Background
Over 352 million cases and 5.6 million deaths from coronavirus disease (COVID-19) have been recorded worldwide as of January 24, 2022 [
1]. With the first case on the African continent reported in Egypt on February 14, 2020 [
2], Africa accounts for 3.4 and 4.2% of measured COVID-19 cases and deaths, respectively [
3]. While some countries have reported more infections than others, with South Africa, Morocco, Tunisia, Ethiopia, Egypt, Libya, Kenya and Zambia accounting for nearly 70% of all COVID-19 cases in Africa by January 24, 2022 [
3], the relatively low burden of COVID-19 in Africa compared to Europe and North America has been, in part, attributed to demographic factors such as younger and more rurally located populations, differences in case and death detection capacity, and environmental factors such as higher temperatures [
4‐
6], as well as countries’ previous experience with outbreak prone diseases [
7].
In the absence of effective pharmaceutical interventions early in the pandemic, non-pharmaceutical interventions (NPIs), or public health and social measures, were adopted to limit transmission and person-to-person contacts. African countries’ success in containing and delaying the first wave of SARS-CoV-2 infections was partially attributed to the prompt and early introduction of NPIs. However, the timing and intensity of those measures varied between countries. By March 16, 2020, when the total reported cases had reached 64, South Africa declared a national disaster, banned travel from the worst-affected countries and public gatherings, and closed schools. By March 26, 2020, when cases neared 927 and deaths were zero, the government announced a three-week national lockdown [
8]. Closure of educational institutions and travel restrictions were also implemented in Kenya, followed by a partial lockdown on April 6, 2020, with 158 cases and 6 deaths [
8]. In the second half of March 2020, a 48-h lockdown was also announced in DRC’s second largest city following the identification of two cases, whereas Senegal imposed a dusk-to-dawn curfew from March 24, 2020, with 79 cases and zero reported deaths [
9]. In contrast, Ethiopia did not introduce lockdown measures but relied on an extensive testing and screening programme combined with public awareness campaigns [
10].
Given the current challenge of ensuring sustained access to vaccines, many African countries continue to rely on NPIs, which can be economically burdensome, unsustainable, or challenging for people to adhere to. The effectiveness of many control measures can be assessed through social contact studies, which estimate the number and type of person-to-person contacts. These contact estimates are also important components of mathematical models of respiratory infectious disease dynamics [
11]. Few contact studies have been conducted in Africa, which means that models must rely on estimates from so-called synthetic matrices, which adjust high-income country contact patterns with population and household structures in low- and middle-income countries (LMICs) [
12]. Exceptions to this since the start of the pandemic are a study in informal settlements in Kenya [
13], one in KwaZulu-Natal in South Africa [
14], and one in different settings in Ethiopia [
15]. However, these are not nationally representative. This is also the case for other pre-COVID-19 contact studies—in rural coastal Kilifi in Kenya [
16,
17], rural Uganda [
18], rural Senegal [
19], in Cape Town and in a rural community and urban area in South Africa [
20,
21], and in a rural and a peri-urban site of Manicaland in Zimbabwe [
22].
We used cross-sectional data collected by the Partnership for Evidence-based Responses to COVID-19 (PERC) and described and quantified social contacts from nationally representative samples of Africa Union Member States gathered from August 4 to 17, 2020 and February 8 to 25, 2021. We compared mean and median contacts by different demographic characteristics and COVID-19 perceptions. We related contact patterns to the intensity of restrictions during the survey periods using the Oxford Government Response Stringency Index as well as Google Community Mobility Reports. We further considered how well these indicators capture changes in contacts. This paper provides one of the first comprehensive measurements of person-to-person contact behaviour across Africa during the COVID-19 pandemic.
Discussion
These data show that social contacts differ markedly between and within countries. We also find an inverse relationship between the stringency of restrictions and mean contacts, suggesting that the implementation of NPIs follows their intent. We find that contacts were the lowest in Ethiopia (survey 1) and Morocco and Zimbabwe (survey 2), and the highest in Cameroon (survey 1) and Sudan (survey 2). The majority of contacts occurred in the household or work/study settings for both surveys. Most contacts occurred with people of working age (18–55), reflecting the generally higher social interactions associated with participating in economic activities. On average, men had more contacts than women and, in most countries, mean contacts did not differ by the respondent’s area of residence, except in Cameroon and Kenya (where respondents in urban locations reported more contacts than those in rural) and in Senegal and Zambia (where the opposite was observed).
Notably, considering that the surveys were conducted when contacts were discouraged, the reported mean number of contacts is consistently higher compared to previous estimates, for the countries where prior contact data are available [
13,
15,
16,
18,
19,
21,
22]. The median number of reported contacts is closer to the past mean estimates. This could be due to a number of reasons. Firstly, while the respondent sample is close to nationally representative, respondents of working age (18–55) are slightly oversampled, potentially leading to a higher number of contacts. Additionally, many other studies report primarily on rural or semi/peri-urban communities or selected regions [
13,
16,
18,
19,
21,
22], making direct comparisons challenging. Finally, variations in the surveys’ methodologies could also be driving the observed differences—in the current study contacts were estimated rather than listed leading to reduced precision and potentially increased variability.
Other studies have also found that men have more contacts than women, reflecting gender differences in economic activities outside the home in some communities [
15,
30]. The finding that Senegalese respondents in urban areas make less contacts than in rural areas is consistent with findings of a prior Senegalese study on face-to-face contacts [
19]. Surprisingly (and in contrast to our findings), available contact studies in Kenya, South Africa, and Zimbabwe report more contacts for rural rather than urban areas [
14,
16,
21,
22]. This could be due to differences in the sampled populations and/or the definition of a contact.
We find that mean contacts increased between the two time points in Ethiopia, Ghana, Liberia, Nigeria, Sudan, and Uganda and decreased in Cameroon, DRC, and Tunisia, reflecting the opposite change in the restrictions between the two time points in all countries but Cameroon and DRC. We cannot exclude external factors that may confound this relationship. While the Cameroon and DRC findings are surprising, this could be due to, among many reasons, chance, poor quality stringency data, or because people are not responding to the restrictions.
There was a lack of significant difference in the number of contacts according to respondents’ perception of the COVID-19 risk to health in both surveys. This contrasts prior studies in high-income settings—a study in the UK reported fewer mean contacts for those who agreed that COVID-19 would have a high impact on their health than those who did not agree [
31]. However, risk perceptions might impact contact patterns differently in low-and-middle-income countries. For example, a survey in communities of Tamil Nadu, India found that the majority of respondents perceived low level of risk of contracting COVID-19, whereby more common concerns were related to loss of income, inability to travel, or getting sick [
32]. Other studies also support the finding that most contacts occur in non-household settings [
13,
18] and, when tighter restrictions are in place such as a national lockdown, the proportion of household contacts is higher than in other contact settings [
33]. More contacts of respondents in larger households than smaller ones have also been reported elsewhere [
12].
There are some methodological limitations of this study. Firstly, the contact surveys in each country were conducted retrospectively via structured telephone interviews, prone to responder recall bias and fatigue. These may have led to respondents omitting contacts or rounding down or up their number of reported contacts. An improved study design might use prospective contact diaries that record individual contacts, or structured questionnaires that aim to minimise recall bias by thoroughly guiding respondents through their day and listing all contacts [
34]. Additionally, recent technological advances in wearable proximity sensors have enabled researchers to measure epidemiologically important close contacts in resource-limited settings [
17]. This remains an important area for future research. However, such designs were not feasible for this study given that collecting contacts data was not the primary focus of the survey, but also because of the large sample size (over 1000 observations) across the 18 and 19 countries in surveys 1 and 2, respectively. The direction of the responder bias is unknown since contacts could be both rounded up or down and therefore the total contacts per country could be under or overestimated. Nevertheless, a strength of this survey, in addition to the large sample size, is that it captured a range of characteristics about the contacts and the respondents. Collecting data on multiple variables could have made a prospective design more burdensome and introduced a high rate of loss to follow-up.
In addition, participants were asked to report the number of social contacts they had, having been asked about COVID-19 transmission risk and attitudes, and we cannot rule out that social desirability bias may have affected responses. While the extent of this occurring is difficult to determine, such bias would result in respondents underreporting the number of contacts leading to overall lower average number of contacts recorded for a given country.
Thirdly, a prerequisite for participation in the telephone survey was access to a functioning telephone, connectivity, and electricity, which might be limited in some countries, particularly in rural areas. Indeed, comparing the proportion of rural and urban participants to the World Urbanisation Prospects database showed that the rural population was under-sampled in DRC, Ethiopia, Guinea, and Sudan in survey 1 and in DRC and Kenya in survey 2, which might affect the generalisability of the reported mean contacts in those countries.
Fourthly, the participation rate for the two surveys was satisfactory given the survey design of randomly digit-dialling a great number of people and given that the two samples remained balanced against the general population on observed characteristics such as gender and location. Thus, it is unlikely that this introduced a strong bias in the results. However, there may unobserved characteristics which drive people to participate and which may have introduced bias in the contacts data. If this was the case, the direction and magnitude of the bias remains unknown.
Regarding the regression analysis, this relies on a small sample size of 18–19 countries and in the case of mobility even fewer since no mobility data existed for DRC, Ethiopia, Guinea, Liberia, Sudan, and Tunisia. While the observed relationship followed the expected pattern (e.g. mean/median contacts increased with increasing mobility and decreased with increasing intensity of the restrictions across countries, albeit with weak significance), the analysis may have been underpowered to detect statistically significant patterns in the data and did not account for confounding factors. Nevertheless, the observed relationship validates that the collected data on contact patterns are of good quality.
Finally, google mobility data patterns over-represent urban areas given the limited access to mobile phone, Google maps, and internet connection in rural areas. While we found no significant differences between the number of contacts in rural compared to urban areas in most countries, the observed relationship between mobility (biased towards urban areas) and the number of mean contacts per country is likely weakened due to the inherently different sample characteristics. Nevertheless, this analysis illustrates an informative finding that there is a correlation between the number of contacts and mobility change.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.