key: cord-0968018-ubzyfpqn
authors: Kolebaje, O.T.; Vincent, O.R.; Vincent, U.E.; McClintock, P.V.E.
title: Nonlinear growth and mathematical modelling of COVID-19 in some African countries with the Atangana-Baleanu fractional derivative
date: 2021-10-19
journal: Commun Nonlinear Sci Numer Simul
DOI: 10.1016/j.cnsns.2021.106076
sha: 85c303a2803d718751e3635059e19c0de60ed3e7
doc_id: 968018
cord_uid: ubzyfpqn

We analyse the time-series evolution of the cumulative number of confirmed cases of COVID-19, the novel coronavirus disease, for some African countries. We propose a mathematical model, incorporating non-pharmaceutical interventions to unravel the disease transmission dynamics. Analysis of the stability of the model’s steady states was carried out, and the reproduction number [Formula: see text] , a vital key for flattening the time-evolution of COVID-19 cases, was obtained by means of the next generation matrix technique. By dividing the time evolution of the pandemic for the cumulative number of confirmed infected cases into different regimes or intervals, hereafter referred to as phases, numerical simulations were performed to fit the proposed model to the cumulative number of confirmed infections for different phases of COVID-19 during its first wave. The estimated [Formula: see text] declined from 2.452–9.179 during the first phase of the infection to 1.374–2.417 in the last phase. Using the Atangana-Baleanu fractional derivative, a fractional COVID-19 model is proposed and numerical simulations performed to establish the dependence of the disease dynamics on the order of the fractional derivatives. An elasticity and sensitivity analysis of [Formula: see text] was carried out to determine the most significant parameters for combating the disease outbreak. These were found to be the effective disease transmission rate, the disease diagnosis or case detection rate, the proportion of susceptible individuals taking precautions, and the disease infection rate. Our results show that if the disease infection rate is less than 0.082/day, then [Formula: see text] is always less than 1; and if at least 55.29% of the susceptible population take precautions such as regular hand washing with soap, use of sanitizers, and the wearing of face masks, then the reproduction number [Formula: see text] remains below unity irrespective of the disease infection rate. Keeping [Formula: see text] values below unity leads to a decrease in COVID-19 prevalence.

Towards the end of December 2019, the infectious Coronavirus disease known as COVID-19 was first detected in Wuhan, the capital city of the Hubei province in China. Caused by the severe acute respiratory syndrome coronavirus SARS-CoV-2 [1] , COVID-19 has caused a global health emergency. The World Health Organization (WHO) declared it to be a public health emergency of international concern on 30 January 2020 [2] , and as a pandemic on 11 March 2020 [3] . By 15 June 2020, the outbreak had infected around 7.8 million people globally with total fatalities of around 430,000 people. Following Africa's first case recorded in Egypt on 14 February 2020, there had been over 246,636 confirmed cases with over 6571 deaths by 16 June 2020. COVID-19 is a rapidly spreading contagious zoonotic disease with symptoms that manifest after an incubation period of approximately 5 days following infection. Symptoms are highly variable, but range from fever, dry cough, and fatigue to less common ones like aches, sore throat, conjunctivitis, diarrhoea, and loss of smell and taste. Because efficient vaccination is not yet widely available, and there are few validated medications for treatment, COVID-19 control strategies employed by government agencies are still largely dominated by non-pharmaceutical interventions such as social distancing, wearing of face masks, regular washing of hands with soap, and use of hand sanitizer. However, the efficacy of these control strategies are not yet well-quantified, and their effectiveness is likely to change as new COVID-19 mutants take the stage.

Infectious disease modeling is a very active scientific research field. The activity is motivated, in part, by the need to gain deeper insight into disease dynamics in order to predict the trend of an epidemic outbreak, through being able to validate and test the effectiveness of control measures proposed to check the spread of the disease [4] . In recent years, mathematical modeling has been playing a key role in understanding the dynamics of infectious diseases and their control measures. It has recently been applied to study, for example Ebola [5, 6, 7, 8, 9] , Dengue fever [10, 11, 12, 13, 14] , Zika virus

The data for the countries to be analysed were collected from the daily situation reports published by the World Health Organization (WHO) [61] . The choice of countries to be analysed was made so as to ensure geographical spread, as well as relatively high population densities. Nigeria and Senegal were chosen in the West Africa zone, Ethiopia and Kenya in East Africa, Egypt in North Africa and South Africa in Southern Africa ( Figure 1) . Figure 2 shows the cumulative number of confirmed infected COVID-19 cases as a function of days since first case was recorded up till 20th June 2020 for the selected countries. Note that the values on the horizontal and vertical axes are different for each country due to the different levels of disease progression and different inception dates. The black-continuous and the blue-dashed curves in Figure 2 represent, respectively, the cubic function, α 0 + α 1 t + α 2 t 2 + α 3 t 3 and the power law function, β 1 t β2 that was fitted to the time-series. Numerical values of the fitted parameters α i , β i , i = 1, 2, 3 for the cubic and power law equation for each country are given in Table 1 . It is clear that the cubic equation fits best to the actual data for the cumulative number of confirmed cases in comparison with the power law equation of Manchein et al. [45] . However, the results of the cubic and power law fitting were comparable during the early stages of disease progression for all the countries. It is noteworthy that the fitted cubic equation for each country is such that there exists no maximum point for the curve beyond t > 25 days. Such a maximum point would correspond to a time when a maximum in the cumulative number of infections is being approached so that the curve begins to flatten. The fact that no maximum value is approached reduces the usefulness of the cubic equation for investigating future dynamics of the pandemic and possible actions needed to flatten the curve of disease progression. Nevertheless, it can be adopted as a tool for forecasting the expected number of new infections.

The cubic equation was used to predict the expected cumulative number of confirmed infected cases from June 21 to June 30 with Root Mean Square Error (RMSE) and Mean Percentage Error (MPE) used as performance indicators for the prediction ( Table 2) where Obs. and Est. are, respectively, the observed and estimated values of the cumulative number of confirmed infected cases and n is the number of observations used. In general, the lower the RMSE and MPE, the better the model. Table 2 , we observe that the lowest and highest RMSE were obtained for Ethiopia and South Africa respectively and the MPEs obtained were of magnitude between 1.022% − 4.290% of the actual cumulative number of confirmed infected cases. Cumulative number of confirmed infected cases with COVID-19 as a function of time from inception (first case) for South Africa, Egypt, Nigeria, Senegal, Ethiopia and Kenya. The black-continuous and the blue-dashed curves represent respectively the functions α 0 +α 1 t+α 2 t 2 +α 3 t 3 and β 1 t β2 that fit the actual time-series represented by the red dots. The parameters α i , β i , i = 1, 2, 3 for each country are described in Table I.  Table 1 Parameters of the cubic and power-law fitting curves. 

Here, we propose a new epidemiological model for the COVID-19 epidemic. The proposed model is an extended form of the well-known Susceptible Exposed Infected Recovered (SEIR) compartmental model that takes into account some features such as quarantine, isolation and asymptomatic infections, commonly employed in epidemiological studies of communicable diseases such as, Ebola, Zika, COVID-19, etc [62, 63, 64] . An asymptomatic transmission refers to transmission of the virus through a person, who does not develop any symptoms despite having been infected. The model contains seven epidemiological compartments namely: Susceptible S(t), Exposed E(t), Infected I(t), Asymptomatic I A (t), Quarantined Q(t), Hospitalized H(t) and Recovered R(t). The complete flow chart of the interactions between different classes of the proposed model is shown in Figure 3 . The susceptible population S(t) represents the totality of the entire population that is at risk of being infected with the virus. This population is assumed to be increasing at a constant rate Ω. The increase is not a net increase because µ is the natural death rate common to all the classes of the population. Exposure and transmission of the virus to the susceptible population involves the action of individuals in the infected class I(t) and the asymptomatic class I A (t). We assume that the infected class consists of people that develop symptoms while the asymptomatic class I A (t) involves people that are without symptoms and therefore unaware of their 7 J o u r n a l P r e -p r o o f positive COVID-19 status. β denotes the rate of disease transmission with α representing a measure of the relative (reduced) effectiveness of individuals in the asymptomatic class as disease spreaders. The spread of the disease to the susceptible population can be controlled by several precautionary measures such as use of soap and sanitizers, lockdown, social distancing and use of face masks and other Personal Protective Equipment (PPE). We assume that h (0 < h < 1) represents the portion of the population that maintains these precautions with the disease only transmitted to (1 − h) portion of the susceptible population. θ is the infection rate for the model with pθ and (1 − p)θ the portions of the exposed class E(t) that go into the infected class I(t) and asymptomatic class I A (t), respectively. The quarantine class Q(t) involves the quarantining of exposed individuals usually through contact tracing, at a rate η 1 and people who develop symptoms in quarantine are also hospitalized at a rate ρ 1 . As the current procedure in most African countries is to limit testing mostly to people who develop COVID-19 symptoms, we assume that the infectious and symptomatic class I(t) are tested at a rate η 2 (diagnosis or detection rate of infected symptomatic individuals) and moved into the hospitalized class H(t). Also, quarantined individuals who develop symptoms are moved to the hospitalized class H(t) at a rate ρ 1 while those who do not develop symptoms after 1/ρ 2 days are back into the susceptible population. Let δ 1 and δ 2 be respectively, the recovery rate of the isolated/hospitalized infected population H(t) and untreated asymptomatic population I A (t) into the recovered population R(t). γ 1 and γ 2 denote the COVID-19 induced death from the hospitalized and asymptomatic class, respectively with γ 2 usually very small compared to γ 1 . µ is the natural death rate common to all the classes of the population. We assume, based on current scientific evidence, that the COVID-19 deceased are not infectious, and that individuals develop antibodies and become immune to the disease once they are recovered.

On the basis of the above assumptions, the nonlinear system of differential equations describing the COVID-19 model used to analyzed data from African countries can be written mathematically as:

with the initial conditions S(0) > 0, E(0) ≥ 0, I(0) > 0, I A (0) ≥ 0, Q(0) ≥ 0, H(0) ≥ 0 and R(0) ≥ 0. At every instant of time, the quantity D(t) = γ 1 H(t) + γ 2 I A (t) represents the number of deaths caused by the disease at time t while C(t) = η 2 I(t) + ρ 1 Q(t) + δ 1 H(t) represents the total number of confirmed COVID-19 cases at time t. Furthermore, let the total size of the population be N(t) = S(t)+E(t)+I(t)+I A (t)+Q(t)+H(t)+R(t). Definitions of the model parameters (1) are presented in Table 3 . Table 3 Description and unit of model parameters. 

where

Eq.

(3) means that the solution of system (1) for S(t) is positive. Similar expressions for E(t), I(t), I A (t), Q(t), H(t) and R(t) can be obtained from system (1) as

J o u r n a l P r e -p r o o f Journal Pre-proof

where

This completes the proof.

Theorem 2 All solutions of system (1) that initiate in ℜ 7 + are bounded uniformly in the region

, H(t), R(t)) be any solution of system (1) with any given non-negative initial condition. Also, let

Thus, by the differential inequality theory [65] , we obtain

It thus follows that, for t → ∞,

This implies that, χ is positively invariant so that all solutions of (1) with initial conditions in ℜ 7 + are confined in χ.

Furthermore, the interacting functions f i (S, E, I, I A , Q, H, R), i = 1 − 7 of the system (1) are continuous and have continuous partial derivatives on ℜ 7 + . Hence, they are Lipschitzian on ℜ 7 + . Additionally, Theorem 2 implies that the solutions of Eq. (1) with initial conditions in ℜ 7 + are uniformly bounded. Therefore, the initial value problem (IVP) is well posed.

Here, we employ the next generation matrix technique to determine the basic reproduction number R 0 , representing the number of secondary infections caused by a single infected individual in the entire duration of their infection [59, 60] . The classes which are directly involved in the spread of disease are E, I, I A and Q. Therefore, from the system equation (1), we obtain the reduced system:

In compact matrix form, system (1) can be written as:

where

In epidemiology, the matrix F is referred to as the matrix of new infections and V is the transfer matrix of individuals between compartments. The transition matrices V and F are obtained from the partial derivatives of V and F with respect to E, I, I A and Q, evaluated at the disease-free equilibrium

We define the next generation matrix by F V −1 , where the reproduction number R 0 is given by the spectral radius of F V −1 [60] .

The reproduction number R 0 obtained for the model is the sum of two other reproduction numbers R A 0 and R B 0 , where

,

R A 0 represents the number of secondary infections caused by an infected individual during their time spent in the infected population. It is a measure of the number of the (1 − h)Ω/µ susceptible population that are infected 14 J o u r n a l P r e -p r o o f by θp people in the infected group with a bilinear transmission rate β, with 1/(µ + η 2 ) being the time an infected individual remains in the infected group and 1/(µ + θ + η 1 ) being the time an individual remains in the exposed group. R B 0 represents the number of secondary infections because of an asymptomatic individual during their time spent in the asymptotic group. It represents the number of the (1 − h)Ω/µ susceptible population that are infected by (1 − p)θ people in the asymptomatic group with an enhanced transmission rate αβ with 1/(µ + δ 2 + γ 2 ) being the time an individual remains in the asymptomatic group.

The equilibrium points of the model system (1) are obtained by setting the interacting functions f i (S, E, I, I A , Q, R), i = 1 − 7 = 0. The disease-free equilibrium is given by X 0 (Ω/µ, 0, 0, 0, 0, 0, 0), while the endemic equilibrium is given by

,

with ∆ 1 = ρ 2 η 1 /(µ + ρ 1 + ρ 2 ) and ∆ 2 = η2pθ µ+η2 + ρ1η1 µ+ρ1+ρ2 . We find that the disease-free equilibrium X 0 always exists, whereas, the endemic equilibrium X 1 is only feasible if R 0 > 1 and (µ + θ + η 1 ) > ∆ 1 .

Here, we discuss the local asymptotic stability criteria of the equilibria of system (1) by evaluating the Jacobian or community matrix and the resulting characteristic equation. We then examine the signs of the eigenvalues based on the Routh-Hurwitz conditions and/or Descartes rule of sign. It is easy to 15 J o u r n a l P r e -p r o o f show that the Jacobian of system (1) is given as:

Proof 3 The Jacobian matrix of system (1) evaluated at the disease-free equilibrium, X 0 is given by

where,

The characteristic equation of model system (1) evaluated at the disease free equilibrium point, X 0 , is given by

where,

It is clear from equation (18) that the four eigenvalues, ω 1 , ω 11 , ω 14 and ω 17 have negative values and the remaining eigenvalues can be easily obtained by finding the roots of the cubic polynomial in Eq. (18) . Applying the Routh-Hurwitz criteria on the cubic polynomial in (18) requires that A > 0, C > 0 and AB > C for the other three eigenvalues to be negative or have negative real parts.

J o u r n a l P r e -p r o o f

Hence, the Routh-Hurwitz criterion is satisfied if, R 0 < 1 and we may conclude that the COVID-19 model (1) is locally asymptotically stable at the free equilibrium point, X 0 .

Proof 4 The Jacobian matrix of system (1) evaluated at the endemic equilibrium, X 1 is given by

where,

,

J o u r n a l P r e -p r o o f

The characteristic equation of system (1) evaluated at the disease-endemic equilibrium, X 1 , is given by

where a 0 = ω 5 ω 8 ω 10 (−ω 4 ω 11 + ω 6 ω 12 ), a 1 = −ω 5 (ω 8 ω 10 ω 12 + ω 6 ω 8 ω 10 + (ω 8 + ω 10 )(−ω 4 ω 11 + ω 6 ω 12 ))

−ω 8 ω 12 (ω 6 + ω 10 ) + ω 5 (ω 8 ω 10 + (ω 6 + ω 12 )(ω 8 + ω 10 ))

It is clear from equation (20) that the two eigenvalues, ω 15 and ω 18 are negative, and that the remaining five eigenvalues are roots of the quintic polynomial in (20) . Using R 0 > 1, ω 8 − ω 10 = δ 2 + γ 2 − η 2 > 0 and −ω 4 ω 11 + ω 6 ω 12 > 0 guaranteed by the feasibility condition (µ + θ + η 1 ) > ∆ 1 of the endemic equilibrium, we observe that a i , i = 1 − 5 > 0. By employing the Descartes' rule of signs, we find that the number of positive eigenvalues of the quintic polynomial given in Eq. (20) is equal to the number of sign-changes from a 5 to a 1 , which equals zero. Hence, the system (1) is locally asymptotically stable if R 0 > 1 and δ 2 + γ 2 − η 2 > 0.

Theorem 5 The disease-free equilibrium X 0 (Ω/µ, 0, 0, 0, 0, 0, 0) is globally asymptotically stable if R 0 < 1 and unstable if R 0 > 1. 

SubstitutingĖ andİ from system (1) into Eq. (21) yields

Choosing κ 1 = pθ, κ 2 = (µ + θ + η 1 ) and inserting S = Ω/µ and I A = 0, we have

Since R A 0 < 1 follows from R 0 < 1, therefore, it is clear that dL/dt < 0, when R 0 < 1 and also dL/dt = 0, if I = 0. Hence, by LaSalle's Invariance principle [66, 67] , the disease-free equilibrium X 0 is globally asymptotically stable.

Theorem 6 If R 0 > 1, then there exist a disease-endemic equilibrium X 1 and it is globally asymptotically stable in the interior of χ.

Proof 6 Given that R 0 > 1, then the existence and local asymptotic stability of the disease-endemic equilibrium is guaranteed. Consider the Lyapunov function

The time derivative of L is given bẏ

J o u r n a l P r e -p r o o f

However, in the endemic state, we have

Then using Eq. (26) in Eq. (25), we havė

Because E ≤ E * , I ≤ I * , I A ≤ I * A , Q ≤ Q * and H ≤ H * , Eq. (27) then becomeṡ

It follows from arithmetic-geometric inequality that

Therefore,L ≤ 0 and alsoL = 0, only if E = E * , I = I * , I A = I * A , Q = Q * and H = H * . Hence, by LaSalle's Invariance principle [66, 67] , the endemic equilibrium X 1 is globally asymptotically stable.

We now carry out numerical simulations to compare our proposed COVID-19 Africa model (1) to the data for cumulative number of confirmed infected cases obtained from the World Health Organization (WHO) [61] for South Africa, Egypt, Nigeria, Senegal, Ethiopia and Kenya. The starting point of our simulation will be a day before the index case was recorded in each country. The demographical parameters Ω and µ are estimated from the total population size (N 0 ) and life expectancy (L.E.) data obtained from the 2018 United Nations data bank [68] . For example, Nigeria has a life expectancy of 54.332 years with population size estimate of 200, 963, 599. Hence, the average death rate, (µ), used for the simulation will be 1/(54.332 × 365) = 5.04 × 10 −5 /day with constant population growth rate, Ω = µN 0 = 10, 128.57. We assume that the proportion of new infections that are symptomatic, (p), is 60% and the portion of the susceptible population taking precautionary measures, (h), is 30% for all the countries except 40% used for South Africa and Egypt [69, 70] . The relative infectiousness of the asymptomatic class, (α), is set to 0.5 [40, 71] . It is assumed that individuals in quarantine, either develop symptoms after an average of 7 days and are moved to isolation/hospitalization at a rate ρ 1 = (1/7) = 0.143/day or are released from quarantine, after an average of 14 days without developing symptoms, into the susceptible population at a rate ρ 2 = (1/14) = 0.0714/day. The average remission time is set to 14 days and 7 days for individuals in the hospitalized and asymptomatic classes respectively. Hence, the recovery rates δ 1 = 0.0714/day and δ 2 = 0.143/day were used in the model simulations. The COVID-19-induced death rates (γ 1 , γ 2 ) are estimated from the percentage of case fatalities recorded. The remaining parameters, the effective disease transmission rate (β), infection rate (θ), quarantine rate of exposed individuals (η 1 ) and diagnosis/case detection rate (η 2 ) 21 J o u r n a l P r e -p r o o f are obtained from fitting the model to the data using the NonlinearModelFit function in Mathematica. Table 4 gives the values of the parameters used in the simulations. The following values of the initial conditions were also used for all the simulated countries: S(0) = N 0 , E(0) = 0, I(0) = 1, I A (0) = 0, Q(0) = 0, H(0) = 0 and R(0) = 0. Moreover, the simulations and parameter estimations were performed such that new initial conditions and new values of fitting parameters were obtained whenever the percentage daily increase in cases was more than 30%. This approach divided the time evolution of the pandemic for the cumulative number of confirmed infected cases into different regimes or intervals, which we refer to as phases. (Note that this definition of phase is quite distinct from an interval of constant conditions determined by e.g. a particular "Tier" or level of lockdown restrictions as used in the UK.) South Africa, Egypt, Nigeria and Kenya all have two phases of the infection, and Senegal has three phases, while Ethiopia has only one phase of the infection. Table 5 gives the estimated values of the parameters β, θ, η 1 and η 2 , as well as the calculated reproduction number R 0 for the different phases of the infection for all the countries. Figure 4 gives the results of the numerical simulations for the different phases of the infection for all the countries. From Figure 4 , we observe that our model (1) was well-fitted to the actual data for cumulative number of confirmed infected cases for all the countries, with values of R 0 between 1.374 and 9.179 and R 0 highest during the first phase of the infection for all the countries. The observed decrease in the R 0 values beyond the first phase may be due to the impact of the strict lockdown measure and other preventive policies enforced by the authorities. J o u r n a l P r e -p r o o f Here, we analyse the elasticity and sensitivity of the reproduction number, R 0 . Sensitivity analysis is a well known technique for identifying the critical parameters or inputs of a model and quantifying their importance relative to one another [58] . For the purpose of elasticity and sensitivity analysis, we used data from Nigeria as a case study. The baseline values and ranges of the system parameters used here are given in Table 6 . In order to perform the elasticity analysis of R 0 , we first calculated the normalized forward sensitivity index. In general, the elasticity (normalized forward sensitivity index) of a variable, u, that depends differentiably on a parameter, ϕ, is given as [72] :

A negative (or positive) sensitivity index indicates a decrease (or an increase) in the value of the parameter ϕ resulting in a decrease (or an increase) in the value of u. For, R 0 , we obtain the following:

J o u r n a l P r e -p r o o f

The elasticity or sensitivity index shows the influence of change in one parameter while keeping all other parameters constant. Note that the sensitivity indices of Ω and β do not depend on any parameter value. Interestingly, the reproduction number R 0 does not depend on the recovery rate of the hospitalized individuals (δ 1 ), or on the isolation rate of quarantined individuals (ρ 1 ), or on the transition rate from quarantine class to susceptible class after quarantine (ρ 2 ), or on the COVID-19-induced death rate of the hospitalized cases (γ 1 ). We proceed to evaluate the above sensitivity indices using the baseline parameter values in Table 6 . A plot showing the sensitivity indices for R 0 with respect to its constituent parameters is presented in Figure 5 , from which we can state that R 0 increases whenever Ω, β α, θ, or p increase. On the other hand, whenever µ, h, η 1 , η 2 , δ 2 or γ 2 increase, then R 0 decreases. For example, ̟ R0 α = 0.271 implies that increasing α by 10% will increase R 0 by 2.71%. Hence, a similar interpretation can be inferred for the remaining parameters in Table 6 . Following the elasticity analysis, the most positive sensitivity index was obtained for β with a 10% increase in β leading to the same proportional increase in R 0 . Note that the same value of positive sensitivity index as β was obtained for Ω. However, Ω is a demographic variable that cannot be changed in the field. The controllable parameter with the most negative sensitivity index was η 2 . With a 10% increase in η 2 , a decrease of 7.29% in R 0 was found. We observe that the most significant parameters are the effective disease transmission rate (β), the disease diagnosis or case detection rate (η 2 ), and the proportion of susceptible individuals taking precautions (h). Therefore, we can conclude that efforts at controlling the disease should concentrate on decreasing the transmission rate through contact reduction via lockdown and social distancing measures. In addition, h can be increased by encouraging social campaigns aimed at increasing the number of individuals taking precautionary measures such as regular hand washing with soap and use of sanitizers, use of face mask and other PPEs.

For the sensitivity analysis of the reproduction number, R 0 , the Latin Hypercube Sampling (LHS) technique [73] was implemented for the parameters of the model. The correlation between R 0 and the input parameter values obtained from 1000 simulations was quantified using Partial Rank Correlation Coefficients (PRCC). The sensitivity analysis results of R 0 , shown in Figure 6 demonstrate that the parameters with the greatest influence on R 0 are β, η 2 , h, δ 2 , and θ in order of decreasing sensitivity. Clearly, the sensitivity analysis results are similar to those of the elasticity analysis, as the parameters β, η 2 , and h, identified via sensitivity indices as being the most significant parameters, are also the most sensitive parameters. It is noteworthy that, though the demographic parameters, Ω and µ had sensitivity indices of 1 and −1, respectively, the results of the sensitivity analysis shows that they have insignificant influence on R 0 due to their very low PRCC values. Figure 7 and 8 provide contour plots of the reproduction number R 0 as a function of the parameter pairs (β, η 2 ) and (θ, h), respectively. From Figure 7 , we observe that for an effective disease transmission rate, β < 6.839 × 10 −10 , R 0 is always less than 1, indicating that, eventually the disease will die out in the population regardless of the value of the diagnosic or case detection rate (η 2 ). For increase in the value of β, R 0 can be kept under 1 by also increasing η 2 . If β > 1.368 × 10 −9 and η 2 < 0.187/day, then R 0 > 2. Also, if β > 2.052 × 10 −9 and η 2 < 0.109/day, then R 0 > 3. Figure 8 shows that if the disease infection rate, θ is less than 0.082/day, then R 0 is always less than 1 regardless of the proportion of susceptible individuals taking precautions (h). Another remarkable observation from Figure 8 is that, if at least 55.29% of the susceptible population adheres to the prescribed precautions, such as regular hand washing with the use of soap, use of sanitizers and face masks, then the reproduction number R 0 can be kept below unity regardless of the value of the disease infection rate. This result is in agreement with previous studies carried out for Lagos, Nigeria, where at least 55% of the population effectively making use of face masks while in public was recommended for the reproduction number of the disease to be brought below 1 [40] . Moreover, if less than 10.58% of the susceptible population take the prescribed precautions with a disease infection rate greater than 0.483, then R 0 > 2. Figure 9 shows the influence of the variation in the model parameters on the progression of the number of active cases. The plots in Figure 9 were obtained by simulating the model (1) numerically using different values of the system parameters, β, η 2 , θ and h, while other parameters were kept constant. We observe that variations in the parameters have significant influence on the maximum of the infection and the number of days taken to reach this maximum. Notably, for an increase in the values of β and θ, there is a corresponding increase in the maximum infection in addition to this value being attained on a later day. For infection rates of 0.4, 0.5, 0.6 and 0.7, maxima of about 850, 000, 750, 000, 610, 000 and 460, 000 may be reached after about 280, 310, 350 and 420 days, respectively. For diagnostic or case detection rates of 0.10, 0.13, 0.16 and 0.18, maxima of about 1.5 million, 1 million, 590, 000 and 370, 000 may be reached after about 220, 280, 370 and 460 days, respectively. If 15% of the susceptible population takes precautions, then the projection in Figure 9 shows that the number of active cases may attain about 1. precautionary measures, then the number of active cases may reach around 320, 000 by around the 510 th day of the infection.

Fractional models with the Atangana-Balenau fractional derivative provide more efficient results than the ordinary derivative models [46, 47, 51, 57, 74, 75, 76, 77, 78] . We begin here by defining the Atangana-Balenau fractional derivative and its integral. a 1 , a 2 ) , a 2 > a 1 , σ ∈ [0, 1],. The Atangana-Balenau fractional derivative is then defined [46] as: 

where B(σ) is the normalization function satisfying B(0) = B(1) = 1, and E σ (.) is the Mittag-Leffer function with one parameter.

The Mittag-Leffler function with one parameter is defined [46] as We generalize the model (1) by applying the Atangana-Baleanu derivative and thus obtain the following fractional COVID-19 model for Africa: 

where D σ t is the fractional derivative and σ represents the fractional order parameter.

The numerical results for the fractional model (34) were obtained by following the procedure described in the Appendix Section 8 in which the modified Adams-Bashforth scheme developed by Toufik and Atangana [78] was adopted. Readers are referred to some very recent applications of the modified Adams-Bashforth scheme [47, 48, 49, 79] . Figure 10 shows the dependence of the number of active cases, infectious and symptomatic class, asymptomatic class and the hospitalized class from the fractional model (36) on the magnitude of the order of the fractional derivative, σ. Fig. 10 . The dynamics of the active cases, infectious and symptomatic class I(t), asymptomatic class I A (t) and hospitalized class H(t) of the fractional model (36) for different values of the order of the fractional derivative σ. model (36) using the modified Adams-Bashforth scheme (39) -(45) for σ = 1.0, 0.95 and 0.9. The result shows that the magnitude of σ has a marked impact on the day the maximum is reached, with a right shift in the time at which this happens as σ decreases from 1.0. However, the order of the fractional derivative (σ) has only an insignificant effect on the projected peak numbers of active cases. Specifically, for σ = 1.0, 0.95 and 0.9, the peak numbers of active cases were approximately 590, 000, 570, 000, 550, 000 by about the 370 th , 440 th and 540 th day after the first case of the infection was recorded, respectively. Hence, the order of the fractional derivative (σ) can be used as an effective delay variable for the peak of the infection.

The results of the numerical simulations showing the effect of the order of the fractional derivative σ on the cumulative number of confirmed infected cases by COVID-19 for the different phases of the infection for all the countries are presented in Figure 11 .

The effective reproduction number R e (t), defined as the actual average number of secondary cases per primary case at time t (for t > 0) is a useful timevarying threshold in epidemiology for measuring the trajectory and rate of spread of the disease at any point in time during the course of the epidemic. The effective reproduction number for Eq. (1) is given by In general, the number of disease cases rises when R e (t) > 1, attains a peak when R e (t) = 1, and declines when R e (t) < 1 [80, 81, 82] . Figures 12 and  13 illustrate the impact of different orders of the fractional derivative σ on the effective reproduction number R e (t). It is evident from Figures 12 and  13 that the COVID-19 epidemic had an effective reproduction number that generally decreases with time for the different phases of the infection for all the countries. In addition, the effective reproduction number falls more rapidly with time for σ = 1.00 and less so for σ = 0.95. This observation is consistent with Figure 10 which shows that the peak number of active cases was reached earliest for σ = 1.00.

In this paper, a mathematical model for the novel coronavirus (COVID-19) disease which incorporates some non-pharmaceutical interventions was pro- posed and used to investigate the transmission dynamics in selected African countries, namely, South Africa, Egypt, Nigeria, Senegal, Ethiopia and Kenya. The model contains seven epidemiological compartments namely: Susceptible S(t), Exposed E(t), Infected I(t), Asymptomatic I A (t), Quarantine Q(t), Hospitalized H(t) and Recovered R(t) and also takes into account some specific features of the COVID-19 epidemic such as quarantine, isolation and asymptomatic infections. We obtained the critical points, identified the disease-free states and the endemic states. The basic reproduction number, R 0 was computed using the next generation matrix approach. Numerical simulations to fit the proposed model to the actual data for cumulative number of confirmed infected cases was performed for the different phases of the infection for all the countries, with values of R 0 between 1.311 and 9.179 obtained. Notably, the condition R 0 < 1 is necessary for the stability (or instability) of the diseasefree(or endemic state). A fractional version of the model was introduced using the Atangana-Baleanu derivative and numerical simulations were performed for better understanding of the dependence of the dynamics of the disease on the order of the fractional derivative, σ. The result shows that the magnitude of σ has a pronounced effect on the day the maximum is reached with a right shift observed in the time taken for the maximum to be attained as σ decreases from 1.0. However, the order of the fractional derivative has insignificant effect on the projected peak number of active cases. Hence, σ can be used as an effective delay variable for the peak of the infection. Elasticity and sensitivity analyses show that the most significant parameters are the effective disease transmission rate (β), disease diagnosis or case detection rate (η 2 ), proportion of susceptible taking precautions (h) and the disease infection rate (θ). If the disease infection rate, θ is less than 0.082/day, then R 0 is always less than 1 regardless of the proportion of susceptible taking precautions (h). Another remarkable inference from the study is that, if at least 55.29% of the suscep-

J o u r n a l P r e -p r o o f tible population take precautions such as regular hand washing with the use of soap, use of sanitizers and wearing of face masks, then, the reproduction number R 0 can be kept below 1 irrespective of the value of the disease infection rate. The most important practical conclusion is that efforts to control the disease should concentrate on decreasing the transmission rate by contact reduction, via lockdown and social distancing measures. In addition, h can be increased by encouraging social campaigns aimed at increasing the number of individuals taking prescribed precautionary measures.

Naming the coronavirus disease (COVID-19) and the virus that causes it

WHO, Statement on the second meeting of the international health regulations (2005) emergency committee regarding the outbreak of novel coronavirus (2019-ncov)

Director-General's opening remarks at the media briefing on COVID-19 -11

Preliminary estimation of the basic reproduction number of SARS-CoV-2 in the Middle East

Mathematical modeling of Ebola Virus dynamics as a step towards rational vaccine design

Mathematical modeling of the West Africa Ebola epidemic

Modeling the Ebola zoonotic dynamics: Interplay between enviroclimatic factors and bat ecology

A fractionalorder model for Ebola Virus infection with delayed immune response on heterogeneous complex networks

Transmission dynamics and control of Ebola Virus disease (EVD): A review

Dengue fever: Mathematical modelling and computer simulation

The incubation periods of Dengue viruses

Assessing the potential of a candidate Dengue vaccine with mathematical modeling

The impact of the demographic transition on Dengue in Thailand: Insights from a statistical analysis and mathematical modeling

A fractional calculus based model for the simulation of an outbreak of Dengue fever

Mathematical modeling of Zika virus

Mathematical modeling, analysis and simulation of the spread of Zika with influence of sexual transmission and preventive measures

Mathematical modeling of within-host Zika virus dynamics

Dynamics of Zika virus outbreaks: an overview of mathematical modeling approaches

Mathematical modeling of Zika virus disease with nonlinear incidence and optimal control

Mathematical model of Zika Virus dynamics with vector control and sensitivity analysis

Mathematical modeling of tuberculosis data of China

Mathematical modeling of pulmonary tuberculosis therapy: Insights from a prototype model with rifampin

Tuberculosis control in South African gold mines: Mathematical modeling of a trial of community-wide isoniazid preventive therapy

Mathematical modeling of tuberculosis propagation and patient detection

Outbreak dynamics of COVID-19 in Europe and the effect of travel restrictions

Monitoring transmissibility and mortality of COVID-19 in Europe

Potential scenarios for the progression of a COVID-19 epidemic in the European Union and the European Economic Area

COVID-19 pandemic: a mobility-dependent SEIR model with undetected cases in Italy

Estimation of COVID-19 prevalence in Italy, Spain, and France

On the uncertainty of real-time predictions of epidemic growths: A COVID-19 case study for China and Italy

Mathematical model of infection kinetics and its analysis for COVID-19, SARS and MERS

Trend analysis and forecasting of COVID-19 outbreak in India

Forecasting the Worldwide spread of COVID-19 based on Logistic Model and SEIR Model

Estimating the reproductive number and the outbreak size of COVID-19 in Korea

Covid-19 risks and response in South Asia

Assessing the risk of spread of COVID-19 to the Asia Pacific region

Prediction on COVID-19 epidemic for different countries: Focusing on South Asia under various precautionary measures

What will be the economic impact of COVID-19 in the US? Rough estimates of disease scenarios

Transmission potential of the novel coronavirus (COVID-19) onboard the Diamond Princess Cruises ship

Analysis of a mathematical model for COVID-19 population dynamics in Lagos

Modeling Nigerian COVID-19 cases: A comparative analysis of models and estimators

Machine-learning models for predicting survivability in COVID-19 patients

A two-level deterministic reasoning pattern to curb the spread of COVID-19 in Africa

Preparedness and vulnerability of African countries against importations of COVID-19: a modelling study

Strong correlations between power-law growth of COVID-19 in four continents and the inefficiency of soft quarantine strategies

New fractional derivatives with nonlocal and non-singular kernel: Theory and application to heat transfer model

A new fractional model for tuberculosis with relapse via Atangana-Baleanu derivative

Thounthong, Fractional investigations of zoonotic visceral leishmaniasis disease with singular and non-singular kernel

Chaos in a calcium oscillation model via Atangana-Baleanu operator with strong memory

Modeling and simulation of nonlinear dynamical system in the frame of nonlocal and nonsingular derivatives

Mathematical analysis and computational experiments for an epidemic system with nonlocal and nonsingular derivative

Modeling the mechanics of viral kinetics under immune control during primary infection of HIV-1 with treatment in fractional order

A fractional order model for Hepatitis B virus with treatment via Atangana-Baleanu derivative

Modeling the transmission of Dengue infection through fractional derivatives

A comparison study of bank data in fractional calculus

The dynamics of a new chaotic system through the Caputo-Fabrizio and Atanagan-Baleanu fractional operators

An epidemiological approach to insurgent population modeling with the Atangana-Baleanu fractional derivative

On the definition and the computation of the basic reproduction ratio R 0 in models for 43 in heterogeneous populations

Mathematical Epidemiology of Infectious Diseases: Model Building

COVID-2019) Situation Reports, Situation reports, World Health Organization

Modeling the effect of quarantine and vaccination on Ebola disease

Mathematical analysis of the role of hospitalization/isolation in controlling the spread of Zika fever

Modeling the impact of quarantine during an outbreak of Ebola virus disease

Differential Equations and Dynamical Systems

Lyapunov functions for a dengue disease transmission model

United Nations Data Bank

Containment, contact tracing and asymptomatic transmission of novel coronavirus disease (COVID-19): A modelling study

Proportion of asymptomatic infection among COVID-19 positive persons and their transmission potential: A systematic review and meta-analysis

A mathematical model for simulating the phase-based transmissibility of a novel coronavirus

Determining important parameters in the spread of Malaria through the sensitivity analysis of a mathematical model

Sensitivity and uncertainty analysis of complex models of disease transmission: an HIV model, as an example

Blood vessel detection based on fractional hessian matrix with nonsingular Mittag-Leffler Gaussian kernel

FPGA implementation and control of chaotic systems involving the variable-order fractional operator with Mittag-Leffler law

Decolonisation of fractional calculus rules: breaking commutativity and associativity to capture more natural phenomena

Fractional dynamics and synchronization of Kuramoto oscillators with nonlocal, nonsingular and strong memory

New numerical approximation of fractional derivative with non-local and non-singular kernel: application to chaotic models

Modeling the dynamics of Hepatitis E with optimal control

The effective reproduction number as a prelude to statistical estimation of time-dependent epidemic trends

Could masks curtail the postlockdown resurgence of COVID-19 in the US?

A primer on using mathematics to understand COVID-19 dynamics: Modeling, analysis and simulations

We are grateful for support from the Engineering and Physical Sciences Research Council (United Kingdom) under research grants Nos. EP/D000610/1 and EP/M015831/1. 

We now describe the numerical procedure used for solution of the fractional model (34) by adopting the modified Adams-Bashforth scheme developed by Toufik and Atangana [78] . Some recent applications of the modified Adams-Bashforth scheme, include [47, 48, 49, 79] . Before applying the procedure in Toufik and Atangana [78] , we write the fractional COVID-19 model (34) in the following form: Following the procedure in Toufik and Atangana [78] , the fractional model can take the form:Using t = t n+1 , n = 0, 1, 2, . . ., in (37), we obtain:

J o u r n a l P r e -p r o o f− (n − k) σ (n − k + 2 + 2σ)) − φ σ f 7 (t k−1 , R) Γ(σ + 2) × (n + 1 − k) σ+1 − (n − k) σ (n − k + 1 + σ) ,where φ = t n+1 − t n .

We declare that we have no known competing financial interests or personal relationships that could have influenced the work reported in this paper.