Characterizing social contact patterns in rural and urban Mozambique: the GlobalMix study, 2021-2022

2. Background

Human social contact patterns drive the transmission of pathogens that spread through proximity to infectious individuals. Data on social contact patterns are critical to understand “who-contacts-whom” and infer who-acquires-infection-from-whom, providing insight on potential control measures such as physical distancing and vaccination. Underlying the patterns of contact are demographic, sociocultural, and economic determinants which vary within and across regions resulting to corresponding variation in contact patterns. Unfortunately, such critical data are not as widely available in low-and-middle-income countries (LMICs), including Mozambique [1], as they are in high-income countries (HICs) [2]. Where data are available, they were collected across e.g., rural-urban divide [3–6] or informal settlement [7,8] limiting the representativeness of the data. More recent studies in LMICs have incorporated innovative methods to obtain data from hard-to-reach populations especially children and illiterate adults by either using shadows [3,9] interviewer-led questionnaires [4,5], or wireless proximity sensors [10,11]. In the absence of locally collected data, simulated contact rates for LMIC populations may also be derived by projecting empirical data collected from HICs (e.g., POLYMOD data) and scaled using local demographic patterns [12]. However, these extrapolations likely mischaracterize contact patterns in important ways when they differ for reasons aside from demographics.

During the early phase of the COVID-19 pandemic, it became clear the extent to which human contact patterns could be modified, and at essentially a global scale [13–17] In the absence of vaccines or pharmaceutical interventions, populations responded by “physical distancing”, i.e., drastically reducing the number and riskiness of contacts. Work contacts and those out of the home decreased and masking increased. Again, however, relatively little data were collected from LMICs, limiting our ability to quantify these changes and use such data to develop models of interventions.

Starting 2021, during the COVID-19 pandemic, we launched the “GlobalMix Study” to collect social contact data from four LMICs: Mozambique, Guatemala, India, and Pakistan. Across all four countries, we are collecting data from selected rural and urban areas using standardized questionnaires and methods that were customized for each context considering social and cultural norms [18]. In this paper, we present the methods and results from Mozambique for which we have complete datasets.

3. Methods

4. Results

4.2 Contact patterns

Participants reported a total of 17674 contacts over two days, with 41% unique contacts (n=3904 day 1 only and n=3250 day 2 only) and 59% (n=10304) reported on both days (repeat contacts). Participants reported an overall mean of 13.1 (95% CI 12.6–13.5) contacts over two days (Table 2). We observe a significant difference in the mean number of contacts reported on day 1 compared to day 2 (8.3 [8.0–8.6] vs 5.5 [5.3–5.7], respectively, paired t-test p<0.01). Since diary completion dates were randomly assigned, the actual mean contacts should not vary between the first and second date of diary completion. Therefore, we believe that the observed difference is a result of reporting bias resulting from participant fatigue; from here henceforth we report the mean and median number of contacts on day 1 only. The rural mean contact rate was significantly higher than the urban rate (9.8 [9.4–10.2] vs 6.8 [6.5–7.1], p<0.01) (Figure 1 A & B). Contact rates were higher in rural areas for each age group (Figure 1 C). While rural mean number of contacts with non-household members was significantly higher that contacts with household members (6.6 vs 3.9, p<0.01), there was marginal difference in the urban site (4.2 vs 3.6, p=0.46). Corresponding median values for day 2 by site are available in Supplementary table 1.

• Download figure
• Open in new tab

Figure 1. Distribution patterns of number of contacts in rural and urban Mozambique

Panels A and B show the density distribution of the number of contacts per person in the rural and urban site, respectively, with the mean and 95% CI shown by the black vertical line. Panel C shows boxplots of the distribution of number of contacts by site. The middle line shows the median number of contacts and the lower and upper line in the boxplot shows the lower and upper quartile of the distribution, respectively. Panels D and E show the age specific contact matrices depicting the average mean number of individuals in age-group (y-axis) with whom a

View this table:

• View inline
• View popup
• Download powerpoint

Table 2. Median (IQR) and mean (bootstrapped 95% CI) number of contacts on day 1 only

Physical contacts were, on average, more numerous than conversation only contacts in both rural (6.7 vs 4.9, p<0.01) and urban areas (5.3 vs 3.3, p<0.01). Participants aged ≤18 years were the main drivers of the higher physical contacts. Out of all participants, 803 (59%) reported having the same number of social contacts compared to periods before the COVID-19 pandemic. From these, urban participants reported either significantly fewer (n=238, mean = 5.9 [95% CI 5.5– 6.3]) or more (n=28, 11.7 [9.3–14.1]) mean number of contacts compared to those who reported no change (n=410, mean = 7.0 (6.7–7.4)). In the rural site, 74 (11%) of the participants reported more mean contacts than usual (12.6, 95% CI 10.9–14.3) and this was significantly different from those who reported either no change (n=384, mean = 9.6 [9.1–10.1]) or fewer contacts (n=215, mean = 9.2 [8.5–9.9]).

4.2.1 Contact matrices

The patterns of age-specific social contacts are generally similar when comparing the urban and rural sites (Figure 1 panel D and E). The urban matrix, however, suggests lower mean number of contacts across all ages compared to the rural site. Participants aged 5-14 years (school-going children) and older working adults aged 30–59 years in both rural and urban areas reported higher assortative (with same age) mean contacts. Older school goers aged 15–19 years also reported on average, high number of contacts with 10–14-year-olds in both sites. Another peak in mean number of contacts is observed between 30–39- and 40–59-year-olds, driven mostly by conversation only contacts (Supplementary figure 2 panels (C) and (D)). In both sites, few interactions are reported between the other ages and little to no contacts are reported with infants. Despite low mean contacts reported by infants, the mean number of contacts between them and other ages generally increased with age to peak at 10–14 years in rural and 30–39 years among urban dwellers.

4.2.2 Patterns of contact by location

In both urban and rural areas, the estimated number of co-located individuals by far exceeded the number of contacts reported by participants. Overall, participants reported a mean of 26.9 (95% CI 23.5–30.3) proximity contacts compared to 6.8 (6.5–7.1) close contacts in the urban site, and 23.1 (20.3–25.9) proximity contacts and 9.8 (9.4–10.2) close contacts in the rural site. The three locations with highest mean number of contacts were places of worship, schools, and transport hubs (Table 3). In addition, rural participants were more likely (n=752 visits, [48%]) to visit other homes compared to urban participants (n=288 visits [29%]). Despite overall numbers being similar, the locations where contact occurred was meaningfully different between urban and rural sites.

View this table:

• View inline
• View popup
• Download powerpoint

Table 3. Number of times that participants reported visiting a location (N) and the median (IQR) number of people reported per location on day 1

4.2.3 Sensitivity of transmission model disease dynamics to our empirical contact matrices

To recap, we restructured the age-specific matrices presented in Figure 1 D and E into 7 age groups (see Methods) depicted in Figure 2 panels A and B. Assortative contacts in empirical data were highest among 10–19-year-olds (higher in rural compared to urban) compared to synthetic values for 0–9-year-olds (panel C, data available from [12]).

• Download figure
• Open in new tab

Figure 2. Contact matrices and modeled age-specific vaccine effects (VE)

Panels A and B show rural and urban contact matrices from our empirical data, respectively, while panel C shows the synthetic contact matrix derived from Mozambique-specific demographic data by Prem et al [12]. Panels D and E shows the overall vaccine effects (VE) of a respiratory infection comparing synthetic and empirical contact rates in the rural and urban site, respectively. Legend key: ER=Empirical Rural, SR=Synthetic Rural, EU=Empirical Urban, SU=Synthetic Urban.

In the rural and urban site, the values of the empirical overall vaccine effects (VE) overlapped in most ages. However, synthetic VE were marginally higher in 10–19 and 40–49-year-olds in the rural site (Figure 2 D), and higher in all ages (particularly adults aged 30–59-year-olds) apart from children aged 0–9-year old (Figure 2 E) in the urban site. Specifically, we report higher attack rates in unvaccinated (AR) compared to vaccinated (ARv) individuals (AR: 94%, ARv: 77%) and lower overall vaccine effects (18%) for 0–9-year-olds compared to attack rates using empirical data (rural AR: 84%, rural ARv, 62%, rural vaccine effects: 26%; urban AR: 84%, urban ARv: 62%, urban vaccine effects: 26%). Additionally, synthetically contacts among 60+ years old were underestimated compared to empirical values, producing notably lower attack rates among this age group (Synthetic AR: 49%, Synthetic ARv: 31%; Rural AR: 40%, Rural ARv: 24%; Urban AR: 36%, Urban ARv: 21%).

5. Discussion

We present results from a two-day cross-sectional study aiming to quantify social contact rates among residents of a rural and urban site in Mozambique during the COVID-19 pandemic. We first engaged with the local community to get their views on the suitability and acceptability of our tools and study procedures. We made several key observations. First, we used the qualitative outcomes to modify the format and content of the paper diaries to make them more user friendly (see additional details in Supplementary text 1). Second, in our sample population, participants from the rural site had significantly higher average number of contacts compared to the urban site. Third, in each site, reported mean contacts increased with age to peak at school-going children/ teenagers aged 15-19 years, and mean contacts were higher among adults (>18 years) compared to children <5 years. Fourth, mixing was assortative (increased frequency of contacts within the same ages) among school-going children, and less pronounced inter-generational mixing particularly in the urban site. Finally, we quantify the importance of using empirical contact data to model disease dynamics and interventions. In model simulations of an emerging respiratory pathogen, we find meaningfully different attack rates and vaccine impact among both child and elderly groups when using our local data compared with widely used contact matrices modelled from other settings.

In the earlier phases of the COVID-19 pandemic, Mozambique adopted physical distancing policies including school closures and restrictions on social gatherings, but with no express requirement to stay at home [23]. This was similar to measures implemented globally to reduce transmission of SARS-CoV-2. Few countries in sub-Saharan Africa e.g., South Africa [24], Kenya [3,5], Zimbabwe [4], Uganda [9], and Somaliland [8] had existing empirical social contact data collected in various settings prior to 2020. There are sparse data on contact patterns during the pandemic from LMICs (examples are from Kenya [7] and Malawi [6]) compared to high income countries such as the United Kingdom and Europe [17] and USA [16]. Longitudinal data during the pandemic would have been critical for a better understanding of the transmission pathways of the novel virus in LMICs taking account of demographic, social and economic contexts, and provide insights on how to enhance NPI use in place and identify priority groups for immunization once vaccines became available albeit in limited supply and use in LMICs.

Compared to pandemic periods in Kenya [7] and Malawi [6], we observed lower mean number of contacts in Mozambique but a higher number of contacts reported by participants in the rural compared to the urban site. However, we interpret this with care since data from Kenya and Malawi were collected from high-density settlements where individuals may have been unable to fully adhere to physical distancing mainly due to economic reasons. The Government of Mozambique periodically revised physical distancing policies to curb the spread of SARS-CoV-2 [25], and we also think that the measures were not strictly adhered to particularly among school-going children and working adults.

We implemented several innovations in the GlobalMix study. First, we collected contact data from participants over two consecutive days. The average number of contacts stratified across different covariates of interest remained constant over two days, suggesting the stability of participant’s recall as well as the number and nature of contacts made over multiple days. The stability of contact networks across days has been previously demonstrated in rural households in Kenya and Malawi using autonomous methods that minimize recall bias [10,11]. A second innovation of our study was an estimation of group proximity contacts at locations frequently visited by participants. It is important to note that participants report almost four times the number of proximity contacts compared to detailed individually reported contacts. This suggests the potential to substantially underestimate the number of interactions that could lead to transmission events particularly in highly mobile age groups congregating in social venues.

Finally, our transmission model simulation demonstrates the importance of contextual empirical social contact data. Although advanced methods for projecting social contact patterns onto regions without data exist (e.g., [12]), we found that age-specific infection attack rates from a model using empirical contact data differed meaningfully compared to a model parameterized with synthetic contact rates. Importantly, we found the largest differences in attack rates (comparing vaccinated versus unvaccinated individuals) resulted in increased vaccine impact in the youngest (0-9 years) age group, who often represent the most vulnerable group. These findings were consistent with a Uganda model where use of local contact pattern data resulted in larger epidemics in young children and smaller epidemics in adults aged >35 years compared to using UK-based contact data [9]. It is also notable that we observed distinct contact patterns resulting in divergent model results for the Mozambique rural and urban sites. This highlights the impact of sub-national differences in contact patterns resulting in variation in disease dynamics. Such insights are not possible with widely used contact data which are generally available at the national level.

6. Conclusion

We present empirical results of a cross-sectional study quantifying rates and patterns of human social contacts relevant for the spread of directly transmitted infections in rural and urban areas of Mozambique. We demonstrate the possibility of collecting high quality social contact data from resource-poor settings thus reducing reliance on synthetic data modelled from high-income countries. Finally, we demonstrate the potential advantages of empirical compared to synthetic data in a transmission and vaccine control model and advocate for the use of contextual data in similar studies. Nevertheless, some questions remain including if social contact patterns changed in this setting as non-pharmaceutical interventions were relaxed. Finally, we are completing data collection from three other LMICs and we will make all our data collection tools, data, and analysis scripts findable, accessible, interoperable, and reusable (FAIR).

9. Funding

This work was supported by the National Institutes of Health (NIH) grant number: R01 HD097175-01. The views expressed in this publication are those of the author(s) and not necessarily those of NIH or the US government. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. CISM is supported by the Government of Mozambique and the Spanish Agency for International Development (AECID).

10. Availability of data and materials

All data produced in the present study are available on our GitHub repository: https://github.com/lopmanlab/globalmix-mozambique-aim1. We also include the survey codebooks in English and Portuguese, as well as the Portuguese version of the diary to portray the age group images used in the study.

11. Authors contribution

12. Competing interests

The authors declare they have no competing interest.

13. Supplementary Information

Supplementary text 1: Details of fieldwork, challenges experienced, and practical solutions applied.

This was the first time that self-report paper diaries were being used in the Manhiça and Polana Canhiço HDSS. It is important to note the logistic and technical challenges we experienced during the conduct of our study, and our mitigation measures, to better inform future studies in similar settings. Following extensive consultations with community members, we used paper diaries with contextualized questions and customized visual aids to guide age selection of contacts, including those met for the first time. During implementation, adult males were hard to find since most left their houses early for work, market sales and to tend to their farms. Absenteeism was also reported in a similar study in Kenya [3] and the authors replaced the loss- to-follow-up individuals with others of similar demographic characteristics. Due to the pandemic, both HDSS databases had not been updated and we resorted to door-to-door recruitment for under-5-year-olds and later expanding the study catchment area to additional HDSS blocks in the urban site. During the rainy season, remote areas in the rural site were inaccessible thus hampering recruitment and data collection exercises. Once recruited, some participants were unavailable to present their diary data during the exit visits, and almost half of diaries in both sites were partly filled in, thus requiring assistance from our field staff to complete the diary. Eventually, we resorted to recruiting children from mothers attending health clinics, visits to workplaces for adult males, and following up participants through phone calls. The lapse in time between diary data collection period and exit survey could have contributed to recall bias and it was not possible to ascertain the magnitude and direction of the recall bias. Despite this, we had a good buy in from the community due to our extensive community engagement conducted before and during the study, which also resulted in higher participation rates in general than previously observed in other sub-Saharan African settings. Furthermore, the data was collected during periods of active transmission of SARS-CoV-2 in the community, including certain times when certain restrictive measures were in place, such as the closure of schools and parks, limitations on the number of employees at the workplace and passengers on public transportation, social distancing, and other measures [23]. This context likely caused some disruption, given the contingency scenario at the time, and may have influenced the characteristics of social contact patterns due to the ongoing pandemic.

• Download figure
• Open in new tab

Supplementary figure 1.

Matrices of age–specific average physical only and conversation only contacts by site. The panels are arranged by rural (A) and urban (B) physical contacts and rural (C) and urban (D) conversation contacts. Average age-specific values are generated by dividing the total number of contacts reported by age group pair with the number of participants in that age group.

• Download figure
• Open in new tab

Supplementary figure 2.

Violin plots comparing the distribution of number of contacts versus number of co-located individuals reported by participants at each location in rural (A) versus urban (B) site.