-19 control measures in Africa. BMC Med.20, 344 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Prem, K. et al. Projecting contact matrices in 177 geographical regions: An update and comparison with empirical data for the COVID-19 era. PLOS Comput. Biol.17, e1009098 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Prem, K., Cook, A. R. & Jit, M. Projecting social contact matrices in 152 countries using contact surveys and demographic data. PLOS Comput. Biol.13, e1005697 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Mousa, A. et al. Social contact patterns and implications for infectious disease transmission – a systematic review and meta-analysis of contact surveys. eLife10, e70294 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Klepac, P. et al. Contacts in context: large-scale setting-specific social mixing matrices from the BBC Pandemic project. Preprint at 10.1101/2020.02.16.20023754 (2020).
  • 22.Melegaro, A. et al. Social contact structures and time use patterns in the Manicaland Province of Zimbabwe. PloS One12, e0170459 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Leung, K., Jit, M., Lau, E. H. Y. & Wu, J. T. Social contact patterns relevant to the spread of respiratory infectious diseases in Hong Kong. Sci. Rep.7, 7974 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Kiti, M. C. et al. Quantifying age-related rates of social contact using diaries in a rural coastal population of Kenya. PLOS ONE9, e104786 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Melegaro, A., Jit, M., Gay, N., Zagheni, E. & Edmunds, W. J. What types of contacts are important for the spread of infections? Using contact survey data to explore European mixing patterns. Epidemics3, 143–151 (2011). [DOI] [PubMed] [Google Scholar]
  • 26.Oguz, M. M., Camurdan, A. D., Aksakal, F. N., Akcaboy, M. & Acoglu, E. A. Social contact patterns of infants in deciding vaccination strategy: a prospective, cross-sectional, single-centre study. Epidemiol. Infect.146, 1157–1166 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Neal, E. F. G. et al. Associations between ethnicity, social contact, and pneumococcal carriage three years post-PCV10 in Fiji. Vaccine38, 202–211 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Simon, A. K., Hollander, G. A. & McMichael, A. Evolution of the immune system in humans from infancy to old age. Proc. R. Soc. B Biol. Sci.282, 20143085 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Aguolu, O. G. et al. Comprehensive profiling of social mixing patterns in resource poor countries: a mixed methods research protocol. PLOS One19, e0301638 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Kenyon, T. A., Valway, S. E., Ihle, W. W., Onorato, I. M. & Castro, K. G. Transmission of multidrug-resistant Mycobacterium tuberculosis during a long airplane flight. N. Engl. J. Med.334, 933–938 (1996). [DOI] [PubMed] [Google Scholar]
  • 31.Driver, C. R., Valway, S. E., Morgan, W. M., Onorato, I. M. & Castro, K. G. Transmission of Mycobacterium tuberculosis associated with air travel. JAMA272, 1031–1035 (1994). [PubMed] [Google Scholar]
  • 32.Butler, A. J., Thomas, M. K. & Pintar, K. D. M. Expert elicitation as a means to attribute 28 enteric pathogens to foodborne, waterborne, animal contact, and person-to-person transmission routes in Canada. Foodborne Pathog. Dis.12, 335–344 (2015). [DOI] [PubMed] [Google Scholar]
  • 33.DuPont, H. L., Levine, M. M., Hornick, R. B. & Formal, S. B. Inoculum size in shigellosis and implications for expected mode of transmission. J. Infect. Dis.159, 1126–1128 (1989). [DOI] [PubMed] [Google Scholar]
  • 34.Nelson, K., Shiiba, M., Kim, S. & Prasad, P. GlobalMix. Zenodo10.5281/zenodo.17094566 (2025).

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Reporting Summary (75.3KB, pdf)

Data Availability Statement

The questionnaires and data for this analysis are available at https://github.com/lopmanlab/GlobalMix and can also be found on Zenodo. (10.5281/zenodo.16754436)34.

The code to reproduce this analysis is available at https://github.com/lopmanlab/GlobalMix and can also be found on Zenodo. (10.5281/zenodo.16754436)34.


Articles from Nature Communications are provided here courtesy of Nature Publishing Group

RESOURCES