 Research
 Open Access
 Published:
Using outbreak data to estimate the dynamic COVID19 landscape in Eastern Africa
BMC Infectious Diseases volume 22, Article number: 531 (2022)
Abstract
Background
The emergence of COVID19 as a global pandemic presents a serious health threat to African countries and the livelihoods of its people. To mitigate the impact of this disease, intervention measures including selfisolation, schools and border closures were implemented to varying degrees of success. Moreover, there are a limited number of empirical studies on the effectiveness of nonpharmaceutical interventions (NPIs) to control COVID19. In this study, we considered two models to inform policy decisions about pandemic planning and the implementation of NPIs based on casedeathrecovery counts.
Methods
We applied an extended susceptibleinfectedremoved (eSIR) model, incorporating quarantine, antibody and vaccination compartments, to time series data in order to assess the transmission dynamics of COVID19. Additionally, we adopted the susceptibleexposedinfectiousrecovered (SEIR) model to investigate the robustness of the eSIR model based on casedeathrecovery counts and the reproductive number (R_{0}). The prediction accuracy was assessed using the root mean square error and mean absolute error. Moreover, parameter sensitivity analysis was performed by fixing initial parameters in the SEIR model and then estimating R_{0}, β and γ.
Results
We observed an exponential trend of the number of active cases of COVID19 since March 02 2020, with the pandemic peak occurring around August 2021. The estimated mean R_{0} values ranged from 1.32 (95% CI, 1.17–1.49) in Rwanda to 8.52 (95% CI: 3.73–14.10) in Kenya. The predicted case counts by January 16/2022 in Burundi, Ethiopia, Kenya, Rwanda, South Sudan, Tanzania and Uganda were 115,505; 7,072,584; 18,248,566; 410,599; 386,020; 107,265, and 3,145,602 respectively. We show that the low apparent morbidity and mortality observed in EACs, is likely biased by underestimation of the infected and mortality cases.
Conclusion
The current NPIs can delay the pandemic pea and effectively reduce further spread of COVID19 and should therefore be strengthened. The observed reduction in R_{0} is consistent with the interventions implemented in EACs, in particular, lockdowns and rollout of vaccination programmes. Future work should account for the negative impact of the interventions on the economy and food systems.
Background
Coronavirus Disease 2019 (COVID19) is a zoonotic disease caused by the Severe Acute Respiratory Syndrome Corona Virus 2 (SARSCoV2), a pathogen that was first discovered in Wuhan, China in 2019 [1,2,3]. Consequently, the disease has spread all over the world, leading to high morbidity and mortality in addition to negatively impacting the healthcare systems and the economy [4, 5]. Following the first case report in Egypt on the 14th February, 2020, a total of 6,543,882 cases and 166,234 deaths had been recorded in 54 African countries by 28th July 2021 [6]. Regionally, Eastern Africa countries (EACs) have not been spared the impact of the pandemic with the following reported cases and mortalities by the 28th July 2021: (Burundi 6573 cases, 8 deaths; Ethiopia 278,920 cases, 4374 deaths; Kenya 198,935 cases, 3882 deaths; Rwanda 66,967 cases, 771 deaths; South Sudan, 11,014 cases, 118 deaths; United Republic of Tanzania, 858 cases, deaths 29 deaths; Uganda, 92,795 cases, 2590 deaths) [6]. Initial reports indicate that majority of the early cases were likely imported from Asia and Europe through trade and tourism [7].
In this regard, policymakers made decisions to mitigate the pandemic future scenarios by implementing NPIs to varying degrees of success [8,9,10]. Similar intervention measures were successfully applied to mitigate the influenza virus transmission [11]. It is nevertheless noteworthy that the time point of implementation of these interventions is key to their success in reducing the peak of the pandemic [12]. Additionally, these measures should be appropriately justified to the population in terms of the optimal time when they could be eased [13].
As the pandemic evolved, mathematical models were developed to estimate the transmission dynamics over time, with the expectation that the pandemic will have a devastating impact across Africa [14,15,16]. For example, the University of Washington, Seattle, developed the Institute of Health Metrics (IHME) for fitting parametrized curves to COVID19 data using extendable nonlinear mixed effects model [17]. The Imperial College London (ICL), developed a semimechanistic Bayesian model to estimate the rate of transmission, total number of cases and deaths at a given time point, and the impact of NPIs on the basic reproduction number (R_{0}) as well as the timevarying reproduction number, R(t) [14]. R_{0} is a measure of contagiousness of infectious agents, and it refers to the number of new infections generated by each infected person [18]. If R_{0} < 1, the disease will decline spreading in the population, and if R_{0} > 1, the disease will spread faster [19]. Moreover, compartmental models have long been used to model the dynamics of infectious diseases including influenza [20,21,22]. These models use ordinary differential equations that mimic infectious disease trajectory, and a three or fourstate Markov chain to solve the equations [23].
Recently, the classical susceptibleinfectiousrecovered (SIR) model was extended to simulate NPIs such as quarantine, and national lockdowns using timevarying functions that modify the transmission rate of the disease [24,25,26]. The eSIR model uses three compartments—susceptible, infected, and removed (sum of recovered and dead) and a Bayesian hierarchical model to simulate future projections of the number of infected and removed population [25, 26]. The standard eSIR model assumes a constant transmission rate through the compartments. This rate can be altered to mimic NPIs by introducing a transmission modifier (π_{(t)}) to allow a timevarying probability of the transmission rate (Fig. 1).
Additionally, the eSIR model also assumes that probabilities of the three compartments follow a Markov transition process with input as the proportions of infected and removed (sum of recovered and dead) cases. The observed proportions of infected and removed cases on day t are denoted by Y_{t}^{I} and Y_{t}^{R}, respectively. The true underlying probabilities of the S, I, and R compartments on day t are denoted by θ_{t}^{S}, θ_{t}^{I}, and θ_{t}^{R}, respectively, and assume that for any t, θ_{t}^{S} + θ_{t}^{I} + θ_{t}^{R} = 1, which can be solved through ordinary differential equations (Eqs. 1–3).
whereby, β > 0 is the disease transmission rate, and γ > 0 is the removal rate. R_{0} = β/γ is the basic reproduction number assuming the whole population is susceptible. The basic eSIR model applies a BetaDirichlet statespace consisting of three observations of infected (Y_{t}^{I}), removed (Y_{t}^{R}) and the latent process at time t [25,26,27].
The latent population prevalence is represented below as a Markov process (Eq. 6).
where \({{\theta }_{t}=({\theta }_{t}^{S}, {\theta }_{t}^{I},{\theta }_{t}^{R})}^{\rm T}\) is the prevalence of susceptible population (\({\theta }_{t}^{S}\)), infectious (\({\theta }_{t}^{I}\)) and removed (\({\theta }_{t}^{R}\)) populations at time t, while \(\tau =(\beta ,\gamma ,{\theta }_{0}^{T},\lambda ,k)\) denotes parameters \({\lambda }^{I}\), \({\lambda }^{R}\) and \(k\) that control variances of the infected, removed and latent processes respectively [25, 26]. The function f is the solution to the standard SIR model using ordinary differential equations (Eqs. 1–3) and a fourth order Runge–Kutta (RK4) approximation [28, 29].
The Markov chain Monte Carlo (MCMC) algorithm was used to implement this model in order to provide the posterior estimates and credible intervals of the unknown parameters, R_{0}, β, and γ [19, 25]. Previously, Mkhatshwa et al. and Wangping et al. reported that MCMC prior distributions can be initialized according to the SARS data from Hong Kong [26, 30]. The MCMC algorithm samples the latent Markov processes and estimates the infection prevalence (\({\theta }_{t}^{I}\)) and the probability of removal (\({\theta }_{t}^{R}\)) from the underlying latent dynamics of COVID19 infection (Fig. 2) [25, 26]. These estimates determine the epidemic turning points and R_{0} of the target population. Nevertheless, it is noteworthy that the details of the eSIR model formulation are described in detail in [25, 26].
Mathematical models leverage available data to predict transmission dynamics of the epidemic and the impact of different policy interventions. These models have been a critical tool for COVID19 policy decisions [31]. However, foreseen risks include underestimation of the disease extend due to asymptomatic cases that account for the majority of the transmission [32]. Models such as the classical SIR and eSIR do not account for the presymptomatic and asymptomatic cases. Indeed, similar studies have used SEIR extensions to account for the presymptomatic and asymptomatic infection [32,33,34,35,36,37,38].
Contemporary models have consistently predicted that the ongoing COVID19 pandemic will have a devastating impact across Africa [14,15,16]. Beyond health risks, the socioeconomic implications of the pandemic motivated the current research to exploit a datadriven approach for deducing the transmission dynamics of the pandemic, infection prevention and evaluating policy implementation. This study sought to predict COVID19 epidemiological trends under current and future scenarios in seven EACs and to quantify the impact of the interventions in flattening the pandemic curve. However, we did not consider the impact of the interventions on the economy and food systems.
Methods
In this work, we applied the extended SusceptibleExposedRemoved (eSIR) compartmental model to project epidemiological trends of COVID19 infections and the impact of government interventions in Burundi, Ethiopia, Kenya, Rwanda, South Sudan, Tanzania and Uganda [25].
Data sources
We used publicly available COVID19 daily recorded timeseries data of the seven EACs collated from the WHO and the Johns Hopkins University Center for Systems Science and Engineering (JHU CCSE) to estimate the transmission of the epidemic and to present the trend of infections and fatalities [39, 40]. These datasets include daily counts of confirmed cases, recovered cases, and deaths from 22nd January 2020 to 30th July 2021.
Epidemiological modelling
Modelling the impact of NPIs was implemented in R (version 4.0.4) using the eSIR model to simulate future projections of casesdeathsrecovery counts [26]. The resulting differential equations were solved by the fourthorder Runge–Kutta approximation [28, 29]. The input data was segmented into two starting from March 02/2020 to May 01/2020 and the same time period for the year 2021. The R_{0} and R(t) were estimated using MCMC algorithm implemented in RJAGS and presented using the resulting posterior mean and 95% credible interval (CI) [41].
Model predictions were interpreted based on the turning points of the projected epidemiological trend of COVID19. The first turning point refers to the mean predicted time when the daily proportion of infected cases becomes lower than the previous infected cases, while the second turning point refers to the mean predicted time when the daily proportion of removed cases (sum of recovered and dead) becomes higher than the infected cases. Similarly, an end point refers to the time point when the median proportion of current infected cases turns to zero [25, 26]. The basic SIR model does not consider NPIs in the estimation of the epidemic trajectory, hence we used the timevarying transmission (tvt) rate SIR model, SIR with timevarying quarantine, antibody (herd immunity) and vaccination compartments to project future scenarios (Fig. 3).
SIR model with a timevarying transmission rate
A timedependent rate parameter [π(t)] was introduced to vary the transmission rate (β) and the average removal rate (γ) according to the NPIs that were introduced in each country. The prior distribution of beta (β) was set to 0.2586, and gamma (γ) was set to 0.0821 [25, 26]. The likelihood of disease transmission under the tvt rate model when a susceptible individual comes into contact with an infected individual is presented below (Eq. 7).
where, \(\pi \left(t\right) :=\left\{1{q}^{S}\left(t\right)\right\}\left\{1{q}^{I}\left(t\right)\right\}\in \left[\mathrm{0,1}\right]\). The parameter \({q}^{S}\left(t\right)\in \left[\mathrm{0,1}\right]\) is the likelihood of a susceptible individual being put under isolation. Similarly, \({q}^{I}\left(t\right)\in \left[\mathrm{0,1}\right]\) is the likelihood of an infected individual being put under isolation.
When applied as a step function, \(\pi \left(t\right)\) assumes four values that correspond to the dates when NPIs were introduced in each country. For example, the first confirmed cases of COVID19 in the seven EACs were reported in early March 2020 and thereafter, several intervention measures were introduced. For example, \({\pi }_{01}\) corresponds to the time period when no interventions had been initiated, \({\pi }_{02}\) to the time period when national/city lockdowns were initiated, \({\pi }_{03}\) corresponds to enhanced quarantine measures while \({\pi }_{04}\) corresponds to opening of new medical facilities for covid patients.
Alternatively, \(\pi \left(t\right)\) can be applied as an exponential function to reflect gradual increase in awareness to the pandemic and response to government interventions, where \(\pi \left(t\right)=\mathrm{exp}({\lambda }_{0}t)\). However, at any given time, \(\pi \left(t\right)=1\) if no intervention measure has been imposed. The function f is the solution the eSIR model using Eq. 8 and RK4 approximation.
The following parameters were used to run the tvt rate model: the transition rate modifier, π(t), was allowed to vary between (1.0, 0.9, 0.5, and 0.1) according to actual governmental interventions. This was set at π(t) = 0.95 if t < Mar 10, for city lockdown; π(t) = 0.9 if t ∈ (Mar 10, Mar 22), country lockdown; π(t) = 0.5 if t ∈ (Mar 15, April 01), shutdown of schools and nonessential businesses; π(t) = 0.1 if t > Mar 31, which corresponds to more enhanced quarantine protocols [26]. The value of \(\pi \left(t\right)\) was estimated from Eq. 7, where\(\pi \left(t\right) :=\left\{1{q}^{S}\left(t\right)\right\}\left\{1{q}^{I}\left(t\right)\right\}\in \left[\mathrm{0,1}\right]\). The proportion of deaths within the removed compartment was estimated from a preselected ratio of 0.0184 and the initial infection fatality ratio was set to 0.01 [42]. MCMC simulation was performed using four parallel chains, with the number of draws in each chain, M = 5e5 and a burnin period of nburnin = 2e3 under 2 × 10^{5} iteration number of adaptation in the MCMC (nadapt = 2 × 10^{5}) [25]. The output of these runs provided estimates of posterior parameters and prevalence of the disease in the six compartments of the modified eSIR model and proportions of the infected and the removed individuals (Fig. 3).
SIR with timevarying quarantine
We simulated the impact of quarantine measures by including a fourth compartment of the population under quarantine [25, 26]. A vector phi (ϕ) that assumes a Dirac delta function (a point mass prior at 0.1–0.4) was used to alter transition from susceptible to the quarantine compartment at time points corresponding to the days when quarantine measures were imposed in each country [20]. The timevarying quarantine model was simulated using parameters described in the tvt rate model (Eq. 9).
The quarantine rate, \(\phi \left(t\right)\), was specified according to the time points when NPIs were enforced in each country.
Herd immunity
We introduced an antibody (A) compartment to simulate the presence of natural acquired immunity against COVID19 within the population and thereby altering the eSIR to eSAIR model [26, 27]. The A compartment consists of infected (I) but recovered/selfimmunized individuals, with rate constants determining transition between the four compartments of Susceptible, Antibody, Infected and Removed (SAIR). The model was run using timevarying transmission rate parameters described above with the assumption that about 20% (α = 0.2) of the susceptible population had acquired neutralizing antibodies against SARSCoV2 (Eq. 10).
The probability of having neutralizing antibodies against COVID19 was denoted by theta (\(d{\theta }_{t}^{A}\)) at time point t, where α\(\left(t\right)\) is a function that determines the proportion of people moved into the antibody (A) compartment from the susceptible compartment.
Vaccination
The vaccination (V) compartment was integrated into the basic SIR model and thereby transforming the eSIR into eSVIR model. The model was run using timevarying transmission rate parameters under the assumption that about 2% (α = 0.02) of the susceptible population was vaccinated (Eq. 11). However, Tanzania only began their vaccination campaign in July 2021, while Burundi was yet to receive vaccine doses.
Validation of the model prediction accuracy
The reliability and usefulness of our approach, was evaluated by comparing model predictions of casedeathrecovery counts against the observed data between 06/16/20 and 04/11/2021 in Ethiopia, Kenya, Rwanda and Uganda using two metrics, the Root Mean square error (RMSE), and Mean Absolute Error (MAE) [27]. RMSE is a measure of the differences between predicted and the observed values for a given variable in a regression analysis (Eq. 12a) while MAE measures the accuracy of the model fit in terms of performance in its predictions (Eq. 12b) [43]. The input data was split into two sets for training and validation of the model [44]. The training dataset ranged from 16th June 2020 to 12th January 2021 while the validation dataset was from 13th January 2021 to 2nd April 2021. Specifically, the model was calibrated using observed data of confirmed casedeathrecovery counts (“training set”) starting from the date of implementation of the intervention up to 7–14 days prior to the peaks. Thereafter, model predictions (“testing set”) of casecounts after the training period were then compared with the observed trends to evaluate the prediction accuracy. A total of 291 datapoints were used to compute the RMSE and MAE values [43, 45].
where, n is the total number of observations, \({Y}_{i}\) the predicted value and \({\widehat{Y}}_{i}\) the observed value for the i^{th} observation.
Comparison of the eSIR and SEIR models
The eSIR model used in this study does not account for the presymptomatic and asymptomatic cases. We applied a modified SEIR extension implemented in the SEIRfansy (faLSE nEGATIVE rate and syMPTOM) package to account for the presymptomatic and asymptomatic infections [46]. Additionally, the modified SEIR model also takes into account the false negative rates of COVID19 RTPCR tests and the unreported/untested cases [46]. COVID19 casesdeathsrecovery count data reported in seven EACs from April 01/2020 to June 31/2020 was used to train the model and generate predictions from August 01 to December 01/2020. Similarly, training data from April 01/2021 to June 31/2021 was used to train the model and generate predictions from August 01 to July 13/2022. The two models were compared with respect to the predicted R(t) values, active cases and deaths.
Results
Scenario analysis of COVID19 epidemic development
Simulation of NPIs estimated the effectiveness of government intervention in curbing the spread of the disease. The predicted R_{0}, time varying reproduction number, R(t) (also known as the effective reproduction number, R_{e}) and casesdeathsrecovery counts provide an insight into the epidemiological trend of the disease for the year ending 2020/2021 (Table 1). Our results show an exponential increase of casesdeathsrecovery counts since March 02 2020/2021 and a steady decline during the 2021/2022 window. We found that the epidemic peak occurred between March–April and July–August 2021. The estimated posterior values of the timevarying reproduction number R(t) ranged between 2.70–3.10 and 1.32–8.52 under the exponential growth model for the time period of 2020/2021 and 2021/2022 respectively. The mean R(t) was lowest in Kenya in 2020/2021 (R_{0} = 2.70, 95% CI: 1.54—4.67) (Table 1). However, moving into 2021/2022 window, Rwanda had the lowest R(t) (R_{0} = 1.32, 95% CI: 1.17–1.49), (Table 2). The estimated count of infected and removed compartments by January 2022 is alarming, however it includes missed cases, presymptomatic and asymptomatic cases (Table 2).
Timevarying changes caused by government interventions
COVID19 pandemic has progressed across EACs with varying impacts. Hyperparameters introduced into the model allowed for inference of the impact of government interventions at specific time points to control the pandemic. For example, the exponential model simulated gradual community awareness of interventions by regional governments (Fig. 4) while the stepwise model simulated NPIs such as school closure, lockdowns and suspension of social gatherings for the year ending 2020/2021 and 2021/2022 (Fig. 5).
We observed an overlap of the epidemiological trend between the exponential and stepwise models. Under the existing preventions in Kenya, the exponential model indicates that the first and second turning points occurred on March 14 and April 01 during the 2020/2021 window (Fig. 4A, B) and Jul 29 and Aug 01 in the 2021/2022 window (Fig. 4C, D). Similarly, the stepwise model indicates that the first and second turning points appeared on April 01 and April 04 (Fig. 5A, B) during the 2020/2021 window and July 27 and July 31 in the 2021/2022 window (Fig. 5C, D). It is noteworthy that the first turning point refers to the mean predicted time when the daily proportion of infected cases becomes lower than the previous infected cases, while the second turning point refers to the mean predicted time when the daily proportion of removed cases (sum of recovered and dead) becomes higher than the infected cases [25, 26]. Additional exponential model projection results are available as additional information (Additional file 1: Figs. S1–S6), while the stepwise model outputs are shown for Burundi, Ethiopia, Rwanda, South Sudan, Tanzania, and Uganda respectively (Additional file 1: Figs. S7–S12).
While NPIs had a substantial impact in mitigating the pandemic, simulation of the standard SIR model without interventions indicated rampant prevalence of the infection (R_{0} > 1) and the endpoints were prolonged (Fig. 6). On the contrary, simulation using SIR with timevarying quarantine produced a decline in timevarying reproduction number due to the introduction of quarantine measures (Tables 1 and 2). With timevarying R_{0} remaining above 1, most EACs are still under threat from the disease, with Kenya (R_{0} = 8.59) facing a higher risk (Fig. 7). However, the SEIR model estimated a lower R_{0} value (R_{0} ≤ 2.49) during the same time period in Kenya (Table 4). Further projections of casesdeathsremoved counts using the standard statespace SIR model without interventions are available as Additional file 2: Figs. S13–S18). Based on our results, we observed a decline in R_{0} and the infection prevalence during 2021/2022 time period in contrast with the 2020/2021 time period, in particular, when time varying quarantine measures were introduced into the model (Additional file 3: Figs. S19–S24).
Epidemiological trends with a subset of the population having COVID19 antibodies and the impact of vaccination campaigns
Herd immunity was simulated using SIR model with a proportion of the population assumed to have neutralizing antibodies against COVID19. We observed a decline in R_{0} under the assumption that 20% of the population in EACs had achieved herd immunity by 2021/2022. Furthermore, we also simulated the impact of vaccination on the dynamics of COVID19. R_{0} declined from 8.52 to 2.14 under the assumption that 2% of the Kenyan population was vaccinated (Fig. 8). While vaccination eventually contributes to the achievement of herd immunity, our simulations showed that vaccination had a bigger impact than herd immunity in lowering the time varying reproduction number, R(t) (Additional file 3: Figs. S25–S30). Vaccination campaigns target to reduce the susceptible population and thereby lowering contacts between infectious and susceptible population [47].
Validation of the model prediction accuracy
A reliable model results in predicted values being close to the observed data values, which implies a good model fit [45]. We observed a good model fit between the forecasted cases and the actual observed cases of COVID19 across four EACs (Table 3). Larger RMSE values indicate a wider difference between the predicted and observed values, which means poor regression model fit [43]. Similarly, the computed MAE values for the model ranged between 1.24 and 10.52 (Table 3). In general, lower RMSE and MAE values provide better support for the model fit (Fig. 9).
Comparison of the eSIR and SEIR models
We generated estimates for the transmission of COVID19 using a SEIR model extension implemented in the SEIRfansy package [46]. The two models were compared with respect to the predicted R(t), active cases and deaths. SEIR model is considered superior to the standard SIR model because it takes into account the presymptomatic, asymptomatic and unreported cases as well as the high false negative rates of COVID19 RTPCR tests [46]. Moreover, the SIR model tends to overestimate the R_{0} because strict enforcement of NPIs causes the isolation of a large proportion of susceptible cases [48].
Our findings show that the mean values returned by the SEIR model corroborate the eSIR model predictions except for wide margin of variability observed in R(t) estimates in Kenya, Rwanda and South Sudan (Table 4). Additionally, the SEIR model also considered the false positive/negative rates of tests, the unreported and untested case counts (Fig. 10). Uganda had the least number untested case counts and false negative rates in contrast to other EACs (Fig. 10G). We observed less variability in R(t) values projected by the SEIR model, in contrast to the eSIR model. For example, the eSIR model estimates of the R_{0} consistently remained above 1 across most EACs with Kenya (R_{0} = 8.59) facing a higher risk. However, the SEIR model projected a lower R_{0} value (R_{0} ≤ 2.49) during the same time period in Kenya (Additional file 3: Fig. S31 C). The two models projected the peak of the pandemic to have occurred between March–April and July–August 2021. Overall, we observed a decline in R(t) values. We anticipate that the COVID19 curve has flattened for most EACs except in South Sudan and Tanzania where projected R(t) values remain high (Table 4 and Fig. 11).
Parameter sensitivity analyses
We conducted parameter sensitivity analyses to evaluate the robustness of the model to prior settings. Parameter sensitivity analyses was performed using COVID19 pandemic data between April 01 and June 30 2021. Initial parameters in the SEIR model were fixed and then followed by estimation of R_{0}, β and γ parameters using the multinomial, Poisson and Binomial models (Table 5). We observed variations in R_{0} during the first phase (01–14 April) between the Multinomial, Poisson and Binomial models. However, the estimated R_{0} values were within the same range during later stages of the pandemic [27]. Our findings show that both models are robust enough to provide reliable predictions. However, the eSIR model tends to overestimate R_{0} values while the SEIR model has less variability in R_{0} estimates [27]. Previously, Ray et al. conducted an indepth parameter sensitivity analyses using a range of scenarios in the context of the pandemic outbreak in India [49]. The reader is referred to Ray et al. for a detailed explanation of parameter sensitivity analyses [49].
Discussion
Following the first reported case of COVID19 in Egypt, the number of cases gradually increased across the continent causing human and economic losses. However, fatalities have remained low particularly during the initial phases of the pandemic. Several arguments to this observation have been put forward including experience with previous pandemics (Ebola virus disease, human immunodeficiency virus, polio, and tuberculosis), demographic factors, host genetics factors, climate and environmental factors [7]. Beyond health risks, the socioeconomic implications of the pandemic motivated many countries to implement NPIs such as wearing masks, lockdown of cities, stop transports, school closure, social distancing, and hand washing [13].
In this study, we applied the eSIR compartmental model to project epidemiological trends of COVID19 and the impact of NPIs in seven EACs [25, 26]. Publicly available data from JHU as at 30th July 2021 were used to estimate the transmission rate of the epidemic and to present the trend of infections and fatalities following government interventions [40]. Parameters such as the R_{0} and R(t) are of great importance for policy makers to adopt the most efficient and effective interventions in order to contain the pandemic and minimize human and economic damages [50].
Our findings show that the epidemic trend of COVID19 differs among EACs with infections remaining high while fatalities are low [51]. The R_{0} posterior values and endpoints in EACs during the 2020/2021 and 2021/2022 window provided a snapshot of the trajectories of the disease. However, foreseen risks include underestimation of the disease extend due to asymptomatic cases and unreported cases as well as the high false negative rates of COVID19 RTPCR tests [46]. To circumvent these risks, we applied a SEIR model implemented in the SEIRfansy package to account for the presymptomatic and asymptomatic infection and transmission of COVID19 [46]. We found that interventions that were implemented during the initial stages of the pandemic had a strong impact on reducing the transmission of the disease. For example, after calibrating the model using timeseries data from March 02/2020 to May 01/2020, our predictions revealed a modest R_{0} value of 2.71 (95% CI: 1.48–4.58), 2.75 (95% CI: 1.57–4.65), 2.70 (95% CI: 1.54–4.67), 3.10 (95% CI: 3.10–5.22), 2.71 (95% CI: 2.71–4.59), 2.82 (95% CI: 2.82–4.90), 2.87 (95% CI: 2.87–4.79) for Burundi, Ethiopia, Kenya, Rwanda, South Sudan, Tanzania, and Uganda respectively. However, R_{0} marginally decreased under the same time period in 2021/2022 projections, except in Burundi and Kenya where the value increased to a mean of 2.84 and 8.52 respectively. Previous studies of the pandemic in Kenya, reported a range of R_{0} values between 1.78 (95% CI: 1.44–2.14) to 3.46 (95% CI: 2.81–4.17) [52,53,5452,53,54]. Indeed, our findings (R_{0} = 2.70, CI: 1.54–4.67) lie within this range.
The exponential model mimicking increased community awareness of NPIs, had more impact in lowering the transmission rate of the disease than the stepwise model that mimics governmental interventions at specific timepoints. As the pandemic evolves, the public perceptions and attitudes towards the interventions change and strict adherence to public policies is practiced [13]. Moreover, the time point of implementation of NPIs is key to their success in reducing the peak of the epidemic [12]. Overall, the 2021/2022 epidemic trajectories indicate that EACs are facing challenges in their efforts to contain community transmission of COVID19. Countryspecific mean R_{0} values remain above 2 (R_{0} > 2) with the exception of Ethiopia, Rwanda and South Sudan. This is further compounded by the weak health systems, inadequate preparedness and capacity to respond to emerging epidemics. Based on these results, we suggest strict implementation of intervention policies, such as enforcement of lockdowns, facemask wearing, longterm surveillance and COVID19 vaccine rollout to contain the pandemic. However, we recommend careful interpretation of the R_{0} values because of the unforeseen risks such as underestimation of the disease extend due to asymptomatic cases and low testing rate which is not randomized.
Under the current intervention measures, the longterm projection of the eSIR exponential model indicates that about 0.97, 6.15, 33.94, 3.17, 3.45, 0.18, 6.88% of the population will be infected by 16th January 2022 in Burundi, Ethiopia, Kenya, Rwanda, South Sudan, Tanzania, and Uganda respectively. The high number of recorded cases of COVID19 could be attributed to the weak health infrastructure, crowded social life and poor personal hygiene. Moreover, disease comorbidities like hypertension, obesity, type II diabetes, HIV, tuberculosis and malaria are highly prevalent in Africa and may contribute to the weak immune response to COVID19 [7, 55, 567, 55, 56]. The comorbid individuals must be prioritized in terms of healthcare and vaccine rollout.
Previous predictive models suggested that Africa could be the next hotspot of COVID19, yet todate, recorded cases and deaths have remained low. Multiple factors have been attributed to the low COVID19 reported cases and fatalities in Africa including herd immunity due to antibodies against SARSCOV2, climate, comorbidities, parasite exposure, and young population structure [51, 57, 58]. Indeed, a recent study by the WHO revealed that over twothirds (65% or 800 million infections) of Africans were exposed to SARSCOV2 virus by September 2021 against a backdrop of 8.2 million reported cases [59]. Seroprevalence varied between countries, being highest in Eastern, Western and Central African regions. Currently, the reported seroprevalence of antibodies against SARSCoV2 range from 0.4% in Cape Verde and 49% in antenatal care clinics in Kenya [60, 61]. Despite these reports, most of these factors attributed to low mortalities have not been studied conclusively to establish their interaction with COVID19 [7]. Multiple studies have associated the low mortality rates of COVID19 in Africa to host immunity [51]. For example, the “trained immunity” hypothesis suggests that the Bacillus CalmetteGuérin (BCG) vaccine against tuberculosis confers protection against COVID19 [51]. Brewster et al. documented that Africans have genetic mutations in the SARSCoV2 receptor, angiotensinconverting enzyme2 (ACE2) gene, which confers low response to ACE inhibitors and therefore linked to low prevalence of COVID19 [62]. Furthermore, previous exposure to Plasmodium falciparum and other pathogens is associated with protective immunity and has been linked to a lower prevalence of COVID19 in malariaendemic areas [57, 63]. Additionally, the demographic structure of Africa’s population that has a predominantly young population aged below 35 years, and with few comorbidities has been linked to low prevalence to COVID19. However, such population can be super spreaders of the virus because they are largely asymptomatic [64].
We estimated the herd effect due to genetic factors and COVID19 vaccination campaigns by incorporating assumptions (about the percentage of the population with antiSARSCoV2 antibodies and the percentage of the population that had been vaccinated) into the simulation of infection dynamics. By assuming that about 20% of the population in each country had neutralizing antibodies against COVID19, we observed a significant decline in R_{0} from 8.52 to 2.62 by January 16th 2021/2022 in all the EACs. Similarly, R_{0} declined from 8.52 to 2.14 under the assumption that 2% of the Kenyan population is vaccinated. While vaccination eventually contributes to the achievement of herd immunity, vaccination had a bigger impact than herd immunity in lowering R_{0} and hence the number of cases and deaths.
During the initial phases of pandemic, the entire African population had no immunity against COVID19, hence the virus spread quickly across communities. However, as the disease evolved, gradual immunity developed aided by genetic factors, previous parasite exposure, and a young population structure with few underlying comorbidities. The COVID19 vaccine has been rolledout in Africa with 49 countries having administered at least one dose. However, the vaccination coverage required to establish herd immunity against COVID19 is quite heterogeneous, ranging from 0, 2.0, 2.2, 3.5, 0.46, 0.18 and 2.5% of the population having received at least one dose of the vaccine in Burundi, Ethiopia, Kenya, Rwanda, South Sudan, Tanzania, and Uganda respectively as of 12th August 2021 [65, 66]. Flattening the curve requires a significant percentage of population to be immunized. In particular, we recommend that countries with high basic reproduction number (R_{0} > 1) such as Kenya (8.52), Burundi (2.84), Uganda (2.34) and Tanzania (2.57) should increase vaccine coverage required to establish herd immunity against COVID19 and strictly enforce interventions. However, the current situation is further complicated by weak health systems in EACs, the inequitable vaccine distribution, vaccine hesitancy and negative perception of government interventions. Furthermore, the emergence of COVID19 variants, such as B.1.617 (“Delta”) and BA.2 Omicron variants, has led to upsurge of cases due to declining protective immunity or the circulation of immune escape viral variants [7, 67,68,69].
Epidemiological models for projecting infectious disease spread have been used to inform public health policy [22, 70,71,72]. To evaluate the reliability and usefulness of our model, we compared model predictions of the casecounts against the observed data for COVID19 in Ethiopia, Kenya, Rwanda and Uganda using the Root Mean square error (RMSE) and Mean Absolute Error (MAE). The metrics provided good support to the model fit to the observed COVID19 cases with larger values indicative of a wider difference between the predicted and observed values, hence poor model fit. The modelling techniques that we used in this study to characterize the epidemic dynamics has been successfully applied to the data in India and Wuhan, China, separately [25,26,27]. A reliable model results in predicted values close to the observed data values [73, 74].
The original eSIR epidemiology model does not provide validation of the predictions [25]. One of the novel contributions to the model was to validate the predictions made by the model using subsequent data from Ethiopia, Kenya, Rwanda and Uganda. A second additional strength was the incorporation of a vaccination compartment into the model to account for vaccineinduced immunity over time. However, we acknowledge that some aspects of these analyses have limitations. For example, the model did not account for under estimation of the reported cases, asymptomatic cases, the population structure, health systems, climate and environmental factors that can affect predictions and forecasts [14, 17, 75, 76].
Conclusions
The current intervention measures can efficaciously prevent the further spread of COVID19 and should be strengthened. However, the impact of these interventions is highly heterogeneous across EACs. Close collaboration between regional governments, the scientific community, and health care providers is required to manage the pandemic. Moreover, comparison of the basic reproduction number (R_{0}) between countries should take into consideration the under estimation of the reported cases, asymptomatic cases, demographic factors such as the population structure, health systems, host genetics factors, climate and environmental factors. The observed reduction in R_{0} is consistent with intervention measures implemented in EACs, in particular, lockdowns and rollout of vaccination programmes. Future work should account for the negative impact of the interventions to the economy and food systems.
Availability of data and materials
The data of cumulative number of COVID19 infected cases are available from COVID19 Data Repository by the Johns Hopkins University Center for Systems Science and Engineering (JHU CCSE) at https://github.com/CSSEGISandData/COVID19. The R packages used in tjis study are publicly available at https://github.com/lilywang1988/eSIR and https://github.com/umichbiostatistics/SEIRfansy.
Abbreviations
 CI:

Confidence interval
 Beta (β):

The disease transmission rate
 Gamma (γ):

The disease removal rate
 WHO:

World Health Organization
 RTPCR:

Reverse transcription polymerase chain reaction
 COVID19:

Coronavirus disease 2019
 SARSCoV2:

Severe Acute Respiratory Syndrome Coronavirus 2
 NPIs:

Nonpharmaceutical intervention(s)
 EACs:

Eastern Africa countries
 SIR:

Susceptible, infected, recovered
 eSIR:

Extended susceptible, infected, removed
 SAIR:

Susceptible, antibody, infected and removed
 SEIR:

Susceptible, exposed, infected, recovered
 R_{0} :

The basic reproduction number
 R(t):

Timevarying reproduction number
 MCMC:

Markov chain Monte Carlo
 RMSE:

Root Mean square error
 MAE:

Mean Absolute Error
 JHU:

Johns Hopkins University
 IHME:

Institute of Health Metrics, Washington, Seattle
 ICL:

Imperial College London
References
Gorbalenya AE, Baker SC, Baric RS, et al. The species Severe acute respiratory syndromerelated coronavirus: classifying 2019nCoV and naming it SARSCoV2. Nat Microbiol. 2020;5:536–44.
Hui DS, Azhar EI, Madani TA, et al. The continuing 2019nCoV epidemic threat of novel coronaviruses to global health—the latest 2019 novel coronavirus outbreak in Wuhan, China. Int J Infect Dis. 2020;91:264–6.
Lu H, Stratton CW, Tang YW. Outbreak of pneumonia of unknown etiology in Wuhan, China: the mystery and the miracle. J Med Virol. 2020;92:401–2.
Fauci AS, Lane HC, Redfield RR. COVID19—navigating the uncharted. N Engl J Med. 2020;382:1268–9.
Wang C, Horby PW, Hayden FG, et al. A novel coronavirus outbreak of global health concern. The Lancet. 2020;395:470–3.
Africa CDC. Coronavirus Disease 2019 (COVID19)—Africa CDC. Africa CDC Dashboard. 2020;2019:1–7.
Tessema SK, Nkengasong JN. Understanding COVID19 in Africa. Nat Rev Immunol. 2021;21:469–70.
Mboera LEG, Akipede GO, Banerjee A, et al. Mitigating lockdown challenges in response to COVID19 in SubSaharan Africa. Int J Infect Dis. 2020;96:308–10.
Gilbert M, Pullano G, Pinotti F, et al. Preparedness and vulnerability of African countries against importations of COVID19: a modelling study. Lancet. 2020;395:871–7.
Hagan JE, Ahinkorah BO, Seidu AA, et al. Africa’s COVID19 situation in focus and recent happenings: a mini review. Front Public Health. 2020;8:937.
Wu D, Lu J, Liu Y, et al. Positive effects of COVID19 control measures on influenza prevention. Int J Infect Dis. 2020;95:345–6.
Imai N, Gaythorpe KAM, Abbott S, et al. Adoption and impact of nonpharmaceutical interventions for COVID19. Wellcome Open Res. 2020. https://doi.org/10.12688/wellcomeopenres.15808.1.
Doogan C, Buntine W, Linger H, et al. Public perceptions and attitudes toward COVID19 nonpharmaceutical interventions across six countries: a topic modeling analysis of twitter data. J Med Internet Res. 2020. https://doi.org/10.2196/21419.
Flaxman S, Mishra S, Gandy A, et al. Estimating the effects of nonpharmaceutical interventions on COVID19 in Europe. Nature. 2020; 1–8.
Chaudhry R, Dranitsaris G, Mubashir T, et al. A country level analysis measuring the impact of government actions, country preparedness and socioeconomic factors on COVID19 mortality and related health outcomes. EClinicalMedicine. 2020. https://doi.org/10.1016/j.eclinm.2020.100464.
Jacobi L, Joshi M, Zhu D. Automated sensitivity analysis for Bayesian inference via Markov Chain Monte Carlo: applications to Gibbs sampling. SSRN Electron J. 2018. https://doi.org/10.2139/ssrn.2984054.
Murray CJ. Forecasting COVID19 impact on hospital beddays, ICUdays, ventilatordays and deaths by US state in the next 4 months. MedRxiv. 2020. https://doi.org/10.1101/2020.03.27.20043752.
Delamater PL, Street EJ, Leslie TF, et al. Complexity of the basic reproduction number (R0). Emerg Infect Dis. 2019;25:1–4.
Shetty RM, Achaiah NC, Subbarajasetty SB. R0 and re of COVID19: can we predict when the pandemic outbreak will be contained? Indian J Crit Care Med. 2020;24:1125–7.
Osthus D, Hickmann KS, Caragea PC, et al. Forecasting seasonal influenza with a statespace SIR model. Ann Appl Stat. 2017;11:202–24.
Baroyan OV, Rvachev LA, Basilevsky UV, et al. Computer modelling of influenza epidemics for the whole country (USSR). Adv Appl Probab. 1971;3:224–6.
Coburn BJ, Wagner BG, Blower S. Modeling influenza epidemics and pandemics: insights into the future of swine flu (H1N1). BMC Med. 2009;7:30.
Mbuvha R, Marwala T. Bayesian inference of COVID19 spreading rates in South Africa. PLoS ONE. 2020;15: e0237126.
Kermack WO, McKendrick AG. Contributions to the mathematical theory of epidemicsI. Bull Math Biol. 1991;53:33–55.
Wang L, Zhou Y, He J, et al. An epidemiological forecast model and software assessing interventions on the COVID19 epidemic in China. J Data Sci. 2021;18:409–32.
Wangping J, Ke H, Yang S, et al. Extended SIR prediction of the epidemics trend of COVID19 in Italy and compared With Hunan, China. Front Med. 2020;7:169.
Purkayastha S, Bhattacharyya R, Bhaduri R, et al. A comparison of five epidemiological models for transmission of SARSCoV2 in India. BMC Infect Dis. 2021;21:1–23.
Butcher JC. Runge–Kutta Methods. In: Numerical Methods for Ordinary Differential Equations. John Wiley & Sons, Ltd, 2008, pp. 137–316.
Yu X, Dai Q. The RungeKutta DG finite element method and the KFVS scheme for compressible flow simulations. Numer Methods Partial Differ Equ. 2006;22:1455–78.
Mkhatshwa T, Mummert A. Modeling superspreading events for infectious diseases: case study SARS. IAENG Int J Appl Math. 2011;41:82–8.
Choi Y, Chan AP. PROVEAN web server: a tool to predict the functional effect of amino acid substitutions and indels. Bioinformatics. 2015;31:2745–7.
Johansson MA, Quandelacy TM, Kada S, et al. SARSCoV2 transmission from people without COVID19 symptoms. JAMA Netw Open. 2021;4:e2035057–e2035057.
Lourenço J, Paton R, Ghafari M, et al. Fundamental principles of epidemic spread highlight the immediate need for largescale serological surveys to assess the stage of the SARSCoV2 epidemic. medRxiv. 2020; 2020.03.24.20042291.
Li R, Pei S, Chen B, et al. Substantial undocumented infection facilitates the rapid dissemination of novel coronavirus (SARSCoV2). Science. 2020;368:489–93.
Yuan HY, Han G, Yuan H, et al. The importance of the timing of quarantine measures before symptom onset to prevent COVID19 outbreaks—illustrated by Hong Kong’s intervention model. medRxiv. 2020; 2020.05.03.20089482.
Mizumoto K, Kagaya K, Zarebski A, et al. Estimating the asymptomatic proportion of coronavirus disease 2019 (COVID19) cases on board the Diamond Princess cruise ship, Yokohama, Japan, 2020. Eurosurveillance. 2020;25:2000180.
Subramanian R, He Q, Pascual M. Quantifying asymptomatic infection and transmission of COVID19 in New York City using observed cases, serology, and testing capacity. Proc Natl Acad Sci U S A. 2021. https://doi.org/10.1073/pnas.2019716118.
Elizabeth Halloran M, Levin BR. Infectious diseases of humans: dynamics and control (pbk edn). Trends Microbiol. 1993;1:202–3.
World Health Organization. WHO Coronavirus (COVID19) Dashboard. Who. 2021; 1–5.
Johns Hopkins University. COVID19 Map—Johns Hopkins Coronavirus Resource Center. Johns Hopkins Coronavirus Resource Center. 2020; 1.
Plummer M, Stukalov A, Denwood M. Bayesian Graphical Models using MCMC—package ‘rjags’. Comprehensive R Archive Network (CRAN), 2019.
Verity R, Okell LC, Dorigatti I, et al. Estimates of the severity of coronavirus disease 2019: a modelbased analysis. Lancet Infect Dis. 2020;20:669–77.
GraceMartin K. Assessing the fit of regression models. The Analysis Factor. 2016; 1–13.
Gholamy A, Kreinovich V, Kosheleva O. Why 70/30 or 80/20 relation between training and testing sets: a pedagogical explanation. Dep Tech Reports. 2018; 1–6.
Batool H, Tian L. Correlation determination between COVID19 and weather parameters using time series forecasting: a case study in Pakistan. Math Probl Eng. 2021;2021:1–9.
Bhaduri R, Kundu R, Purkayastha S, et al. Extending the SusceptibleExposedInfectedRemoved(SEIR) Model to handle the high false negative rate and symptombased administration of COVID19 diagnostic tests: SEIRfansy. medRxiv Prepr Serv Heal Sci 2020; 2020.09.24.20200238.
Achaiah NC, Subbarajasetty SB, Shetty RM. R0 and re of COVID19: can we predict when the pandemic outbreak will be contained? Indian J Crit Care Med. 2020;24:1125–7.
You C, Deng Y, Hu W, et al. Estimation of the timevarying reproduction number of COVID19 outbreak in China. Int J Hyg Environ Health. 2020;228: 113555.
Ray D, Salvatore M, Bhattacharyya R, et al. Predictions, role of interventions and effects of a historic national lockdown in India’s response to the COVID19 Pandemic: data science call to arms. Harvard Data Sci Rev. 2020. https://doi.org/10.1162/99608f92.60e08ed5.
Taghizadeh L, Karimi A, Heitzinger C. Uncertainty quantification in epidemiological models for the COVID19 pandemic. Comput Biol Med. 2020;125: 104011.
Ghosh D, Jonathan A, Mersha TB. COVID19 pandemic: the African paradox. J Glob Health. 2020;10:1–6.
Quaife M, Van Zandvoort K, Gimma A, et al. The impact of COVID19 control measures on social contacts and transmission in Kenyan informal settlements. BMC Med. 2020. https://doi.org/10.1186/s12916020017794.
Brand SPC, Aziza R, Kombe IK, et al. Forecasting the scale of the COVID19 epidemic in Kenya. medRxiv. 2020; 2020.04.09.20059865.
Mwalili S, Kimathi M, Ojiambo V, et al. Agestructured impact of mitigation strategies on COVID19 severity and deaths in Kenya. ResearchSquare 2020; 1–14.
Das S, Anu KR, Birangal SR, et al. Role of comorbidities like diabetes on severe acute respiratory syndrome coronavirus2: a review. Life Sci. 2020. https://doi.org/10.1016/j.lfs.2020.118202.
Ejaz H, Alsrhani A, Zafar A, et al. COVID19 and comorbidities: deleterious impact on infected patients. J Infect Public Health. 2020;13:1833–9.
Iesa MAM, Osman MEM, Hassan MA, et al. SARSCoV2 and Plasmodium falciparum common immunodominant regions may explain low COVID19 incidence in the malariaendemic belt. New Microbes New Infect. 2020;38: 100817.
Kong JD, Tekwa EW, GignouxWolfsohn SA. Social, economic, and environmental factors influencing the basic reproduction number of COVID19 across countries. PLoS ONE. 2021. https://doi.org/10.1371/journal.pone.0252373.
WHO Africa. Over twothirds of Africans exposed to virus which causes COVID19: WHO study  WHO  Regional Office for Africa, https://www.afro.who.int/news/overtwothirdsafricansexposedviruswhichcausesCOVID19whostudy (accessed 8 April 2022).
Gomez LF, Mendes C, Silva JS, et al. Seroepidemiological survey and profile of SARSCoV2 infection in cape verde. SSRN Electron J. 2021. https://doi.org/10.2139/ssrn.3762489.
Lucinde R, Mugo D, Bottomley C, et al. Serosurveillance for IgG to SARSCoV2 at antenatal care clinics in two Kenyan referral hospitals Corresponding author + Contributed equally KEMRIWellcome Trust Research Programme. medRxiv 2021; 2021.02.05.21250735.
Brewster LM, Seedat YK. Why do hypertensive patients of African ancestry respond better to calcium blockers and diuretics than to ACE inhibitors and βadrenergic blockers? A systematic review. BMC Med. 2013;11:1–16.
Anjorin AA, Abioye AI, Asowata OE, et al. Comorbidities and the COVID19 pandemic dynamics in Africa. Trop Med Int Heal. 2021;26:2–13.
Kronbichler A, Kresse D, Yoon S, et al. Asymptomatic patients as a source of COVID19 infections: a systematic review and metaanalysis. Int J Infect Dis. 2020;98:180–6.
Africa CDC. COVID19—Africa CDC. Africa CDC, 2020. https://africacdc.org/COVID19/.
Mathieu E, Ritchie H, OrtizOspina E, et al. A global database of COVID19 vaccinations. Nat Hum Behav. 2021. https://doi.org/10.1038/s41562021011228.
Zhan XY, Zhang Y, Zhou X, et al. Molecular evolution of SARSCoV2 structural genes: Evidence of positive selection in spike glycoprotein. bioRxiv. 2020. https://doi.org/10.1101/2020.06.25.170688.
Volz E, Hill V, McCrone JT, et al. Evaluating the effects of SARSCoV2 spike mutation D614G on transmissibility and pathogenicity. Cell. 2021;184:6475.e11.
Harvey WT, Carabelli AM, Jackson B, et al. SARSCoV2 variants, spike mutations and immune escape. Nat Rev Microbiol. 2021;19:409–24.
Alonso WJ, Viboud C, Simonsen L, et al. Seasonality of influenza in Brazil: a traveling wave from the amazon to the subtropics. Am J Epidemiol. 2007;165:1434–42.
Martins LD, da Silva I, Batista WV, et al. How socioeconomic and atmospheric variables impact COVID19 and influenza outbreaks in tropical and subtropical regions of Brazil. Environ Res. 2020;191: 110184.
Unwin JT, Mishra S, Bradley VC, et al. Report 23: Statelevel tracking of COVID19 in the United States. 2020. https://doi.org/10.25561/79231.
Karen GraceMartin. Assessing the Fit of Regression Models, 2016. https://www.theanalysisfactor.com/AssessingtheFitofRegressionModels/.
Scarpone C, Brinkmann ST, Große T, et al. A multimethod approach for countyscale geospatial analysis of emerging infectious diseases: a crosssectional case study of COVID19 incidence in Germany. Int J Health Geogr. 2020. https://doi.org/10.1186/S12942020002251.
Kimathi M, Mwalili S, Ojiambo V, et al. Agestructured model for COVID19: effectiveness of social distancing and contact reduction in Kenya. Infect Dis Model. 2021;6:15–23.
Mwalili S, Kimathi M, Ojiambo V, et al. SEIR model for COVID19 dynamics incorporating the environment and social distancing. BMC Res Notes. 2020. https://doi.org/10.1186/s13104020051921.
Acknowledgements
The authors would like to thank BioInnovate Africa, International Centre of Insect Physiology and Ecology (icipe) for funding that supported this research. The authors appreciate the valuable support and advice offered by Dr. Beatrice Muriithi, Chelsea Kilonzo Mbeke and Brian Kanji.
Funding
This work received financial support from the following organizations and agencies: BioInnovate Africa, International Centre of Insect Physiology and Ecology (icipe), Grant number: B8401F; The European Union; the Swedish International Development Cooperation Agency (Sida); the Swiss Agency for Development and Cooperation (SDC); the Federal Democratic Republic of Ethiopia; and the Government of the Republic of Kenya. The views expressed herein do not necessarily reflect the official opinion of the donors.
Author information
Authors and Affiliations
Contributions
HEZ Tonnang conceptualized and designed the study. MW acquired, analyzed and interpreted the data. Both authors contributed equally in drafting, reviewing of the article and final approval of the version to be published. Both authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
This is an epidemiological modelling research and therefore presents secondary data. No ethical approvals were required.
Consent for publication
Not applicable.
Competing interests
The authors have declared that no competing interests exist.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Additional file 1.
Supplementary figures. Scenario analysis of COVID19 pandemic using the exponential model (Figure S1  S6) and the stepwise model (Figure S7  S12) in Burundi, Ethiopia, Rwanda, South Sudan, Tanzania, and Uganda respectively.
Additional file 2.
Supplementary figures. Projections of COVID19 epidemic trends in Burundi, Ethiopia, Rwanda, South Sudan, Tanzania, and Uganda using the standard statespace SIR model without interventions (Figure S13 – S18).
Additional file 3.
Supplementary figures. Estimation of COVID19 epidemic trends using a timevarying quarantine model (Figure S19  S24) and scenario projection of herd immunity and vaccination campaign in Burundi, Ethiopia, Rwanda, South Sudan, Tanzania, and Uganda respectively (Figure S25  S30). Further estimates of R0 values across using the multinomial2parameter SEIR model (Figure S31).
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
About this article
Cite this article
Wamalwa, M., Tonnang, H.E.Z. Using outbreak data to estimate the dynamic COVID19 landscape in Eastern Africa. BMC Infect Dis 22, 531 (2022). https://doi.org/10.1186/s12879022075103
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s12879022075103
Keywords
 COVID19
 eSIR model
 Runge–Kutta approximation
 Basic reproduction number
 Epidemic trend