COVID-19 has affected the African region in many ways. We aimed to generate robust information on the transmission dynamics of COVID-19 in this region since the beginning of the pandemic and throughout 2022.
For each of the 47 countries of the WHO African region, we consolidated COVID-19 data from reported infections and deaths (from WHO statistics); published literature on socioecological, biophysical, and public health interventions; and immunity status and variants of concern, to build a dynamic and comprehensive picture of COVID-19 burden. The model is consolidated through a partially observed Markov decision process, with a Fourier series to produce observed patterns over time based on the SEIRD (denoting susceptible, exposed, infected, recovered, and dead) modelling framework. The model was set up to run weekly, by country, from the date the first infection was reported in each country until Dec 31, 2021. New variants were introduced into the model based on sequenced data reported by countries. The models were then extrapolated until the end of 2022 and included three scenarios based on possible new variants with varying transmissibility, severity, or immunogenicity.
Between Jan 1, 2020, and Dec 31, 2021, our model estimates the number of SARS-CoV-2 infections in the African region to be 505·6 million (95% CI 476·0–536·2), inferring that only 1·4% (one in 71) of SARS-CoV-2 infections in the region were reported. Deaths are estimated at 439 500 (95% CI 344 374–574 785), with 35·3% (one in three) of these reported as COVID-19-related deaths. Although the number of infections were similar between 2020 and 2021, 81% of the deaths were in 2021. 52·3% (95% CI 43·5–95·2) of the region’s population is estimated to have some SARS-CoV-2 immunity, given vaccination coverage of 14·7% as of Dec 31, 2021. By the end of 2022, we estimate that infections will remain high, at around 166·2 million (95% CI 157·5–174·9) infections, but deaths will substantially reduce to 22 563 (14 970–38 831).
The African region is estimated to have had a similar number of COVID-19 infections to that of the rest of the world, but with fewer deaths. Our model suggests that the current approach to SARS-CoV-2 testing is missing most infections. These results are consistent with findings from representative seroprevalence studies. There is, therefore, a need for surveillance of hospitalisations, comorbidities, and the emergence of new variants of concern, and scale-up of representative seroprevalence studies, as core response strategies.
The COVID-19 pandemic has had a substantial impact on health, societies, and economies worldwide, including in the African region. As of Dec 31, 2021, the 47 countries of the WHO African region had reported 7·1 million infections and 155 000 deaths, accounting for 2·5% and 2·9% of the global burden of COVID-19, respectively.
When the already increased disease burden and restricted capacity of health systems across countries in the region are considered, this reported burden represents a larger problem than the numbers suggest. The burden is uneven across the region, with relative numbers of infections per 100 000 population ranging from 29·46 (Niger) to 25 061·16 (Seychelles) and deaths from 1·07 (Chad) to 151·66 (South Africa).
The variation in the number of infections within and across countries is driven by socioecological factors, and the number of deaths by biophysical factors.The socioecological factors, primarily population density, age, and hygiene, influence the rate of transmission.
In addition, public health response measures, specifically lockdowns and safe hygiene practices, contribute to minimising the role of these socioecological factors on the rate of SARS-CoV-2 transmission.
By contrast, the biophysical factors, primarily comorbidities such as hypertension, diabetes (type 2), chronic obstructive pulmonary disease, HIV, and obesity, increase the severity and risk of mortality after infection.The prevalence of these comorbidities varies in Africa, and in some instances they afflict a substantial portion of the population—up to 27% for HIV/AIDS, 22·1% for diabetes, 22·1% for chronic obstructive pulmonary disease, and 30·8% for hypertension. Health system responses, such as surveillance and effective case management, aim to mitigate against the effects of these biophysical factors. These are complemented by the use of available vaccines, which enhance antibody development and thereby reduce the severity of disease.
Research in context
Evidence before this study
The COVID-19 pandemic has had widespread effects on all countries. Globally, and particularly in the African region, the reported impact (in terms of infections and deaths) is known to underestimate the actual impact, given that countries are not able to test all suspected cases of infection during widespread community transmission. Knowledge of the true burden of COVID-19 relies on having a comprehensive mortality surveillance framework and regular seroprevalence studies, neither of which are reliably available in most countries of the African region. According to SeroTracker—a dashboard for SARS-CoV-2 serosurveys that systematically monitors and synthesises SARS-CoV-2 serological studies—as of Dec 31, 2021, 211 seroprevalence surveys (of 2812 globally) had been conducted in the WHO African region since the beginning of the pandemic. Only 11 of the 47 countries in this region have conducted seroprevalence surveys that were nationally representative, with only five countries conducting seroprevalence surveys with low risk of bias. Reporting of deaths is similarly underestimated; an assessment of health information systems shows only 10% of deaths in the WHO Africa region were being registered before the COVID-19 pandemic, compared with a global average of 62%. Reliable sources for understanding the true burden of COVID-19, therefore, remain insufficient in the WHO African region.
Added value of this study
Our findings consolidate the available data from the region, together with epidemiological evidence about COVID-19, to statistically build a picture of the true burden of COVID-19 transmission and outcomes in each country of the region. This modelling study offers an analysis of a more comprehensive picture of the progression and true burden of SARS-CoV-2 infection and COVID-19-related deaths, allowing for better understanding of the patterns so far seen in the region.
Implications of all the available evidence
These findings provide information on the variable impact of the pandemic across countries in the WHO African region. We provide an empirical basis to enable better preparedness and response to similar pandemics in the future. Moving forward, the predictions from this study could help to guide policy and practice in terms of where and how to plan response actions to the COVID-19 pandemic in the medium term. Our findings also provide a mechanism for other health programmes to generate evidence for policy when data is incomplete.
Misinformation related to the COVID-19 response activities, together with the negative socioeconomic implications of mitigation measures, has negatively affected the applicability of these interventions across countries.5 Among these, delivery of essential health services were especially impacted, with widespread disruptions to service continuity.10 Weak existing capacities for surveillance and diagnostics further contributed to the nature of the pandemic response in most countries in the region, including under-reporting of infections and deaths.11 Among and within countries in the region, the unique socioecological and biophysical factors among populations have contributed to the varying patterns in transmission and mortality observed. The introduction of vaccines (which reduce the susceptible population) and the emergence of variants of concern, which can influence transmission and disease severity, have also contributed to these unique trends.12, 13 The alpha (B.1.1.7), beta (B.1.351), delta (B.1.617.2), and omicron (B.1.1.529) variants reported in the region have had greater rates of transmission than the reference strain, driving spikes in transmission as they emerge.14 22 of the 47 countries in the region had reported the delta variant in mid 2021, which is postulated to have fuelled the observed third transmission wave and the associated increased mortality. By the end of 2021, the omicron variant dominated—accounting for 71·43% of sequenced samples by Dec 31, 2021. Variants increase the potential for immune escape, potentially leading to increased mortality as they negate protection gained from earlier infections or vaccinations (or both).15 Seroprevalence studies suggest that the reported COVID-19 infections and deaths represent a small fraction of the actual burden of COVID-19 in the WHO African region.16, 17 As of Dec 31, 2021, there were 3179 seroprevalence studies across 134 countries and territories tracked.18 Among the 47 countries of the WHO African region, 31 had seroprevalence studies, of which 11 were nationally representative and five—in Ethiopia, Kenya, Senegal, Sierra Leone, and South Africa—were at low risk of bias, as categorised in the SeroTracker.18 Looking at these countries, their reported infections at the time of the seroprevalence studies represented a small proportion (one in 50 to one in 200) of their actual burden of COVID-19.19, 20 It is, therefore, important to decipher what has happened in the region since the COVID-19 pandemic started. Other efforts have been made to understand the true COVID-19 burden; however, these have been restricted in scope to a country or subregion, based on scarce primary information, or do not reflect country-specific adaptive factors.21, 22 Consequently, the outputs produced tended to be implausible, and not correlated with other known realities in the region. Given the challenges of health information availability in the region, it is important to leverage dynamic, adaptive approaches that make use of existing primary data, and to validate these data with quality information, such as from seroprevalence studies and available mortality statistics. We aimed to present such an approach to derive best estimates of the country-specific burden of COVID-19 in the WHO African region since the pandemic began, and of anticipated patterns for 2022.
We used a standard epidemiological SEIRD (denoting susceptible, exposed, infected, recovered, and dead) compartmental model framework with modifications to include partially observed Markov processes for each of the 47 countries in the WHO African region.
Trends in the data were explored by country, using time plots, from the first infection (in each country) to Dec 31, 2021, before the model fitting. All individuals, regardless of age, were included.
First, we assumed that all the countries recorded widespread, sustained community transmission of COVID-19 and face a potential for emergence of new variants. However, due to different capacities and policies for detection and reporting, we assume that the officially reported infections and deaths only represent an observed layer of COVID-19, beneath which are unreported or undetected infections and deaths (unobserved layer). To account for these, we conceptualised the SEIRD model with modification to include a partially observed Markov decision process, which is a statistical approach whereby the hidden states are assumed to have the Markov property and emit symbols or observations. For the compartment model (SEIRD), the hidden states are the number of individuals in the compartments.
Second, the susceptible population, which was increased by the population birth rates and diminished by the population death rates, was divided into immune and non-immune states to reflect different paths for each population subgroup. The immune population was those who had been vaccinated or who had recovered from COVID-19. Equation (1) was used to model growth in vaccination rates up to the achieved vaccination coverage, and this draws from the total susceptible population.
Third, the exposed population was again subdivided into multiple states to reflect different COVID-19 variants of concern, since each has a unique epidemiology. The exposure to the different variants of concern was factored in by considering their respective incubation periods as the input into the model (appendix p 2). The overall effect in the model was a weighted average of the variants of concerns’ inverse of the latent periods, with the weights being the proportions of their prevalence.
Fourth, the individuals with COVID-19 were subdivided into observed and unobserved or hidden states using the partially observed Markov decision process, with the observed infections being a function of the force of infection for both the immune and non-immune populations, respectively.
Finally, the infected population was captured depending on the different states’ severity of illness. This fifth stage involved determining the outcome of illness and represented the final states arising from both immune and non-immune subpopulations, which was determined by the infectious periods for either subpopulation. The framework assumes that all individuals who recover will rejoin the immune subpopulation and are at risk of being exposed again, although this risk is mitigated by the reduced risk of severe disease due to immunity gained, either through vaccination against or recovery from COVID-19. Once a person enters a state, their projection is defined subsequently based on the outcome.
Figure 1 shows the model framework and different transition states. 14 Markov states are defined for either the non-immune or immune subpopulations, reflecting the two situations every individual is in. Either of these groups could be infected with a circulating variant at a given timepoint. This infection could be recorded and reported (observed) or missed (hidden) depending on the testing approaches and capacity available. The final outcome of an infection is represented by four mutually exclusive Markov states: asymptomatic, moderate, severe, or critical. These were derived using evidence from the existing evidence base (appendix p 2). We did not consider interstate transition—only the final state—due to absence of quality information on their magnitude. Each of these four states can progress to either recovery or death. All deaths from both immune and non-immune subpopulations represent an absorbing Markov state, whereas all the recovered individuals return to the immune pool to continue the cycle after a time, depending on the level of immunity accrued. The probability of transitioning from one of the compartments (states) to another was derived from published evidence and is listed in the appendix (p 2).