Characterising social contacts under COVID-19 control measures in Africa

Telephone interviews were conducted with adult respondents (aged 18 and over) with access to a landline or a mobile phone in 18 countries (Cameroon, Democratic Republic of Congo, Egypt, Ethiopia, Ghana, Guinea, Ivory Coast, Kenya, Liberia, Mozambique, Nigeria, Senegal, South Africa, Sudan, Tunisia, Uganda, Zambia, and Zimbabwe) during August 4 to 17, 2020 (survey 1, S1) and these countries plus Morocco during February 2 to 28, 2021 (survey 2, S2). The survey was designed as a quota sample survey through Random Digit Dial (RDD) with quotas on gender and urban/rural location within each country to achieve a sample closely matching the profile of each country. Multiple call-backs to unanswered numbers dialled was not part of the design [23].

Respondents were first asked questions relating to their knowledge about COVID-19, their risk perceptions of catching COVID-19, their general health, and the potential impact of SARS-CoV-2 on their heath (Additional file 1). They were subsequently asked about their attitude towards the existing control measures, their confidence in the government’s response, their attitude towards vaccination (in survey 2 only) and the burden of the pandemic on them and their household. Respondents were finally asked whether they had been diagnosed or believed to have had COVID-19, about their socio-demographic characteristics, and the number of contacts they had had in the last 24 h with contactees (people who respondents had a contact with) aged 0–17, 18–55, or over 55. A contact was defined as “anyone with whom [the respondent] exchanged at least a few words and was close enough to not need to raise [their] voice or [who they] had direct physical contact with (including handshaking or other contact).” Contacts were categorised according to contacts with members of the household; with external visitors to the household; with people at work, school, or university; or with people in other places. Demographic data such as the respondent’s self-reported age, residence, household size, monthly income, and educational level were also collected.

R version 4.1.1 and Stata version 16.1 were used for the cleaning and preparation of the survey datasets and the subsequent analysis [24, 25]. Data analysis was performed at the level of individual countries due to the anticipated variability between countries and given the novelty of such country-level analysis.

Respondents’ gender, age, residence area, household head education level, monthly income and household size were characterised (Table 1 and Additional file 2: Table S1 for surveys 1 and 2, respectively). Survey 2 had not collected monthly income in a common currency, preventing comparisons across countries. The age structure of the sampled populations was also compared to United Nations’ World Population Prospects (WPP) data on the number of people by age group (Additional file 2: Table S2) and the percentage of female and rural population according to World Bank estimates (Additional file 2: Table S3) in each country to assess the representativeness of the sample [26].

The distribution of contacts by contactees’ age group (0–17, 18–55, 56–100) and contact setting (household, visitor, work or study, and other) was characterised (Figs. 1 and 2). In addition, we calculated the mean and median number of total daily contacts per country (Additional file 2: Table S4), according to respondents’ gender (male or female), residence location (urban or rural), household-head education status (no education, primary, secondary, tertiary, including post-graduate), household size (0–3, 4–6, or 7+), self-reported health (“Very bad”/“Bad”, “Fair”, or “Good”/ “Very good”), past or present SARS-CoV-2 infection status (yes, no, or uncertain), perceived COVID-19 self-reported infection risk (“Low”/“Very Low”, “Medium”, or “High”/“Very high”), the self-reported impact of COVID-19 on health (“Not at all”, “Somewhat”, “Very”, or “Extremely” seriously), mask ownership ("Yes" or "No") for surveys 1 and 2, and vaccine acceptance (“Definitely”, “Probably”, “Definitely not”, or “Probably not”) for survey 2 (Additional files 3, 4, 5, 6, 7, 8, 9, 10 and 11). A permutation test was performed for the main results to determine whether the differences in the mean or median number of daily contacts per respondent were significant at the 1% to 0.01% levels or were due to chance, using the country-specific samples for each survey (Additional file 2: Table S5). Entries with missing data on any of these variables were not included in the statistical analysis. Variables with notable or surprising differences are presented in Figs. 3 and 4, while the remainder are included as additional files.

Finally, mean and median daily contacts were regressed against the change in mobility from pre-pandemic levels and the stringency of restrictions (Fig. 5 and Additional file 12: Fig. S10). For the change in mobility, publicly available national-level Google Community Mobility Reports were used, which show mobility trends in different locations (retail and recreation, groceries and pharmacies, parks, transit stations, workplaces, and residential) [27]. The dependent variable was defined as the mean or median reported number of contacts in the surveyed countries. The independent variable was defined as the mean change in mobility recorded across all locations during each survey period. No data were available for the Democratic Republic of Congo (DRC), Ethiopia, Guinea, Liberia, Sudan, and Tunisia. The Oxford Government Response Stringency Index is a composite index of the stringency scores of eight indicators (school closing, workplace closing, cancellation of public events, restrictions on gatherings, closure of public transport, stay-at-home requirements, restrictions on movement, and restrictions on international travel) (Additional file 2: Table S6) [28]. The independent variable for this analysis was defined as the mean absolute stringency index reported during each survey period. The estimated model intercepts and coefficients are reported with their standard errors and p-values.

Participation was voluntary, and data were anonymised. Interviewers explained the purpose of the study in the respondent’s preferred language, and informed consent was obtained before participation. Ipsos Mori, a market research company based in the United Kingdom (UK), conducted the surveys. Ethical approvals from the London School of Hygiene and Tropical Medicines (LSHTM) were obtained for the survey (Ref: 22486-1) and current analysis (Ref: 25945) as well as from all national ethics committees.

Twenty-three thousand nine hundred forty respondents and 563,976 contacts were analysed across the 18 countries for survey 1 and 25,640 respondents and 607,142 contacts across the 19 countries for survey 2. The percentage of people who agreed to participate out of all people randomly called (i.e. the participation rate) was on average 41% for survey 1 and 50% for survey 2 (Additional file 2: Table S7). The lowest participation rate was recorded in Morocco for survey 2 of 11% and the highest for Sudan in survey 2 of 83%. Most countries had a participation rate of at least 30%.

For survey 1, the median age of respondents ranged between 26 in Guinea and Zambia and 41 in Tunisia; between 42% (Guinea) and 52% (Sudan) were female (Table 1). For survey 2, the median age ranged between 26 in Guinea and Sudan and 41 in Morocco and Tunisia; between 45% (Kenya) and 53% (South Africa) were female (Additional file 2: Table S3). The sampled populations followed a similar age structure across countries in both surveys. Considering the absence of the 0–17 age group, adults aged over 60 were under-sampled in all countries for both surveys (Additional file 2: Table S2) [26]. Female respondents were represented well, when compared to population estimates (Additional file 2: Table S3) [26]. Between 24% (Uganda) and 58% (South Africa) resided in urban areas in survey 1 and between 21% (Uganda) and 67% (Tunisia) in survey 2. Compared to the World Urbanisation Prospects estimates (Additional file 2: Table S3) [29], the rural population was under-sampled in DRC, Ethiopia, Guinea, and Sudan and oversampled in Cameroon and South Africa in survey 1. The rural population was somewhat under-sampled in DRC and Kenya and oversampled in Nigeria and South Africa in survey 2. For the remaining countries, the rural and urban populations were represented well.

The median household size ranged between 5 and 9 in survey 1 (Table 1) and between 4 and 10 in survey 2 (Additional file 2: Table S1). In most countries, respondents had primary, secondary or tertiary education representing between 20% and 55% of the survey 1 respondents. Similar was the case for survey 2, except in Morocco and Guinea where over 35% of the respondents had no education (Additional file 2: Table S1). In Survey 1, Guinea, Kenya, Liberia, Mozambique, and Uganda, approximately 30% of respondents had an income of $0–$100. In Zimbabwe, these were 83% of the respondents. In Cameroon, DRC, Egypt, Ethiopia, Ghana, Ivory Coast, Nigeria, Sudan, Tunisia, and Zambia, those with an income of $101–$500 USD represented 40% or more of the sample. These in turn represented 35% and 28% of the samples in Senegal and South Africa, respectively. 20% of the observations for Cameroon, Mozambique, Nigeria, South Africa, and Senegal were missing income data. Reported income was not comparable across countries for survey 2 as it was recorded in the local currency.

The majority of contacts in each country occurred with contactees aged 18 to 55, representing 50% or over of total contacts for both surveys (Fig. 1). Between 20% and 40% of all contacts were with contactees aged 0 to 17. A similar trend was observed for both surveys. Contacts with those over 55 were less than 20% of total contacts in most countries. For both surveys, most contacts occurred in a household or work, school, or university settings (Fig. 2). Household contacts accounted for between 20% and 40% of all contacts. Contacts in work, school or university settings accounted for between 30% and 40%+ of all contacts. The smallest number of contacts occurred in the remaining two settings (other and visitors), except in Kenya, Mozambique, Sudan, and Zimbabwe, where other contacts accounted for the majority in survey 1. Most countries displayed a similar pattern of contacts across survey rounds (e.g. Cameroon, DRC, Egypt, Ghana, Guinea, Ivory Coast, Liberia, Senegal, South Africa, Sudan, and Zambia). However, there were some notable changes across survey rounds where most contacts occurred: from a household to a work/study setting in Ethiopia, Nigeria, Tunisia, and Uganda and from other to households in Mozambique and Zimbabwe.

For survey 1, the number of mean daily reported contacts per respondent varied greatly between countries ranging between 9 (SD=16, median = 4, IQR = 8) in Ethiopia and 41 (SD=56, median = 20, IQR = 41) in Cameroon (Additional file 2: Table S4). For survey 2, mean contacts ranged between 13 in Morocco (SD=15, median = 8, IQR = 11) and in Zimbabwe (SD=20, median = 7, IQR = 10) and 50 in Sudan (SD=53, median = 33, IQR = 40).

There was strong evidence to suggest that mean contacts increased between the surveys in Ethiopia (from 9 to 21), Ghana (from 24 to 34), Liberia (from 27 to 35), Nigeria (from 19 to 28), Sudan (from 40 to 50), and Uganda (from 16 to 34) (p-value < 0.0001). For these, the increase in the median number of contacts was also significant for all but Liberia. Mean contacts decreased from 41 to 16 in Cameroon, 28 to 23 in DRC, and 25 to 16 in Tunisia (p-value < 0.0001). Of these, the decrease in the median number of contacts was significant for Cameroon and Tunisia at this significance level. This was consistent with changes in the stringency of restrictions (Additional file 2: Table S6), except for Cameroon and DRC.

In general, male respondents reported more mean or median contants than female respondents (Fig. 3 and Additional file 2: Table S5). In survey 1, male respondents reported significantly more mean contacts than female respondents in Cameroon (48 vs 33), Egypt (29 vs 15), Ethiopia (10 vs 7), Ghana (28 vs 20), Ivory Coast (26 vs 17), Kenya (35 vs 22), Mozambique (24 vs 15), Nigeria (22 vs 16), Sudan (53 vs 29), Tunisia (32 vs 18), and Uganda (18 vs 13) (p<0.0001). Only in South Africa, no significant difference in the number of mean or median contacts was observed between the two genders. The overall trend was similar but no significant differences were observed for Ghana (mean contacts for both genders of 34) and Ivory Coast (26 vs 25) in survey 2. Male respondents in Morocco had double the contacts of female respondents (17 vs 8, p<0.0001) in survey 2. The differences in the median reported contacts for both genders were less significant in comparison, but followed the same trend.

The variation in the mean and median contacts by respondent’s location was less pronounced compared to that by gender (Fig. 4 and Additional file 2: Table S5). However, respondents in urban locations had more contacts on average than those in rural locations in Cameroon (survey 1: 47 vs 35; survey 2: 18 vs 13, p<0.0001) and Kenya (survey 1: 35 vs 26; survey 2: 28 vs 23, p<0.01). The opposite was the case in Senegal (survey 1: 17 vs 23; survey 2: 17 vs 21, p<0.0001) and Zambia (survey 2: 23 vs 30, p<0.001).

Variations in the mean and median contacts by different demographic characteristics or COVID-19 perceptions over the two surveys are further reported in Additional files 3, 4, 5, 6, 7, 8, 9, 10 and 11. There were no substantial differences across countries in the number of contacts by the household head education level (Additional file 3: Fig. S1), the respondent’s self-reported general health (Additional file 4: Fig. S2), the respondent’s past SARS-CoV-2 infection status (Additional file 5: Fig. S3), vaccine acceptance in survey 2 (Additional file 6: Fig. S4), or possession of a mask (Additional file 7: Fig. S5). Generally, respondents living in households of 7 or more members reported more non-household contacts (work/study, visitor, or other) than those in smaller households (e.g. of 1–3 or 4–6 members) (Additional file 8: Fig. S6). In some countries (e.g. Cameroon, DRC, Sudan), this applied to all contacts (Additional file 9: Fig. S7). Somewhat surprisingly, we did not find consistent differences across countries and waves in the mean number of contacts between respondents with low and high levels of perceived risk of catching SARS-CoV-2 (Additional file 10: Fig. S8), and perceived risk to health (Additional file 11: Fig. S9).

The relationship between mean daily contacts and mobility change and stringency index follows the expected pattern across survey waves (Fig. 5). Mean daily contacts increase as mobility increases and decline as the stringency of restrictions increases. In particular, an increase of one percentage point in mobility is associated with an increase of 0.57 (standard errors (SE): 0.26, p=0.0513) contacts for survey 1 and 0.28 (SE: 0.25, p=0.2910) for survey 2, though neither are significant at the 5% level. A decrease of one unit in the stringency index was associated with a decrease of 0.12 (SE: 0.11, p=0.3041) contacts for survey 1 and 0.33 (SE: 0.10, p=0.0045) for survey 2, significant at the 5% level. A similar relationship is observed between median daily contacts and mobility change and the stringency index (Additional file 12: Fig. S10). However, the magnitude of the coefficient on mobility and stringency and their standard errors decreased compared to when mean contacts were used. A potential explanation for this is that using the meadian estimate instead of the mean decreases the variation in observations in the contacts by country. This in turn leads to a more significant but potentially diluted coefficient estimate.